xref: /openbmc/linux/mm/page_alloc.c (revision adb80490)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/page_alloc.c
4  *
5  *  Manages the free list, the system allocates free pages here.
6  *  Note that kmalloc() lives in slab.c
7  *
8  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
9  *  Swap reorganised 29.12.95, Stephen Tweedie
10  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
11  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
12  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
13  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
14  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
15  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
16  */
17 
18 #include <linux/stddef.h>
19 #include <linux/mm.h>
20 #include <linux/highmem.h>
21 #include <linux/swap.h>
22 #include <linux/interrupt.h>
23 #include <linux/pagemap.h>
24 #include <linux/jiffies.h>
25 #include <linux/memblock.h>
26 #include <linux/compiler.h>
27 #include <linux/kernel.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/memremap.h>
46 #include <linux/stop_machine.h>
47 #include <linux/random.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/debugobjects.h>
54 #include <linux/kmemleak.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <trace/events/oom.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
63 #include <linux/sched/mm.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
66 #include <linux/memcontrol.h>
67 #include <linux/ftrace.h>
68 #include <linux/lockdep.h>
69 #include <linux/nmi.h>
70 #include <linux/psi.h>
71 
72 #include <asm/sections.h>
73 #include <asm/tlbflush.h>
74 #include <asm/div64.h>
75 #include "internal.h"
76 #include "shuffle.h"
77 
78 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
79 static DEFINE_MUTEX(pcp_batch_high_lock);
80 #define MIN_PERCPU_PAGELIST_FRACTION	(8)
81 
82 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
83 DEFINE_PER_CPU(int, numa_node);
84 EXPORT_PER_CPU_SYMBOL(numa_node);
85 #endif
86 
87 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
88 
89 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
90 /*
91  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
92  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
93  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
94  * defined in <linux/topology.h>.
95  */
96 DEFINE_PER_CPU(int, _numa_mem_);		/* Kernel "local memory" node */
97 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
98 int _node_numa_mem_[MAX_NUMNODES];
99 #endif
100 
101 /* work_structs for global per-cpu drains */
102 struct pcpu_drain {
103 	struct zone *zone;
104 	struct work_struct work;
105 };
106 DEFINE_MUTEX(pcpu_drain_mutex);
107 DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
108 
109 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
110 volatile unsigned long latent_entropy __latent_entropy;
111 EXPORT_SYMBOL(latent_entropy);
112 #endif
113 
114 /*
115  * Array of node states.
116  */
117 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
118 	[N_POSSIBLE] = NODE_MASK_ALL,
119 	[N_ONLINE] = { { [0] = 1UL } },
120 #ifndef CONFIG_NUMA
121 	[N_NORMAL_MEMORY] = { { [0] = 1UL } },
122 #ifdef CONFIG_HIGHMEM
123 	[N_HIGH_MEMORY] = { { [0] = 1UL } },
124 #endif
125 	[N_MEMORY] = { { [0] = 1UL } },
126 	[N_CPU] = { { [0] = 1UL } },
127 #endif	/* NUMA */
128 };
129 EXPORT_SYMBOL(node_states);
130 
131 atomic_long_t _totalram_pages __read_mostly;
132 EXPORT_SYMBOL(_totalram_pages);
133 unsigned long totalreserve_pages __read_mostly;
134 unsigned long totalcma_pages __read_mostly;
135 
136 int percpu_pagelist_fraction;
137 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
138 #ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON
139 DEFINE_STATIC_KEY_TRUE(init_on_alloc);
140 #else
141 DEFINE_STATIC_KEY_FALSE(init_on_alloc);
142 #endif
143 EXPORT_SYMBOL(init_on_alloc);
144 
145 #ifdef CONFIG_INIT_ON_FREE_DEFAULT_ON
146 DEFINE_STATIC_KEY_TRUE(init_on_free);
147 #else
148 DEFINE_STATIC_KEY_FALSE(init_on_free);
149 #endif
150 EXPORT_SYMBOL(init_on_free);
151 
152 static int __init early_init_on_alloc(char *buf)
153 {
154 	int ret;
155 	bool bool_result;
156 
157 	if (!buf)
158 		return -EINVAL;
159 	ret = kstrtobool(buf, &bool_result);
160 	if (bool_result && page_poisoning_enabled())
161 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc\n");
162 	if (bool_result)
163 		static_branch_enable(&init_on_alloc);
164 	else
165 		static_branch_disable(&init_on_alloc);
166 	return ret;
167 }
168 early_param("init_on_alloc", early_init_on_alloc);
169 
170 static int __init early_init_on_free(char *buf)
171 {
172 	int ret;
173 	bool bool_result;
174 
175 	if (!buf)
176 		return -EINVAL;
177 	ret = kstrtobool(buf, &bool_result);
178 	if (bool_result && page_poisoning_enabled())
179 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_free\n");
180 	if (bool_result)
181 		static_branch_enable(&init_on_free);
182 	else
183 		static_branch_disable(&init_on_free);
184 	return ret;
185 }
186 early_param("init_on_free", early_init_on_free);
187 
188 /*
189  * A cached value of the page's pageblock's migratetype, used when the page is
190  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
191  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
192  * Also the migratetype set in the page does not necessarily match the pcplist
193  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
194  * other index - this ensures that it will be put on the correct CMA freelist.
195  */
196 static inline int get_pcppage_migratetype(struct page *page)
197 {
198 	return page->index;
199 }
200 
201 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
202 {
203 	page->index = migratetype;
204 }
205 
206 #ifdef CONFIG_PM_SLEEP
207 /*
208  * The following functions are used by the suspend/hibernate code to temporarily
209  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
210  * while devices are suspended.  To avoid races with the suspend/hibernate code,
211  * they should always be called with system_transition_mutex held
212  * (gfp_allowed_mask also should only be modified with system_transition_mutex
213  * held, unless the suspend/hibernate code is guaranteed not to run in parallel
214  * with that modification).
215  */
216 
217 static gfp_t saved_gfp_mask;
218 
219 void pm_restore_gfp_mask(void)
220 {
221 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
222 	if (saved_gfp_mask) {
223 		gfp_allowed_mask = saved_gfp_mask;
224 		saved_gfp_mask = 0;
225 	}
226 }
227 
228 void pm_restrict_gfp_mask(void)
229 {
230 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
231 	WARN_ON(saved_gfp_mask);
232 	saved_gfp_mask = gfp_allowed_mask;
233 	gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
234 }
235 
236 bool pm_suspended_storage(void)
237 {
238 	if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
239 		return false;
240 	return true;
241 }
242 #endif /* CONFIG_PM_SLEEP */
243 
244 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
245 unsigned int pageblock_order __read_mostly;
246 #endif
247 
248 static void __free_pages_ok(struct page *page, unsigned int order);
249 
250 /*
251  * results with 256, 32 in the lowmem_reserve sysctl:
252  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
253  *	1G machine -> (16M dma, 784M normal, 224M high)
254  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
255  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
256  *	HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
257  *
258  * TBD: should special case ZONE_DMA32 machines here - in those we normally
259  * don't need any ZONE_NORMAL reservation
260  */
261 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
262 #ifdef CONFIG_ZONE_DMA
263 	[ZONE_DMA] = 256,
264 #endif
265 #ifdef CONFIG_ZONE_DMA32
266 	[ZONE_DMA32] = 256,
267 #endif
268 	[ZONE_NORMAL] = 32,
269 #ifdef CONFIG_HIGHMEM
270 	[ZONE_HIGHMEM] = 0,
271 #endif
272 	[ZONE_MOVABLE] = 0,
273 };
274 
275 static char * const zone_names[MAX_NR_ZONES] = {
276 #ifdef CONFIG_ZONE_DMA
277 	 "DMA",
278 #endif
279 #ifdef CONFIG_ZONE_DMA32
280 	 "DMA32",
281 #endif
282 	 "Normal",
283 #ifdef CONFIG_HIGHMEM
284 	 "HighMem",
285 #endif
286 	 "Movable",
287 #ifdef CONFIG_ZONE_DEVICE
288 	 "Device",
289 #endif
290 };
291 
292 const char * const migratetype_names[MIGRATE_TYPES] = {
293 	"Unmovable",
294 	"Movable",
295 	"Reclaimable",
296 	"HighAtomic",
297 #ifdef CONFIG_CMA
298 	"CMA",
299 #endif
300 #ifdef CONFIG_MEMORY_ISOLATION
301 	"Isolate",
302 #endif
303 };
304 
305 compound_page_dtor * const compound_page_dtors[] = {
306 	NULL,
307 	free_compound_page,
308 #ifdef CONFIG_HUGETLB_PAGE
309 	free_huge_page,
310 #endif
311 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
312 	free_transhuge_page,
313 #endif
314 };
315 
316 int min_free_kbytes = 1024;
317 int user_min_free_kbytes = -1;
318 #ifdef CONFIG_DISCONTIGMEM
319 /*
320  * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges
321  * are not on separate NUMA nodes. Functionally this works but with
322  * watermark_boost_factor, it can reclaim prematurely as the ranges can be
323  * quite small. By default, do not boost watermarks on discontigmem as in
324  * many cases very high-order allocations like THP are likely to be
325  * unsupported and the premature reclaim offsets the advantage of long-term
326  * fragmentation avoidance.
327  */
328 int watermark_boost_factor __read_mostly;
329 #else
330 int watermark_boost_factor __read_mostly = 15000;
331 #endif
332 int watermark_scale_factor = 10;
333 
334 static unsigned long nr_kernel_pages __initdata;
335 static unsigned long nr_all_pages __initdata;
336 static unsigned long dma_reserve __initdata;
337 
338 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
339 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
340 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
341 static unsigned long required_kernelcore __initdata;
342 static unsigned long required_kernelcore_percent __initdata;
343 static unsigned long required_movablecore __initdata;
344 static unsigned long required_movablecore_percent __initdata;
345 static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
346 static bool mirrored_kernelcore __meminitdata;
347 
348 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
349 int movable_zone;
350 EXPORT_SYMBOL(movable_zone);
351 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
352 
353 #if MAX_NUMNODES > 1
354 unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
355 unsigned int nr_online_nodes __read_mostly = 1;
356 EXPORT_SYMBOL(nr_node_ids);
357 EXPORT_SYMBOL(nr_online_nodes);
358 #endif
359 
360 int page_group_by_mobility_disabled __read_mostly;
361 
362 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
363 /*
364  * During boot we initialize deferred pages on-demand, as needed, but once
365  * page_alloc_init_late() has finished, the deferred pages are all initialized,
366  * and we can permanently disable that path.
367  */
368 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
369 
370 /*
371  * Calling kasan_free_pages() only after deferred memory initialization
372  * has completed. Poisoning pages during deferred memory init will greatly
373  * lengthen the process and cause problem in large memory systems as the
374  * deferred pages initialization is done with interrupt disabled.
375  *
376  * Assuming that there will be no reference to those newly initialized
377  * pages before they are ever allocated, this should have no effect on
378  * KASAN memory tracking as the poison will be properly inserted at page
379  * allocation time. The only corner case is when pages are allocated by
380  * on-demand allocation and then freed again before the deferred pages
381  * initialization is done, but this is not likely to happen.
382  */
383 static inline void kasan_free_nondeferred_pages(struct page *page, int order)
384 {
385 	if (!static_branch_unlikely(&deferred_pages))
386 		kasan_free_pages(page, order);
387 }
388 
389 /* Returns true if the struct page for the pfn is uninitialised */
390 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
391 {
392 	int nid = early_pfn_to_nid(pfn);
393 
394 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
395 		return true;
396 
397 	return false;
398 }
399 
400 /*
401  * Returns true when the remaining initialisation should be deferred until
402  * later in the boot cycle when it can be parallelised.
403  */
404 static bool __meminit
405 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
406 {
407 	static unsigned long prev_end_pfn, nr_initialised;
408 
409 	/*
410 	 * prev_end_pfn static that contains the end of previous zone
411 	 * No need to protect because called very early in boot before smp_init.
412 	 */
413 	if (prev_end_pfn != end_pfn) {
414 		prev_end_pfn = end_pfn;
415 		nr_initialised = 0;
416 	}
417 
418 	/* Always populate low zones for address-constrained allocations */
419 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
420 		return false;
421 
422 	/*
423 	 * We start only with one section of pages, more pages are added as
424 	 * needed until the rest of deferred pages are initialized.
425 	 */
426 	nr_initialised++;
427 	if ((nr_initialised > PAGES_PER_SECTION) &&
428 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
429 		NODE_DATA(nid)->first_deferred_pfn = pfn;
430 		return true;
431 	}
432 	return false;
433 }
434 #else
435 #define kasan_free_nondeferred_pages(p, o)	kasan_free_pages(p, o)
436 
437 static inline bool early_page_uninitialised(unsigned long pfn)
438 {
439 	return false;
440 }
441 
442 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
443 {
444 	return false;
445 }
446 #endif
447 
448 /* Return a pointer to the bitmap storing bits affecting a block of pages */
449 static inline unsigned long *get_pageblock_bitmap(struct page *page,
450 							unsigned long pfn)
451 {
452 #ifdef CONFIG_SPARSEMEM
453 	return section_to_usemap(__pfn_to_section(pfn));
454 #else
455 	return page_zone(page)->pageblock_flags;
456 #endif /* CONFIG_SPARSEMEM */
457 }
458 
459 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
460 {
461 #ifdef CONFIG_SPARSEMEM
462 	pfn &= (PAGES_PER_SECTION-1);
463 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
464 #else
465 	pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
466 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
467 #endif /* CONFIG_SPARSEMEM */
468 }
469 
470 /**
471  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
472  * @page: The page within the block of interest
473  * @pfn: The target page frame number
474  * @end_bitidx: The last bit of interest to retrieve
475  * @mask: mask of bits that the caller is interested in
476  *
477  * Return: pageblock_bits flags
478  */
479 static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
480 					unsigned long pfn,
481 					unsigned long end_bitidx,
482 					unsigned long mask)
483 {
484 	unsigned long *bitmap;
485 	unsigned long bitidx, word_bitidx;
486 	unsigned long word;
487 
488 	bitmap = get_pageblock_bitmap(page, pfn);
489 	bitidx = pfn_to_bitidx(page, pfn);
490 	word_bitidx = bitidx / BITS_PER_LONG;
491 	bitidx &= (BITS_PER_LONG-1);
492 
493 	word = bitmap[word_bitidx];
494 	bitidx += end_bitidx;
495 	return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
496 }
497 
498 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
499 					unsigned long end_bitidx,
500 					unsigned long mask)
501 {
502 	return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
503 }
504 
505 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
506 {
507 	return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
508 }
509 
510 /**
511  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
512  * @page: The page within the block of interest
513  * @flags: The flags to set
514  * @pfn: The target page frame number
515  * @end_bitidx: The last bit of interest
516  * @mask: mask of bits that the caller is interested in
517  */
518 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
519 					unsigned long pfn,
520 					unsigned long end_bitidx,
521 					unsigned long mask)
522 {
523 	unsigned long *bitmap;
524 	unsigned long bitidx, word_bitidx;
525 	unsigned long old_word, word;
526 
527 	BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
528 	BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
529 
530 	bitmap = get_pageblock_bitmap(page, pfn);
531 	bitidx = pfn_to_bitidx(page, pfn);
532 	word_bitidx = bitidx / BITS_PER_LONG;
533 	bitidx &= (BITS_PER_LONG-1);
534 
535 	VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
536 
537 	bitidx += end_bitidx;
538 	mask <<= (BITS_PER_LONG - bitidx - 1);
539 	flags <<= (BITS_PER_LONG - bitidx - 1);
540 
541 	word = READ_ONCE(bitmap[word_bitidx]);
542 	for (;;) {
543 		old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
544 		if (word == old_word)
545 			break;
546 		word = old_word;
547 	}
548 }
549 
550 void set_pageblock_migratetype(struct page *page, int migratetype)
551 {
552 	if (unlikely(page_group_by_mobility_disabled &&
553 		     migratetype < MIGRATE_PCPTYPES))
554 		migratetype = MIGRATE_UNMOVABLE;
555 
556 	set_pageblock_flags_group(page, (unsigned long)migratetype,
557 					PB_migrate, PB_migrate_end);
558 }
559 
560 #ifdef CONFIG_DEBUG_VM
561 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
562 {
563 	int ret = 0;
564 	unsigned seq;
565 	unsigned long pfn = page_to_pfn(page);
566 	unsigned long sp, start_pfn;
567 
568 	do {
569 		seq = zone_span_seqbegin(zone);
570 		start_pfn = zone->zone_start_pfn;
571 		sp = zone->spanned_pages;
572 		if (!zone_spans_pfn(zone, pfn))
573 			ret = 1;
574 	} while (zone_span_seqretry(zone, seq));
575 
576 	if (ret)
577 		pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
578 			pfn, zone_to_nid(zone), zone->name,
579 			start_pfn, start_pfn + sp);
580 
581 	return ret;
582 }
583 
584 static int page_is_consistent(struct zone *zone, struct page *page)
585 {
586 	if (!pfn_valid_within(page_to_pfn(page)))
587 		return 0;
588 	if (zone != page_zone(page))
589 		return 0;
590 
591 	return 1;
592 }
593 /*
594  * Temporary debugging check for pages not lying within a given zone.
595  */
596 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
597 {
598 	if (page_outside_zone_boundaries(zone, page))
599 		return 1;
600 	if (!page_is_consistent(zone, page))
601 		return 1;
602 
603 	return 0;
604 }
605 #else
606 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
607 {
608 	return 0;
609 }
610 #endif
611 
612 static void bad_page(struct page *page, const char *reason,
613 		unsigned long bad_flags)
614 {
615 	static unsigned long resume;
616 	static unsigned long nr_shown;
617 	static unsigned long nr_unshown;
618 
619 	/*
620 	 * Allow a burst of 60 reports, then keep quiet for that minute;
621 	 * or allow a steady drip of one report per second.
622 	 */
623 	if (nr_shown == 60) {
624 		if (time_before(jiffies, resume)) {
625 			nr_unshown++;
626 			goto out;
627 		}
628 		if (nr_unshown) {
629 			pr_alert(
630 			      "BUG: Bad page state: %lu messages suppressed\n",
631 				nr_unshown);
632 			nr_unshown = 0;
633 		}
634 		nr_shown = 0;
635 	}
636 	if (nr_shown++ == 0)
637 		resume = jiffies + 60 * HZ;
638 
639 	pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
640 		current->comm, page_to_pfn(page));
641 	__dump_page(page, reason);
642 	bad_flags &= page->flags;
643 	if (bad_flags)
644 		pr_alert("bad because of flags: %#lx(%pGp)\n",
645 						bad_flags, &bad_flags);
646 	dump_page_owner(page);
647 
648 	print_modules();
649 	dump_stack();
650 out:
651 	/* Leave bad fields for debug, except PageBuddy could make trouble */
652 	page_mapcount_reset(page); /* remove PageBuddy */
653 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
654 }
655 
656 /*
657  * Higher-order pages are called "compound pages".  They are structured thusly:
658  *
659  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
660  *
661  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
662  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
663  *
664  * The first tail page's ->compound_dtor holds the offset in array of compound
665  * page destructors. See compound_page_dtors.
666  *
667  * The first tail page's ->compound_order holds the order of allocation.
668  * This usage means that zero-order pages may not be compound.
669  */
670 
671 void free_compound_page(struct page *page)
672 {
673 	__free_pages_ok(page, compound_order(page));
674 }
675 
676 void prep_compound_page(struct page *page, unsigned int order)
677 {
678 	int i;
679 	int nr_pages = 1 << order;
680 
681 	set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
682 	set_compound_order(page, order);
683 	__SetPageHead(page);
684 	for (i = 1; i < nr_pages; i++) {
685 		struct page *p = page + i;
686 		set_page_count(p, 0);
687 		p->mapping = TAIL_MAPPING;
688 		set_compound_head(p, page);
689 	}
690 	atomic_set(compound_mapcount_ptr(page), -1);
691 }
692 
693 #ifdef CONFIG_DEBUG_PAGEALLOC
694 unsigned int _debug_guardpage_minorder;
695 
696 #ifdef CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
697 DEFINE_STATIC_KEY_TRUE(_debug_pagealloc_enabled);
698 #else
699 DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
700 #endif
701 EXPORT_SYMBOL(_debug_pagealloc_enabled);
702 
703 DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
704 
705 static int __init early_debug_pagealloc(char *buf)
706 {
707 	bool enable = false;
708 
709 	if (kstrtobool(buf, &enable))
710 		return -EINVAL;
711 
712 	if (enable)
713 		static_branch_enable(&_debug_pagealloc_enabled);
714 
715 	return 0;
716 }
717 early_param("debug_pagealloc", early_debug_pagealloc);
718 
719 static void init_debug_guardpage(void)
720 {
721 	if (!debug_pagealloc_enabled())
722 		return;
723 
724 	if (!debug_guardpage_minorder())
725 		return;
726 
727 	static_branch_enable(&_debug_guardpage_enabled);
728 }
729 
730 static int __init debug_guardpage_minorder_setup(char *buf)
731 {
732 	unsigned long res;
733 
734 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
735 		pr_err("Bad debug_guardpage_minorder value\n");
736 		return 0;
737 	}
738 	_debug_guardpage_minorder = res;
739 	pr_info("Setting debug_guardpage_minorder to %lu\n", res);
740 	return 0;
741 }
742 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
743 
744 static inline bool set_page_guard(struct zone *zone, struct page *page,
745 				unsigned int order, int migratetype)
746 {
747 	if (!debug_guardpage_enabled())
748 		return false;
749 
750 	if (order >= debug_guardpage_minorder())
751 		return false;
752 
753 	__SetPageGuard(page);
754 	INIT_LIST_HEAD(&page->lru);
755 	set_page_private(page, order);
756 	/* Guard pages are not available for any usage */
757 	__mod_zone_freepage_state(zone, -(1 << order), migratetype);
758 
759 	return true;
760 }
761 
762 static inline void clear_page_guard(struct zone *zone, struct page *page,
763 				unsigned int order, int migratetype)
764 {
765 	if (!debug_guardpage_enabled())
766 		return;
767 
768 	__ClearPageGuard(page);
769 
770 	set_page_private(page, 0);
771 	if (!is_migrate_isolate(migratetype))
772 		__mod_zone_freepage_state(zone, (1 << order), migratetype);
773 }
774 #else
775 static inline bool set_page_guard(struct zone *zone, struct page *page,
776 			unsigned int order, int migratetype) { return false; }
777 static inline void clear_page_guard(struct zone *zone, struct page *page,
778 				unsigned int order, int migratetype) {}
779 #endif
780 
781 static inline void set_page_order(struct page *page, unsigned int order)
782 {
783 	set_page_private(page, order);
784 	__SetPageBuddy(page);
785 }
786 
787 /*
788  * This function checks whether a page is free && is the buddy
789  * we can coalesce a page and its buddy if
790  * (a) the buddy is not in a hole (check before calling!) &&
791  * (b) the buddy is in the buddy system &&
792  * (c) a page and its buddy have the same order &&
793  * (d) a page and its buddy are in the same zone.
794  *
795  * For recording whether a page is in the buddy system, we set PageBuddy.
796  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
797  *
798  * For recording page's order, we use page_private(page).
799  */
800 static inline int page_is_buddy(struct page *page, struct page *buddy,
801 							unsigned int order)
802 {
803 	if (page_is_guard(buddy) && page_order(buddy) == order) {
804 		if (page_zone_id(page) != page_zone_id(buddy))
805 			return 0;
806 
807 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
808 
809 		return 1;
810 	}
811 
812 	if (PageBuddy(buddy) && page_order(buddy) == order) {
813 		/*
814 		 * zone check is done late to avoid uselessly
815 		 * calculating zone/node ids for pages that could
816 		 * never merge.
817 		 */
818 		if (page_zone_id(page) != page_zone_id(buddy))
819 			return 0;
820 
821 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
822 
823 		return 1;
824 	}
825 	return 0;
826 }
827 
828 #ifdef CONFIG_COMPACTION
829 static inline struct capture_control *task_capc(struct zone *zone)
830 {
831 	struct capture_control *capc = current->capture_control;
832 
833 	return capc &&
834 		!(current->flags & PF_KTHREAD) &&
835 		!capc->page &&
836 		capc->cc->zone == zone &&
837 		capc->cc->direct_compaction ? capc : NULL;
838 }
839 
840 static inline bool
841 compaction_capture(struct capture_control *capc, struct page *page,
842 		   int order, int migratetype)
843 {
844 	if (!capc || order != capc->cc->order)
845 		return false;
846 
847 	/* Do not accidentally pollute CMA or isolated regions*/
848 	if (is_migrate_cma(migratetype) ||
849 	    is_migrate_isolate(migratetype))
850 		return false;
851 
852 	/*
853 	 * Do not let lower order allocations polluate a movable pageblock.
854 	 * This might let an unmovable request use a reclaimable pageblock
855 	 * and vice-versa but no more than normal fallback logic which can
856 	 * have trouble finding a high-order free page.
857 	 */
858 	if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
859 		return false;
860 
861 	capc->page = page;
862 	return true;
863 }
864 
865 #else
866 static inline struct capture_control *task_capc(struct zone *zone)
867 {
868 	return NULL;
869 }
870 
871 static inline bool
872 compaction_capture(struct capture_control *capc, struct page *page,
873 		   int order, int migratetype)
874 {
875 	return false;
876 }
877 #endif /* CONFIG_COMPACTION */
878 
879 /*
880  * Freeing function for a buddy system allocator.
881  *
882  * The concept of a buddy system is to maintain direct-mapped table
883  * (containing bit values) for memory blocks of various "orders".
884  * The bottom level table contains the map for the smallest allocatable
885  * units of memory (here, pages), and each level above it describes
886  * pairs of units from the levels below, hence, "buddies".
887  * At a high level, all that happens here is marking the table entry
888  * at the bottom level available, and propagating the changes upward
889  * as necessary, plus some accounting needed to play nicely with other
890  * parts of the VM system.
891  * At each level, we keep a list of pages, which are heads of continuous
892  * free pages of length of (1 << order) and marked with PageBuddy.
893  * Page's order is recorded in page_private(page) field.
894  * So when we are allocating or freeing one, we can derive the state of the
895  * other.  That is, if we allocate a small block, and both were
896  * free, the remainder of the region must be split into blocks.
897  * If a block is freed, and its buddy is also free, then this
898  * triggers coalescing into a block of larger size.
899  *
900  * -- nyc
901  */
902 
903 static inline void __free_one_page(struct page *page,
904 		unsigned long pfn,
905 		struct zone *zone, unsigned int order,
906 		int migratetype)
907 {
908 	unsigned long combined_pfn;
909 	unsigned long uninitialized_var(buddy_pfn);
910 	struct page *buddy;
911 	unsigned int max_order;
912 	struct capture_control *capc = task_capc(zone);
913 
914 	max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
915 
916 	VM_BUG_ON(!zone_is_initialized(zone));
917 	VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
918 
919 	VM_BUG_ON(migratetype == -1);
920 	if (likely(!is_migrate_isolate(migratetype)))
921 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
922 
923 	VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
924 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
925 
926 continue_merging:
927 	while (order < max_order - 1) {
928 		if (compaction_capture(capc, page, order, migratetype)) {
929 			__mod_zone_freepage_state(zone, -(1 << order),
930 								migratetype);
931 			return;
932 		}
933 		buddy_pfn = __find_buddy_pfn(pfn, order);
934 		buddy = page + (buddy_pfn - pfn);
935 
936 		if (!pfn_valid_within(buddy_pfn))
937 			goto done_merging;
938 		if (!page_is_buddy(page, buddy, order))
939 			goto done_merging;
940 		/*
941 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
942 		 * merge with it and move up one order.
943 		 */
944 		if (page_is_guard(buddy))
945 			clear_page_guard(zone, buddy, order, migratetype);
946 		else
947 			del_page_from_free_area(buddy, &zone->free_area[order]);
948 		combined_pfn = buddy_pfn & pfn;
949 		page = page + (combined_pfn - pfn);
950 		pfn = combined_pfn;
951 		order++;
952 	}
953 	if (max_order < MAX_ORDER) {
954 		/* If we are here, it means order is >= pageblock_order.
955 		 * We want to prevent merge between freepages on isolate
956 		 * pageblock and normal pageblock. Without this, pageblock
957 		 * isolation could cause incorrect freepage or CMA accounting.
958 		 *
959 		 * We don't want to hit this code for the more frequent
960 		 * low-order merging.
961 		 */
962 		if (unlikely(has_isolate_pageblock(zone))) {
963 			int buddy_mt;
964 
965 			buddy_pfn = __find_buddy_pfn(pfn, order);
966 			buddy = page + (buddy_pfn - pfn);
967 			buddy_mt = get_pageblock_migratetype(buddy);
968 
969 			if (migratetype != buddy_mt
970 					&& (is_migrate_isolate(migratetype) ||
971 						is_migrate_isolate(buddy_mt)))
972 				goto done_merging;
973 		}
974 		max_order++;
975 		goto continue_merging;
976 	}
977 
978 done_merging:
979 	set_page_order(page, order);
980 
981 	/*
982 	 * If this is not the largest possible page, check if the buddy
983 	 * of the next-highest order is free. If it is, it's possible
984 	 * that pages are being freed that will coalesce soon. In case,
985 	 * that is happening, add the free page to the tail of the list
986 	 * so it's less likely to be used soon and more likely to be merged
987 	 * as a higher order page
988 	 */
989 	if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)
990 			&& !is_shuffle_order(order)) {
991 		struct page *higher_page, *higher_buddy;
992 		combined_pfn = buddy_pfn & pfn;
993 		higher_page = page + (combined_pfn - pfn);
994 		buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
995 		higher_buddy = higher_page + (buddy_pfn - combined_pfn);
996 		if (pfn_valid_within(buddy_pfn) &&
997 		    page_is_buddy(higher_page, higher_buddy, order + 1)) {
998 			add_to_free_area_tail(page, &zone->free_area[order],
999 					      migratetype);
1000 			return;
1001 		}
1002 	}
1003 
1004 	if (is_shuffle_order(order))
1005 		add_to_free_area_random(page, &zone->free_area[order],
1006 				migratetype);
1007 	else
1008 		add_to_free_area(page, &zone->free_area[order], migratetype);
1009 
1010 }
1011 
1012 /*
1013  * A bad page could be due to a number of fields. Instead of multiple branches,
1014  * try and check multiple fields with one check. The caller must do a detailed
1015  * check if necessary.
1016  */
1017 static inline bool page_expected_state(struct page *page,
1018 					unsigned long check_flags)
1019 {
1020 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1021 		return false;
1022 
1023 	if (unlikely((unsigned long)page->mapping |
1024 			page_ref_count(page) |
1025 #ifdef CONFIG_MEMCG
1026 			(unsigned long)page->mem_cgroup |
1027 #endif
1028 			(page->flags & check_flags)))
1029 		return false;
1030 
1031 	return true;
1032 }
1033 
1034 static void free_pages_check_bad(struct page *page)
1035 {
1036 	const char *bad_reason;
1037 	unsigned long bad_flags;
1038 
1039 	bad_reason = NULL;
1040 	bad_flags = 0;
1041 
1042 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1043 		bad_reason = "nonzero mapcount";
1044 	if (unlikely(page->mapping != NULL))
1045 		bad_reason = "non-NULL mapping";
1046 	if (unlikely(page_ref_count(page) != 0))
1047 		bad_reason = "nonzero _refcount";
1048 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
1049 		bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1050 		bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
1051 	}
1052 #ifdef CONFIG_MEMCG
1053 	if (unlikely(page->mem_cgroup))
1054 		bad_reason = "page still charged to cgroup";
1055 #endif
1056 	bad_page(page, bad_reason, bad_flags);
1057 }
1058 
1059 static inline int free_pages_check(struct page *page)
1060 {
1061 	if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1062 		return 0;
1063 
1064 	/* Something has gone sideways, find it */
1065 	free_pages_check_bad(page);
1066 	return 1;
1067 }
1068 
1069 static int free_tail_pages_check(struct page *head_page, struct page *page)
1070 {
1071 	int ret = 1;
1072 
1073 	/*
1074 	 * We rely page->lru.next never has bit 0 set, unless the page
1075 	 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
1076 	 */
1077 	BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1078 
1079 	if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1080 		ret = 0;
1081 		goto out;
1082 	}
1083 	switch (page - head_page) {
1084 	case 1:
1085 		/* the first tail page: ->mapping may be compound_mapcount() */
1086 		if (unlikely(compound_mapcount(page))) {
1087 			bad_page(page, "nonzero compound_mapcount", 0);
1088 			goto out;
1089 		}
1090 		break;
1091 	case 2:
1092 		/*
1093 		 * the second tail page: ->mapping is
1094 		 * deferred_list.next -- ignore value.
1095 		 */
1096 		break;
1097 	default:
1098 		if (page->mapping != TAIL_MAPPING) {
1099 			bad_page(page, "corrupted mapping in tail page", 0);
1100 			goto out;
1101 		}
1102 		break;
1103 	}
1104 	if (unlikely(!PageTail(page))) {
1105 		bad_page(page, "PageTail not set", 0);
1106 		goto out;
1107 	}
1108 	if (unlikely(compound_head(page) != head_page)) {
1109 		bad_page(page, "compound_head not consistent", 0);
1110 		goto out;
1111 	}
1112 	ret = 0;
1113 out:
1114 	page->mapping = NULL;
1115 	clear_compound_head(page);
1116 	return ret;
1117 }
1118 
1119 static void kernel_init_free_pages(struct page *page, int numpages)
1120 {
1121 	int i;
1122 
1123 	for (i = 0; i < numpages; i++)
1124 		clear_highpage(page + i);
1125 }
1126 
1127 static __always_inline bool free_pages_prepare(struct page *page,
1128 					unsigned int order, bool check_free)
1129 {
1130 	int bad = 0;
1131 
1132 	VM_BUG_ON_PAGE(PageTail(page), page);
1133 
1134 	trace_mm_page_free(page, order);
1135 
1136 	/*
1137 	 * Check tail pages before head page information is cleared to
1138 	 * avoid checking PageCompound for order-0 pages.
1139 	 */
1140 	if (unlikely(order)) {
1141 		bool compound = PageCompound(page);
1142 		int i;
1143 
1144 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1145 
1146 		if (compound)
1147 			ClearPageDoubleMap(page);
1148 		for (i = 1; i < (1 << order); i++) {
1149 			if (compound)
1150 				bad += free_tail_pages_check(page, page + i);
1151 			if (unlikely(free_pages_check(page + i))) {
1152 				bad++;
1153 				continue;
1154 			}
1155 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1156 		}
1157 	}
1158 	if (PageMappingFlags(page))
1159 		page->mapping = NULL;
1160 	if (memcg_kmem_enabled() && PageKmemcg(page))
1161 		__memcg_kmem_uncharge(page, order);
1162 	if (check_free)
1163 		bad += free_pages_check(page);
1164 	if (bad)
1165 		return false;
1166 
1167 	page_cpupid_reset_last(page);
1168 	page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1169 	reset_page_owner(page, order);
1170 
1171 	if (!PageHighMem(page)) {
1172 		debug_check_no_locks_freed(page_address(page),
1173 					   PAGE_SIZE << order);
1174 		debug_check_no_obj_freed(page_address(page),
1175 					   PAGE_SIZE << order);
1176 	}
1177 	arch_free_page(page, order);
1178 	if (want_init_on_free())
1179 		kernel_init_free_pages(page, 1 << order);
1180 
1181 	kernel_poison_pages(page, 1 << order, 0);
1182 	if (debug_pagealloc_enabled())
1183 		kernel_map_pages(page, 1 << order, 0);
1184 
1185 	kasan_free_nondeferred_pages(page, order);
1186 
1187 	return true;
1188 }
1189 
1190 #ifdef CONFIG_DEBUG_VM
1191 /*
1192  * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed
1193  * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when
1194  * moved from pcp lists to free lists.
1195  */
1196 static bool free_pcp_prepare(struct page *page)
1197 {
1198 	return free_pages_prepare(page, 0, true);
1199 }
1200 
1201 static bool bulkfree_pcp_prepare(struct page *page)
1202 {
1203 	if (debug_pagealloc_enabled())
1204 		return free_pages_check(page);
1205 	else
1206 		return false;
1207 }
1208 #else
1209 /*
1210  * With DEBUG_VM disabled, order-0 pages being freed are checked only when
1211  * moving from pcp lists to free list in order to reduce overhead. With
1212  * debug_pagealloc enabled, they are checked also immediately when being freed
1213  * to the pcp lists.
1214  */
1215 static bool free_pcp_prepare(struct page *page)
1216 {
1217 	if (debug_pagealloc_enabled())
1218 		return free_pages_prepare(page, 0, true);
1219 	else
1220 		return free_pages_prepare(page, 0, false);
1221 }
1222 
1223 static bool bulkfree_pcp_prepare(struct page *page)
1224 {
1225 	return free_pages_check(page);
1226 }
1227 #endif /* CONFIG_DEBUG_VM */
1228 
1229 static inline void prefetch_buddy(struct page *page)
1230 {
1231 	unsigned long pfn = page_to_pfn(page);
1232 	unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1233 	struct page *buddy = page + (buddy_pfn - pfn);
1234 
1235 	prefetch(buddy);
1236 }
1237 
1238 /*
1239  * Frees a number of pages from the PCP lists
1240  * Assumes all pages on list are in same zone, and of same order.
1241  * count is the number of pages to free.
1242  *
1243  * If the zone was previously in an "all pages pinned" state then look to
1244  * see if this freeing clears that state.
1245  *
1246  * And clear the zone's pages_scanned counter, to hold off the "all pages are
1247  * pinned" detection logic.
1248  */
1249 static void free_pcppages_bulk(struct zone *zone, int count,
1250 					struct per_cpu_pages *pcp)
1251 {
1252 	int migratetype = 0;
1253 	int batch_free = 0;
1254 	int prefetch_nr = 0;
1255 	bool isolated_pageblocks;
1256 	struct page *page, *tmp;
1257 	LIST_HEAD(head);
1258 
1259 	while (count) {
1260 		struct list_head *list;
1261 
1262 		/*
1263 		 * Remove pages from lists in a round-robin fashion. A
1264 		 * batch_free count is maintained that is incremented when an
1265 		 * empty list is encountered.  This is so more pages are freed
1266 		 * off fuller lists instead of spinning excessively around empty
1267 		 * lists
1268 		 */
1269 		do {
1270 			batch_free++;
1271 			if (++migratetype == MIGRATE_PCPTYPES)
1272 				migratetype = 0;
1273 			list = &pcp->lists[migratetype];
1274 		} while (list_empty(list));
1275 
1276 		/* This is the only non-empty list. Free them all. */
1277 		if (batch_free == MIGRATE_PCPTYPES)
1278 			batch_free = count;
1279 
1280 		do {
1281 			page = list_last_entry(list, struct page, lru);
1282 			/* must delete to avoid corrupting pcp list */
1283 			list_del(&page->lru);
1284 			pcp->count--;
1285 
1286 			if (bulkfree_pcp_prepare(page))
1287 				continue;
1288 
1289 			list_add_tail(&page->lru, &head);
1290 
1291 			/*
1292 			 * We are going to put the page back to the global
1293 			 * pool, prefetch its buddy to speed up later access
1294 			 * under zone->lock. It is believed the overhead of
1295 			 * an additional test and calculating buddy_pfn here
1296 			 * can be offset by reduced memory latency later. To
1297 			 * avoid excessive prefetching due to large count, only
1298 			 * prefetch buddy for the first pcp->batch nr of pages.
1299 			 */
1300 			if (prefetch_nr++ < pcp->batch)
1301 				prefetch_buddy(page);
1302 		} while (--count && --batch_free && !list_empty(list));
1303 	}
1304 
1305 	spin_lock(&zone->lock);
1306 	isolated_pageblocks = has_isolate_pageblock(zone);
1307 
1308 	/*
1309 	 * Use safe version since after __free_one_page(),
1310 	 * page->lru.next will not point to original list.
1311 	 */
1312 	list_for_each_entry_safe(page, tmp, &head, lru) {
1313 		int mt = get_pcppage_migratetype(page);
1314 		/* MIGRATE_ISOLATE page should not go to pcplists */
1315 		VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1316 		/* Pageblock could have been isolated meanwhile */
1317 		if (unlikely(isolated_pageblocks))
1318 			mt = get_pageblock_migratetype(page);
1319 
1320 		__free_one_page(page, page_to_pfn(page), zone, 0, mt);
1321 		trace_mm_page_pcpu_drain(page, 0, mt);
1322 	}
1323 	spin_unlock(&zone->lock);
1324 }
1325 
1326 static void free_one_page(struct zone *zone,
1327 				struct page *page, unsigned long pfn,
1328 				unsigned int order,
1329 				int migratetype)
1330 {
1331 	spin_lock(&zone->lock);
1332 	if (unlikely(has_isolate_pageblock(zone) ||
1333 		is_migrate_isolate(migratetype))) {
1334 		migratetype = get_pfnblock_migratetype(page, pfn);
1335 	}
1336 	__free_one_page(page, pfn, zone, order, migratetype);
1337 	spin_unlock(&zone->lock);
1338 }
1339 
1340 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1341 				unsigned long zone, int nid)
1342 {
1343 	mm_zero_struct_page(page);
1344 	set_page_links(page, zone, nid, pfn);
1345 	init_page_count(page);
1346 	page_mapcount_reset(page);
1347 	page_cpupid_reset_last(page);
1348 	page_kasan_tag_reset(page);
1349 
1350 	INIT_LIST_HEAD(&page->lru);
1351 #ifdef WANT_PAGE_VIRTUAL
1352 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1353 	if (!is_highmem_idx(zone))
1354 		set_page_address(page, __va(pfn << PAGE_SHIFT));
1355 #endif
1356 }
1357 
1358 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1359 static void __meminit init_reserved_page(unsigned long pfn)
1360 {
1361 	pg_data_t *pgdat;
1362 	int nid, zid;
1363 
1364 	if (!early_page_uninitialised(pfn))
1365 		return;
1366 
1367 	nid = early_pfn_to_nid(pfn);
1368 	pgdat = NODE_DATA(nid);
1369 
1370 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1371 		struct zone *zone = &pgdat->node_zones[zid];
1372 
1373 		if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1374 			break;
1375 	}
1376 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1377 }
1378 #else
1379 static inline void init_reserved_page(unsigned long pfn)
1380 {
1381 }
1382 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1383 
1384 /*
1385  * Initialised pages do not have PageReserved set. This function is
1386  * called for each range allocated by the bootmem allocator and
1387  * marks the pages PageReserved. The remaining valid pages are later
1388  * sent to the buddy page allocator.
1389  */
1390 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1391 {
1392 	unsigned long start_pfn = PFN_DOWN(start);
1393 	unsigned long end_pfn = PFN_UP(end);
1394 
1395 	for (; start_pfn < end_pfn; start_pfn++) {
1396 		if (pfn_valid(start_pfn)) {
1397 			struct page *page = pfn_to_page(start_pfn);
1398 
1399 			init_reserved_page(start_pfn);
1400 
1401 			/* Avoid false-positive PageTail() */
1402 			INIT_LIST_HEAD(&page->lru);
1403 
1404 			/*
1405 			 * no need for atomic set_bit because the struct
1406 			 * page is not visible yet so nobody should
1407 			 * access it yet.
1408 			 */
1409 			__SetPageReserved(page);
1410 		}
1411 	}
1412 }
1413 
1414 static void __free_pages_ok(struct page *page, unsigned int order)
1415 {
1416 	unsigned long flags;
1417 	int migratetype;
1418 	unsigned long pfn = page_to_pfn(page);
1419 
1420 	if (!free_pages_prepare(page, order, true))
1421 		return;
1422 
1423 	migratetype = get_pfnblock_migratetype(page, pfn);
1424 	local_irq_save(flags);
1425 	__count_vm_events(PGFREE, 1 << order);
1426 	free_one_page(page_zone(page), page, pfn, order, migratetype);
1427 	local_irq_restore(flags);
1428 }
1429 
1430 void __free_pages_core(struct page *page, unsigned int order)
1431 {
1432 	unsigned int nr_pages = 1 << order;
1433 	struct page *p = page;
1434 	unsigned int loop;
1435 
1436 	prefetchw(p);
1437 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1438 		prefetchw(p + 1);
1439 		__ClearPageReserved(p);
1440 		set_page_count(p, 0);
1441 	}
1442 	__ClearPageReserved(p);
1443 	set_page_count(p, 0);
1444 
1445 	atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1446 	set_page_refcounted(page);
1447 	__free_pages(page, order);
1448 }
1449 
1450 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1451 	defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1452 
1453 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1454 
1455 int __meminit early_pfn_to_nid(unsigned long pfn)
1456 {
1457 	static DEFINE_SPINLOCK(early_pfn_lock);
1458 	int nid;
1459 
1460 	spin_lock(&early_pfn_lock);
1461 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1462 	if (nid < 0)
1463 		nid = first_online_node;
1464 	spin_unlock(&early_pfn_lock);
1465 
1466 	return nid;
1467 }
1468 #endif
1469 
1470 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1471 /* Only safe to use early in boot when initialisation is single-threaded */
1472 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1473 {
1474 	int nid;
1475 
1476 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1477 	if (nid >= 0 && nid != node)
1478 		return false;
1479 	return true;
1480 }
1481 
1482 #else
1483 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1484 {
1485 	return true;
1486 }
1487 #endif
1488 
1489 
1490 void __init memblock_free_pages(struct page *page, unsigned long pfn,
1491 							unsigned int order)
1492 {
1493 	if (early_page_uninitialised(pfn))
1494 		return;
1495 	__free_pages_core(page, order);
1496 }
1497 
1498 /*
1499  * Check that the whole (or subset of) a pageblock given by the interval of
1500  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1501  * with the migration of free compaction scanner. The scanners then need to
1502  * use only pfn_valid_within() check for arches that allow holes within
1503  * pageblocks.
1504  *
1505  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1506  *
1507  * It's possible on some configurations to have a setup like node0 node1 node0
1508  * i.e. it's possible that all pages within a zones range of pages do not
1509  * belong to a single zone. We assume that a border between node0 and node1
1510  * can occur within a single pageblock, but not a node0 node1 node0
1511  * interleaving within a single pageblock. It is therefore sufficient to check
1512  * the first and last page of a pageblock and avoid checking each individual
1513  * page in a pageblock.
1514  */
1515 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1516 				     unsigned long end_pfn, struct zone *zone)
1517 {
1518 	struct page *start_page;
1519 	struct page *end_page;
1520 
1521 	/* end_pfn is one past the range we are checking */
1522 	end_pfn--;
1523 
1524 	if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1525 		return NULL;
1526 
1527 	start_page = pfn_to_online_page(start_pfn);
1528 	if (!start_page)
1529 		return NULL;
1530 
1531 	if (page_zone(start_page) != zone)
1532 		return NULL;
1533 
1534 	end_page = pfn_to_page(end_pfn);
1535 
1536 	/* This gives a shorter code than deriving page_zone(end_page) */
1537 	if (page_zone_id(start_page) != page_zone_id(end_page))
1538 		return NULL;
1539 
1540 	return start_page;
1541 }
1542 
1543 void set_zone_contiguous(struct zone *zone)
1544 {
1545 	unsigned long block_start_pfn = zone->zone_start_pfn;
1546 	unsigned long block_end_pfn;
1547 
1548 	block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1549 	for (; block_start_pfn < zone_end_pfn(zone);
1550 			block_start_pfn = block_end_pfn,
1551 			 block_end_pfn += pageblock_nr_pages) {
1552 
1553 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1554 
1555 		if (!__pageblock_pfn_to_page(block_start_pfn,
1556 					     block_end_pfn, zone))
1557 			return;
1558 	}
1559 
1560 	/* We confirm that there is no hole */
1561 	zone->contiguous = true;
1562 }
1563 
1564 void clear_zone_contiguous(struct zone *zone)
1565 {
1566 	zone->contiguous = false;
1567 }
1568 
1569 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1570 static void __init deferred_free_range(unsigned long pfn,
1571 				       unsigned long nr_pages)
1572 {
1573 	struct page *page;
1574 	unsigned long i;
1575 
1576 	if (!nr_pages)
1577 		return;
1578 
1579 	page = pfn_to_page(pfn);
1580 
1581 	/* Free a large naturally-aligned chunk if possible */
1582 	if (nr_pages == pageblock_nr_pages &&
1583 	    (pfn & (pageblock_nr_pages - 1)) == 0) {
1584 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1585 		__free_pages_core(page, pageblock_order);
1586 		return;
1587 	}
1588 
1589 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1590 		if ((pfn & (pageblock_nr_pages - 1)) == 0)
1591 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1592 		__free_pages_core(page, 0);
1593 	}
1594 }
1595 
1596 /* Completion tracking for deferred_init_memmap() threads */
1597 static atomic_t pgdat_init_n_undone __initdata;
1598 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1599 
1600 static inline void __init pgdat_init_report_one_done(void)
1601 {
1602 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1603 		complete(&pgdat_init_all_done_comp);
1604 }
1605 
1606 /*
1607  * Returns true if page needs to be initialized or freed to buddy allocator.
1608  *
1609  * First we check if pfn is valid on architectures where it is possible to have
1610  * holes within pageblock_nr_pages. On systems where it is not possible, this
1611  * function is optimized out.
1612  *
1613  * Then, we check if a current large page is valid by only checking the validity
1614  * of the head pfn.
1615  */
1616 static inline bool __init deferred_pfn_valid(unsigned long pfn)
1617 {
1618 	if (!pfn_valid_within(pfn))
1619 		return false;
1620 	if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1621 		return false;
1622 	return true;
1623 }
1624 
1625 /*
1626  * Free pages to buddy allocator. Try to free aligned pages in
1627  * pageblock_nr_pages sizes.
1628  */
1629 static void __init deferred_free_pages(unsigned long pfn,
1630 				       unsigned long end_pfn)
1631 {
1632 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1633 	unsigned long nr_free = 0;
1634 
1635 	for (; pfn < end_pfn; pfn++) {
1636 		if (!deferred_pfn_valid(pfn)) {
1637 			deferred_free_range(pfn - nr_free, nr_free);
1638 			nr_free = 0;
1639 		} else if (!(pfn & nr_pgmask)) {
1640 			deferred_free_range(pfn - nr_free, nr_free);
1641 			nr_free = 1;
1642 			touch_nmi_watchdog();
1643 		} else {
1644 			nr_free++;
1645 		}
1646 	}
1647 	/* Free the last block of pages to allocator */
1648 	deferred_free_range(pfn - nr_free, nr_free);
1649 }
1650 
1651 /*
1652  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
1653  * by performing it only once every pageblock_nr_pages.
1654  * Return number of pages initialized.
1655  */
1656 static unsigned long  __init deferred_init_pages(struct zone *zone,
1657 						 unsigned long pfn,
1658 						 unsigned long end_pfn)
1659 {
1660 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1661 	int nid = zone_to_nid(zone);
1662 	unsigned long nr_pages = 0;
1663 	int zid = zone_idx(zone);
1664 	struct page *page = NULL;
1665 
1666 	for (; pfn < end_pfn; pfn++) {
1667 		if (!deferred_pfn_valid(pfn)) {
1668 			page = NULL;
1669 			continue;
1670 		} else if (!page || !(pfn & nr_pgmask)) {
1671 			page = pfn_to_page(pfn);
1672 			touch_nmi_watchdog();
1673 		} else {
1674 			page++;
1675 		}
1676 		__init_single_page(page, pfn, zid, nid);
1677 		nr_pages++;
1678 	}
1679 	return (nr_pages);
1680 }
1681 
1682 /*
1683  * This function is meant to pre-load the iterator for the zone init.
1684  * Specifically it walks through the ranges until we are caught up to the
1685  * first_init_pfn value and exits there. If we never encounter the value we
1686  * return false indicating there are no valid ranges left.
1687  */
1688 static bool __init
1689 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1690 				    unsigned long *spfn, unsigned long *epfn,
1691 				    unsigned long first_init_pfn)
1692 {
1693 	u64 j;
1694 
1695 	/*
1696 	 * Start out by walking through the ranges in this zone that have
1697 	 * already been initialized. We don't need to do anything with them
1698 	 * so we just need to flush them out of the system.
1699 	 */
1700 	for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1701 		if (*epfn <= first_init_pfn)
1702 			continue;
1703 		if (*spfn < first_init_pfn)
1704 			*spfn = first_init_pfn;
1705 		*i = j;
1706 		return true;
1707 	}
1708 
1709 	return false;
1710 }
1711 
1712 /*
1713  * Initialize and free pages. We do it in two loops: first we initialize
1714  * struct page, then free to buddy allocator, because while we are
1715  * freeing pages we can access pages that are ahead (computing buddy
1716  * page in __free_one_page()).
1717  *
1718  * In order to try and keep some memory in the cache we have the loop
1719  * broken along max page order boundaries. This way we will not cause
1720  * any issues with the buddy page computation.
1721  */
1722 static unsigned long __init
1723 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1724 		       unsigned long *end_pfn)
1725 {
1726 	unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1727 	unsigned long spfn = *start_pfn, epfn = *end_pfn;
1728 	unsigned long nr_pages = 0;
1729 	u64 j = *i;
1730 
1731 	/* First we loop through and initialize the page values */
1732 	for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1733 		unsigned long t;
1734 
1735 		if (mo_pfn <= *start_pfn)
1736 			break;
1737 
1738 		t = min(mo_pfn, *end_pfn);
1739 		nr_pages += deferred_init_pages(zone, *start_pfn, t);
1740 
1741 		if (mo_pfn < *end_pfn) {
1742 			*start_pfn = mo_pfn;
1743 			break;
1744 		}
1745 	}
1746 
1747 	/* Reset values and now loop through freeing pages as needed */
1748 	swap(j, *i);
1749 
1750 	for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
1751 		unsigned long t;
1752 
1753 		if (mo_pfn <= spfn)
1754 			break;
1755 
1756 		t = min(mo_pfn, epfn);
1757 		deferred_free_pages(spfn, t);
1758 
1759 		if (mo_pfn <= epfn)
1760 			break;
1761 	}
1762 
1763 	return nr_pages;
1764 }
1765 
1766 /* Initialise remaining memory on a node */
1767 static int __init deferred_init_memmap(void *data)
1768 {
1769 	pg_data_t *pgdat = data;
1770 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1771 	unsigned long spfn = 0, epfn = 0, nr_pages = 0;
1772 	unsigned long first_init_pfn, flags;
1773 	unsigned long start = jiffies;
1774 	struct zone *zone;
1775 	int zid;
1776 	u64 i;
1777 
1778 	/* Bind memory initialisation thread to a local node if possible */
1779 	if (!cpumask_empty(cpumask))
1780 		set_cpus_allowed_ptr(current, cpumask);
1781 
1782 	pgdat_resize_lock(pgdat, &flags);
1783 	first_init_pfn = pgdat->first_deferred_pfn;
1784 	if (first_init_pfn == ULONG_MAX) {
1785 		pgdat_resize_unlock(pgdat, &flags);
1786 		pgdat_init_report_one_done();
1787 		return 0;
1788 	}
1789 
1790 	/* Sanity check boundaries */
1791 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1792 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1793 	pgdat->first_deferred_pfn = ULONG_MAX;
1794 
1795 	/* Only the highest zone is deferred so find it */
1796 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1797 		zone = pgdat->node_zones + zid;
1798 		if (first_init_pfn < zone_end_pfn(zone))
1799 			break;
1800 	}
1801 
1802 	/* If the zone is empty somebody else may have cleared out the zone */
1803 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1804 						 first_init_pfn))
1805 		goto zone_empty;
1806 
1807 	/*
1808 	 * Initialize and free pages in MAX_ORDER sized increments so
1809 	 * that we can avoid introducing any issues with the buddy
1810 	 * allocator.
1811 	 */
1812 	while (spfn < epfn)
1813 		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
1814 zone_empty:
1815 	pgdat_resize_unlock(pgdat, &flags);
1816 
1817 	/* Sanity check that the next zone really is unpopulated */
1818 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1819 
1820 	pr_info("node %d initialised, %lu pages in %ums\n",
1821 		pgdat->node_id,	nr_pages, jiffies_to_msecs(jiffies - start));
1822 
1823 	pgdat_init_report_one_done();
1824 	return 0;
1825 }
1826 
1827 /*
1828  * If this zone has deferred pages, try to grow it by initializing enough
1829  * deferred pages to satisfy the allocation specified by order, rounded up to
1830  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
1831  * of SECTION_SIZE bytes by initializing struct pages in increments of
1832  * PAGES_PER_SECTION * sizeof(struct page) bytes.
1833  *
1834  * Return true when zone was grown, otherwise return false. We return true even
1835  * when we grow less than requested, to let the caller decide if there are
1836  * enough pages to satisfy the allocation.
1837  *
1838  * Note: We use noinline because this function is needed only during boot, and
1839  * it is called from a __ref function _deferred_grow_zone. This way we are
1840  * making sure that it is not inlined into permanent text section.
1841  */
1842 static noinline bool __init
1843 deferred_grow_zone(struct zone *zone, unsigned int order)
1844 {
1845 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1846 	pg_data_t *pgdat = zone->zone_pgdat;
1847 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1848 	unsigned long spfn, epfn, flags;
1849 	unsigned long nr_pages = 0;
1850 	u64 i;
1851 
1852 	/* Only the last zone may have deferred pages */
1853 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1854 		return false;
1855 
1856 	pgdat_resize_lock(pgdat, &flags);
1857 
1858 	/*
1859 	 * If deferred pages have been initialized while we were waiting for
1860 	 * the lock, return true, as the zone was grown.  The caller will retry
1861 	 * this zone.  We won't return to this function since the caller also
1862 	 * has this static branch.
1863 	 */
1864 	if (!static_branch_unlikely(&deferred_pages)) {
1865 		pgdat_resize_unlock(pgdat, &flags);
1866 		return true;
1867 	}
1868 
1869 	/*
1870 	 * If someone grew this zone while we were waiting for spinlock, return
1871 	 * true, as there might be enough pages already.
1872 	 */
1873 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1874 		pgdat_resize_unlock(pgdat, &flags);
1875 		return true;
1876 	}
1877 
1878 	/* If the zone is empty somebody else may have cleared out the zone */
1879 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1880 						 first_deferred_pfn)) {
1881 		pgdat->first_deferred_pfn = ULONG_MAX;
1882 		pgdat_resize_unlock(pgdat, &flags);
1883 		/* Retry only once. */
1884 		return first_deferred_pfn != ULONG_MAX;
1885 	}
1886 
1887 	/*
1888 	 * Initialize and free pages in MAX_ORDER sized increments so
1889 	 * that we can avoid introducing any issues with the buddy
1890 	 * allocator.
1891 	 */
1892 	while (spfn < epfn) {
1893 		/* update our first deferred PFN for this section */
1894 		first_deferred_pfn = spfn;
1895 
1896 		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
1897 
1898 		/* We should only stop along section boundaries */
1899 		if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
1900 			continue;
1901 
1902 		/* If our quota has been met we can stop here */
1903 		if (nr_pages >= nr_pages_needed)
1904 			break;
1905 	}
1906 
1907 	pgdat->first_deferred_pfn = spfn;
1908 	pgdat_resize_unlock(pgdat, &flags);
1909 
1910 	return nr_pages > 0;
1911 }
1912 
1913 /*
1914  * deferred_grow_zone() is __init, but it is called from
1915  * get_page_from_freelist() during early boot until deferred_pages permanently
1916  * disables this call. This is why we have refdata wrapper to avoid warning,
1917  * and to ensure that the function body gets unloaded.
1918  */
1919 static bool __ref
1920 _deferred_grow_zone(struct zone *zone, unsigned int order)
1921 {
1922 	return deferred_grow_zone(zone, order);
1923 }
1924 
1925 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1926 
1927 void __init page_alloc_init_late(void)
1928 {
1929 	struct zone *zone;
1930 	int nid;
1931 
1932 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1933 
1934 	/* There will be num_node_state(N_MEMORY) threads */
1935 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1936 	for_each_node_state(nid, N_MEMORY) {
1937 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1938 	}
1939 
1940 	/* Block until all are initialised */
1941 	wait_for_completion(&pgdat_init_all_done_comp);
1942 
1943 	/*
1944 	 * We initialized the rest of the deferred pages.  Permanently disable
1945 	 * on-demand struct page initialization.
1946 	 */
1947 	static_branch_disable(&deferred_pages);
1948 
1949 	/* Reinit limits that are based on free pages after the kernel is up */
1950 	files_maxfiles_init();
1951 #endif
1952 
1953 	/* Discard memblock private memory */
1954 	memblock_discard();
1955 
1956 	for_each_node_state(nid, N_MEMORY)
1957 		shuffle_free_memory(NODE_DATA(nid));
1958 
1959 	for_each_populated_zone(zone)
1960 		set_zone_contiguous(zone);
1961 
1962 #ifdef CONFIG_DEBUG_PAGEALLOC
1963 	init_debug_guardpage();
1964 #endif
1965 }
1966 
1967 #ifdef CONFIG_CMA
1968 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1969 void __init init_cma_reserved_pageblock(struct page *page)
1970 {
1971 	unsigned i = pageblock_nr_pages;
1972 	struct page *p = page;
1973 
1974 	do {
1975 		__ClearPageReserved(p);
1976 		set_page_count(p, 0);
1977 	} while (++p, --i);
1978 
1979 	set_pageblock_migratetype(page, MIGRATE_CMA);
1980 
1981 	if (pageblock_order >= MAX_ORDER) {
1982 		i = pageblock_nr_pages;
1983 		p = page;
1984 		do {
1985 			set_page_refcounted(p);
1986 			__free_pages(p, MAX_ORDER - 1);
1987 			p += MAX_ORDER_NR_PAGES;
1988 		} while (i -= MAX_ORDER_NR_PAGES);
1989 	} else {
1990 		set_page_refcounted(page);
1991 		__free_pages(page, pageblock_order);
1992 	}
1993 
1994 	adjust_managed_page_count(page, pageblock_nr_pages);
1995 }
1996 #endif
1997 
1998 /*
1999  * The order of subdivision here is critical for the IO subsystem.
2000  * Please do not alter this order without good reasons and regression
2001  * testing. Specifically, as large blocks of memory are subdivided,
2002  * the order in which smaller blocks are delivered depends on the order
2003  * they're subdivided in this function. This is the primary factor
2004  * influencing the order in which pages are delivered to the IO
2005  * subsystem according to empirical testing, and this is also justified
2006  * by considering the behavior of a buddy system containing a single
2007  * large block of memory acted on by a series of small allocations.
2008  * This behavior is a critical factor in sglist merging's success.
2009  *
2010  * -- nyc
2011  */
2012 static inline void expand(struct zone *zone, struct page *page,
2013 	int low, int high, struct free_area *area,
2014 	int migratetype)
2015 {
2016 	unsigned long size = 1 << high;
2017 
2018 	while (high > low) {
2019 		area--;
2020 		high--;
2021 		size >>= 1;
2022 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2023 
2024 		/*
2025 		 * Mark as guard pages (or page), that will allow to
2026 		 * merge back to allocator when buddy will be freed.
2027 		 * Corresponding page table entries will not be touched,
2028 		 * pages will stay not present in virtual address space
2029 		 */
2030 		if (set_page_guard(zone, &page[size], high, migratetype))
2031 			continue;
2032 
2033 		add_to_free_area(&page[size], area, migratetype);
2034 		set_page_order(&page[size], high);
2035 	}
2036 }
2037 
2038 static void check_new_page_bad(struct page *page)
2039 {
2040 	const char *bad_reason = NULL;
2041 	unsigned long bad_flags = 0;
2042 
2043 	if (unlikely(atomic_read(&page->_mapcount) != -1))
2044 		bad_reason = "nonzero mapcount";
2045 	if (unlikely(page->mapping != NULL))
2046 		bad_reason = "non-NULL mapping";
2047 	if (unlikely(page_ref_count(page) != 0))
2048 		bad_reason = "nonzero _refcount";
2049 	if (unlikely(page->flags & __PG_HWPOISON)) {
2050 		bad_reason = "HWPoisoned (hardware-corrupted)";
2051 		bad_flags = __PG_HWPOISON;
2052 		/* Don't complain about hwpoisoned pages */
2053 		page_mapcount_reset(page); /* remove PageBuddy */
2054 		return;
2055 	}
2056 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
2057 		bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
2058 		bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
2059 	}
2060 #ifdef CONFIG_MEMCG
2061 	if (unlikely(page->mem_cgroup))
2062 		bad_reason = "page still charged to cgroup";
2063 #endif
2064 	bad_page(page, bad_reason, bad_flags);
2065 }
2066 
2067 /*
2068  * This page is about to be returned from the page allocator
2069  */
2070 static inline int check_new_page(struct page *page)
2071 {
2072 	if (likely(page_expected_state(page,
2073 				PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2074 		return 0;
2075 
2076 	check_new_page_bad(page);
2077 	return 1;
2078 }
2079 
2080 static inline bool free_pages_prezeroed(void)
2081 {
2082 	return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
2083 		page_poisoning_enabled()) || want_init_on_free();
2084 }
2085 
2086 #ifdef CONFIG_DEBUG_VM
2087 /*
2088  * With DEBUG_VM enabled, order-0 pages are checked for expected state when
2089  * being allocated from pcp lists. With debug_pagealloc also enabled, they are
2090  * also checked when pcp lists are refilled from the free lists.
2091  */
2092 static inline bool check_pcp_refill(struct page *page)
2093 {
2094 	if (debug_pagealloc_enabled())
2095 		return check_new_page(page);
2096 	else
2097 		return false;
2098 }
2099 
2100 static inline bool check_new_pcp(struct page *page)
2101 {
2102 	return check_new_page(page);
2103 }
2104 #else
2105 /*
2106  * With DEBUG_VM disabled, free order-0 pages are checked for expected state
2107  * when pcp lists are being refilled from the free lists. With debug_pagealloc
2108  * enabled, they are also checked when being allocated from the pcp lists.
2109  */
2110 static inline bool check_pcp_refill(struct page *page)
2111 {
2112 	return check_new_page(page);
2113 }
2114 static inline bool check_new_pcp(struct page *page)
2115 {
2116 	if (debug_pagealloc_enabled())
2117 		return check_new_page(page);
2118 	else
2119 		return false;
2120 }
2121 #endif /* CONFIG_DEBUG_VM */
2122 
2123 static bool check_new_pages(struct page *page, unsigned int order)
2124 {
2125 	int i;
2126 	for (i = 0; i < (1 << order); i++) {
2127 		struct page *p = page + i;
2128 
2129 		if (unlikely(check_new_page(p)))
2130 			return true;
2131 	}
2132 
2133 	return false;
2134 }
2135 
2136 inline void post_alloc_hook(struct page *page, unsigned int order,
2137 				gfp_t gfp_flags)
2138 {
2139 	set_page_private(page, 0);
2140 	set_page_refcounted(page);
2141 
2142 	arch_alloc_page(page, order);
2143 	if (debug_pagealloc_enabled())
2144 		kernel_map_pages(page, 1 << order, 1);
2145 	kasan_alloc_pages(page, order);
2146 	kernel_poison_pages(page, 1 << order, 1);
2147 	set_page_owner(page, order, gfp_flags);
2148 }
2149 
2150 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2151 							unsigned int alloc_flags)
2152 {
2153 	post_alloc_hook(page, order, gfp_flags);
2154 
2155 	if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags))
2156 		kernel_init_free_pages(page, 1 << order);
2157 
2158 	if (order && (gfp_flags & __GFP_COMP))
2159 		prep_compound_page(page, order);
2160 
2161 	/*
2162 	 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
2163 	 * allocate the page. The expectation is that the caller is taking
2164 	 * steps that will free more memory. The caller should avoid the page
2165 	 * being used for !PFMEMALLOC purposes.
2166 	 */
2167 	if (alloc_flags & ALLOC_NO_WATERMARKS)
2168 		set_page_pfmemalloc(page);
2169 	else
2170 		clear_page_pfmemalloc(page);
2171 }
2172 
2173 /*
2174  * Go through the free lists for the given migratetype and remove
2175  * the smallest available page from the freelists
2176  */
2177 static __always_inline
2178 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2179 						int migratetype)
2180 {
2181 	unsigned int current_order;
2182 	struct free_area *area;
2183 	struct page *page;
2184 
2185 	/* Find a page of the appropriate size in the preferred list */
2186 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2187 		area = &(zone->free_area[current_order]);
2188 		page = get_page_from_free_area(area, migratetype);
2189 		if (!page)
2190 			continue;
2191 		del_page_from_free_area(page, area);
2192 		expand(zone, page, order, current_order, area, migratetype);
2193 		set_pcppage_migratetype(page, migratetype);
2194 		return page;
2195 	}
2196 
2197 	return NULL;
2198 }
2199 
2200 
2201 /*
2202  * This array describes the order lists are fallen back to when
2203  * the free lists for the desirable migrate type are depleted
2204  */
2205 static int fallbacks[MIGRATE_TYPES][4] = {
2206 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
2207 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2208 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
2209 #ifdef CONFIG_CMA
2210 	[MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
2211 #endif
2212 #ifdef CONFIG_MEMORY_ISOLATION
2213 	[MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
2214 #endif
2215 };
2216 
2217 #ifdef CONFIG_CMA
2218 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2219 					unsigned int order)
2220 {
2221 	return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2222 }
2223 #else
2224 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2225 					unsigned int order) { return NULL; }
2226 #endif
2227 
2228 /*
2229  * Move the free pages in a range to the free lists of the requested type.
2230  * Note that start_page and end_pages are not aligned on a pageblock
2231  * boundary. If alignment is required, use move_freepages_block()
2232  */
2233 static int move_freepages(struct zone *zone,
2234 			  struct page *start_page, struct page *end_page,
2235 			  int migratetype, int *num_movable)
2236 {
2237 	struct page *page;
2238 	unsigned int order;
2239 	int pages_moved = 0;
2240 
2241 	for (page = start_page; page <= end_page;) {
2242 		if (!pfn_valid_within(page_to_pfn(page))) {
2243 			page++;
2244 			continue;
2245 		}
2246 
2247 		if (!PageBuddy(page)) {
2248 			/*
2249 			 * We assume that pages that could be isolated for
2250 			 * migration are movable. But we don't actually try
2251 			 * isolating, as that would be expensive.
2252 			 */
2253 			if (num_movable &&
2254 					(PageLRU(page) || __PageMovable(page)))
2255 				(*num_movable)++;
2256 
2257 			page++;
2258 			continue;
2259 		}
2260 
2261 		/* Make sure we are not inadvertently changing nodes */
2262 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2263 		VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2264 
2265 		order = page_order(page);
2266 		move_to_free_area(page, &zone->free_area[order], migratetype);
2267 		page += 1 << order;
2268 		pages_moved += 1 << order;
2269 	}
2270 
2271 	return pages_moved;
2272 }
2273 
2274 int move_freepages_block(struct zone *zone, struct page *page,
2275 				int migratetype, int *num_movable)
2276 {
2277 	unsigned long start_pfn, end_pfn;
2278 	struct page *start_page, *end_page;
2279 
2280 	if (num_movable)
2281 		*num_movable = 0;
2282 
2283 	start_pfn = page_to_pfn(page);
2284 	start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2285 	start_page = pfn_to_page(start_pfn);
2286 	end_page = start_page + pageblock_nr_pages - 1;
2287 	end_pfn = start_pfn + pageblock_nr_pages - 1;
2288 
2289 	/* Do not cross zone boundaries */
2290 	if (!zone_spans_pfn(zone, start_pfn))
2291 		start_page = page;
2292 	if (!zone_spans_pfn(zone, end_pfn))
2293 		return 0;
2294 
2295 	return move_freepages(zone, start_page, end_page, migratetype,
2296 								num_movable);
2297 }
2298 
2299 static void change_pageblock_range(struct page *pageblock_page,
2300 					int start_order, int migratetype)
2301 {
2302 	int nr_pageblocks = 1 << (start_order - pageblock_order);
2303 
2304 	while (nr_pageblocks--) {
2305 		set_pageblock_migratetype(pageblock_page, migratetype);
2306 		pageblock_page += pageblock_nr_pages;
2307 	}
2308 }
2309 
2310 /*
2311  * When we are falling back to another migratetype during allocation, try to
2312  * steal extra free pages from the same pageblocks to satisfy further
2313  * allocations, instead of polluting multiple pageblocks.
2314  *
2315  * If we are stealing a relatively large buddy page, it is likely there will
2316  * be more free pages in the pageblock, so try to steal them all. For
2317  * reclaimable and unmovable allocations, we steal regardless of page size,
2318  * as fragmentation caused by those allocations polluting movable pageblocks
2319  * is worse than movable allocations stealing from unmovable and reclaimable
2320  * pageblocks.
2321  */
2322 static bool can_steal_fallback(unsigned int order, int start_mt)
2323 {
2324 	/*
2325 	 * Leaving this order check is intended, although there is
2326 	 * relaxed order check in next check. The reason is that
2327 	 * we can actually steal whole pageblock if this condition met,
2328 	 * but, below check doesn't guarantee it and that is just heuristic
2329 	 * so could be changed anytime.
2330 	 */
2331 	if (order >= pageblock_order)
2332 		return true;
2333 
2334 	if (order >= pageblock_order / 2 ||
2335 		start_mt == MIGRATE_RECLAIMABLE ||
2336 		start_mt == MIGRATE_UNMOVABLE ||
2337 		page_group_by_mobility_disabled)
2338 		return true;
2339 
2340 	return false;
2341 }
2342 
2343 static inline void boost_watermark(struct zone *zone)
2344 {
2345 	unsigned long max_boost;
2346 
2347 	if (!watermark_boost_factor)
2348 		return;
2349 
2350 	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2351 			watermark_boost_factor, 10000);
2352 
2353 	/*
2354 	 * high watermark may be uninitialised if fragmentation occurs
2355 	 * very early in boot so do not boost. We do not fall
2356 	 * through and boost by pageblock_nr_pages as failing
2357 	 * allocations that early means that reclaim is not going
2358 	 * to help and it may even be impossible to reclaim the
2359 	 * boosted watermark resulting in a hang.
2360 	 */
2361 	if (!max_boost)
2362 		return;
2363 
2364 	max_boost = max(pageblock_nr_pages, max_boost);
2365 
2366 	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2367 		max_boost);
2368 }
2369 
2370 /*
2371  * This function implements actual steal behaviour. If order is large enough,
2372  * we can steal whole pageblock. If not, we first move freepages in this
2373  * pageblock to our migratetype and determine how many already-allocated pages
2374  * are there in the pageblock with a compatible migratetype. If at least half
2375  * of pages are free or compatible, we can change migratetype of the pageblock
2376  * itself, so pages freed in the future will be put on the correct free list.
2377  */
2378 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2379 		unsigned int alloc_flags, int start_type, bool whole_block)
2380 {
2381 	unsigned int current_order = page_order(page);
2382 	struct free_area *area;
2383 	int free_pages, movable_pages, alike_pages;
2384 	int old_block_type;
2385 
2386 	old_block_type = get_pageblock_migratetype(page);
2387 
2388 	/*
2389 	 * This can happen due to races and we want to prevent broken
2390 	 * highatomic accounting.
2391 	 */
2392 	if (is_migrate_highatomic(old_block_type))
2393 		goto single_page;
2394 
2395 	/* Take ownership for orders >= pageblock_order */
2396 	if (current_order >= pageblock_order) {
2397 		change_pageblock_range(page, current_order, start_type);
2398 		goto single_page;
2399 	}
2400 
2401 	/*
2402 	 * Boost watermarks to increase reclaim pressure to reduce the
2403 	 * likelihood of future fallbacks. Wake kswapd now as the node
2404 	 * may be balanced overall and kswapd will not wake naturally.
2405 	 */
2406 	boost_watermark(zone);
2407 	if (alloc_flags & ALLOC_KSWAPD)
2408 		set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2409 
2410 	/* We are not allowed to try stealing from the whole block */
2411 	if (!whole_block)
2412 		goto single_page;
2413 
2414 	free_pages = move_freepages_block(zone, page, start_type,
2415 						&movable_pages);
2416 	/*
2417 	 * Determine how many pages are compatible with our allocation.
2418 	 * For movable allocation, it's the number of movable pages which
2419 	 * we just obtained. For other types it's a bit more tricky.
2420 	 */
2421 	if (start_type == MIGRATE_MOVABLE) {
2422 		alike_pages = movable_pages;
2423 	} else {
2424 		/*
2425 		 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2426 		 * to MOVABLE pageblock, consider all non-movable pages as
2427 		 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2428 		 * vice versa, be conservative since we can't distinguish the
2429 		 * exact migratetype of non-movable pages.
2430 		 */
2431 		if (old_block_type == MIGRATE_MOVABLE)
2432 			alike_pages = pageblock_nr_pages
2433 						- (free_pages + movable_pages);
2434 		else
2435 			alike_pages = 0;
2436 	}
2437 
2438 	/* moving whole block can fail due to zone boundary conditions */
2439 	if (!free_pages)
2440 		goto single_page;
2441 
2442 	/*
2443 	 * If a sufficient number of pages in the block are either free or of
2444 	 * comparable migratability as our allocation, claim the whole block.
2445 	 */
2446 	if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2447 			page_group_by_mobility_disabled)
2448 		set_pageblock_migratetype(page, start_type);
2449 
2450 	return;
2451 
2452 single_page:
2453 	area = &zone->free_area[current_order];
2454 	move_to_free_area(page, area, start_type);
2455 }
2456 
2457 /*
2458  * Check whether there is a suitable fallback freepage with requested order.
2459  * If only_stealable is true, this function returns fallback_mt only if
2460  * we can steal other freepages all together. This would help to reduce
2461  * fragmentation due to mixed migratetype pages in one pageblock.
2462  */
2463 int find_suitable_fallback(struct free_area *area, unsigned int order,
2464 			int migratetype, bool only_stealable, bool *can_steal)
2465 {
2466 	int i;
2467 	int fallback_mt;
2468 
2469 	if (area->nr_free == 0)
2470 		return -1;
2471 
2472 	*can_steal = false;
2473 	for (i = 0;; i++) {
2474 		fallback_mt = fallbacks[migratetype][i];
2475 		if (fallback_mt == MIGRATE_TYPES)
2476 			break;
2477 
2478 		if (free_area_empty(area, fallback_mt))
2479 			continue;
2480 
2481 		if (can_steal_fallback(order, migratetype))
2482 			*can_steal = true;
2483 
2484 		if (!only_stealable)
2485 			return fallback_mt;
2486 
2487 		if (*can_steal)
2488 			return fallback_mt;
2489 	}
2490 
2491 	return -1;
2492 }
2493 
2494 /*
2495  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2496  * there are no empty page blocks that contain a page with a suitable order
2497  */
2498 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2499 				unsigned int alloc_order)
2500 {
2501 	int mt;
2502 	unsigned long max_managed, flags;
2503 
2504 	/*
2505 	 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2506 	 * Check is race-prone but harmless.
2507 	 */
2508 	max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2509 	if (zone->nr_reserved_highatomic >= max_managed)
2510 		return;
2511 
2512 	spin_lock_irqsave(&zone->lock, flags);
2513 
2514 	/* Recheck the nr_reserved_highatomic limit under the lock */
2515 	if (zone->nr_reserved_highatomic >= max_managed)
2516 		goto out_unlock;
2517 
2518 	/* Yoink! */
2519 	mt = get_pageblock_migratetype(page);
2520 	if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2521 	    && !is_migrate_cma(mt)) {
2522 		zone->nr_reserved_highatomic += pageblock_nr_pages;
2523 		set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2524 		move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2525 	}
2526 
2527 out_unlock:
2528 	spin_unlock_irqrestore(&zone->lock, flags);
2529 }
2530 
2531 /*
2532  * Used when an allocation is about to fail under memory pressure. This
2533  * potentially hurts the reliability of high-order allocations when under
2534  * intense memory pressure but failed atomic allocations should be easier
2535  * to recover from than an OOM.
2536  *
2537  * If @force is true, try to unreserve a pageblock even though highatomic
2538  * pageblock is exhausted.
2539  */
2540 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2541 						bool force)
2542 {
2543 	struct zonelist *zonelist = ac->zonelist;
2544 	unsigned long flags;
2545 	struct zoneref *z;
2546 	struct zone *zone;
2547 	struct page *page;
2548 	int order;
2549 	bool ret;
2550 
2551 	for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2552 								ac->nodemask) {
2553 		/*
2554 		 * Preserve at least one pageblock unless memory pressure
2555 		 * is really high.
2556 		 */
2557 		if (!force && zone->nr_reserved_highatomic <=
2558 					pageblock_nr_pages)
2559 			continue;
2560 
2561 		spin_lock_irqsave(&zone->lock, flags);
2562 		for (order = 0; order < MAX_ORDER; order++) {
2563 			struct free_area *area = &(zone->free_area[order]);
2564 
2565 			page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2566 			if (!page)
2567 				continue;
2568 
2569 			/*
2570 			 * In page freeing path, migratetype change is racy so
2571 			 * we can counter several free pages in a pageblock
2572 			 * in this loop althoug we changed the pageblock type
2573 			 * from highatomic to ac->migratetype. So we should
2574 			 * adjust the count once.
2575 			 */
2576 			if (is_migrate_highatomic_page(page)) {
2577 				/*
2578 				 * It should never happen but changes to
2579 				 * locking could inadvertently allow a per-cpu
2580 				 * drain to add pages to MIGRATE_HIGHATOMIC
2581 				 * while unreserving so be safe and watch for
2582 				 * underflows.
2583 				 */
2584 				zone->nr_reserved_highatomic -= min(
2585 						pageblock_nr_pages,
2586 						zone->nr_reserved_highatomic);
2587 			}
2588 
2589 			/*
2590 			 * Convert to ac->migratetype and avoid the normal
2591 			 * pageblock stealing heuristics. Minimally, the caller
2592 			 * is doing the work and needs the pages. More
2593 			 * importantly, if the block was always converted to
2594 			 * MIGRATE_UNMOVABLE or another type then the number
2595 			 * of pageblocks that cannot be completely freed
2596 			 * may increase.
2597 			 */
2598 			set_pageblock_migratetype(page, ac->migratetype);
2599 			ret = move_freepages_block(zone, page, ac->migratetype,
2600 									NULL);
2601 			if (ret) {
2602 				spin_unlock_irqrestore(&zone->lock, flags);
2603 				return ret;
2604 			}
2605 		}
2606 		spin_unlock_irqrestore(&zone->lock, flags);
2607 	}
2608 
2609 	return false;
2610 }
2611 
2612 /*
2613  * Try finding a free buddy page on the fallback list and put it on the free
2614  * list of requested migratetype, possibly along with other pages from the same
2615  * block, depending on fragmentation avoidance heuristics. Returns true if
2616  * fallback was found so that __rmqueue_smallest() can grab it.
2617  *
2618  * The use of signed ints for order and current_order is a deliberate
2619  * deviation from the rest of this file, to make the for loop
2620  * condition simpler.
2621  */
2622 static __always_inline bool
2623 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2624 						unsigned int alloc_flags)
2625 {
2626 	struct free_area *area;
2627 	int current_order;
2628 	int min_order = order;
2629 	struct page *page;
2630 	int fallback_mt;
2631 	bool can_steal;
2632 
2633 	/*
2634 	 * Do not steal pages from freelists belonging to other pageblocks
2635 	 * i.e. orders < pageblock_order. If there are no local zones free,
2636 	 * the zonelists will be reiterated without ALLOC_NOFRAGMENT.
2637 	 */
2638 	if (alloc_flags & ALLOC_NOFRAGMENT)
2639 		min_order = pageblock_order;
2640 
2641 	/*
2642 	 * Find the largest available free page in the other list. This roughly
2643 	 * approximates finding the pageblock with the most free pages, which
2644 	 * would be too costly to do exactly.
2645 	 */
2646 	for (current_order = MAX_ORDER - 1; current_order >= min_order;
2647 				--current_order) {
2648 		area = &(zone->free_area[current_order]);
2649 		fallback_mt = find_suitable_fallback(area, current_order,
2650 				start_migratetype, false, &can_steal);
2651 		if (fallback_mt == -1)
2652 			continue;
2653 
2654 		/*
2655 		 * We cannot steal all free pages from the pageblock and the
2656 		 * requested migratetype is movable. In that case it's better to
2657 		 * steal and split the smallest available page instead of the
2658 		 * largest available page, because even if the next movable
2659 		 * allocation falls back into a different pageblock than this
2660 		 * one, it won't cause permanent fragmentation.
2661 		 */
2662 		if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2663 					&& current_order > order)
2664 			goto find_smallest;
2665 
2666 		goto do_steal;
2667 	}
2668 
2669 	return false;
2670 
2671 find_smallest:
2672 	for (current_order = order; current_order < MAX_ORDER;
2673 							current_order++) {
2674 		area = &(zone->free_area[current_order]);
2675 		fallback_mt = find_suitable_fallback(area, current_order,
2676 				start_migratetype, false, &can_steal);
2677 		if (fallback_mt != -1)
2678 			break;
2679 	}
2680 
2681 	/*
2682 	 * This should not happen - we already found a suitable fallback
2683 	 * when looking for the largest page.
2684 	 */
2685 	VM_BUG_ON(current_order == MAX_ORDER);
2686 
2687 do_steal:
2688 	page = get_page_from_free_area(area, fallback_mt);
2689 
2690 	steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2691 								can_steal);
2692 
2693 	trace_mm_page_alloc_extfrag(page, order, current_order,
2694 		start_migratetype, fallback_mt);
2695 
2696 	return true;
2697 
2698 }
2699 
2700 /*
2701  * Do the hard work of removing an element from the buddy allocator.
2702  * Call me with the zone->lock already held.
2703  */
2704 static __always_inline struct page *
2705 __rmqueue(struct zone *zone, unsigned int order, int migratetype,
2706 						unsigned int alloc_flags)
2707 {
2708 	struct page *page;
2709 
2710 retry:
2711 	page = __rmqueue_smallest(zone, order, migratetype);
2712 	if (unlikely(!page)) {
2713 		if (migratetype == MIGRATE_MOVABLE)
2714 			page = __rmqueue_cma_fallback(zone, order);
2715 
2716 		if (!page && __rmqueue_fallback(zone, order, migratetype,
2717 								alloc_flags))
2718 			goto retry;
2719 	}
2720 
2721 	trace_mm_page_alloc_zone_locked(page, order, migratetype);
2722 	return page;
2723 }
2724 
2725 /*
2726  * Obtain a specified number of elements from the buddy allocator, all under
2727  * a single hold of the lock, for efficiency.  Add them to the supplied list.
2728  * Returns the number of new pages which were placed at *list.
2729  */
2730 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2731 			unsigned long count, struct list_head *list,
2732 			int migratetype, unsigned int alloc_flags)
2733 {
2734 	int i, alloced = 0;
2735 
2736 	spin_lock(&zone->lock);
2737 	for (i = 0; i < count; ++i) {
2738 		struct page *page = __rmqueue(zone, order, migratetype,
2739 								alloc_flags);
2740 		if (unlikely(page == NULL))
2741 			break;
2742 
2743 		if (unlikely(check_pcp_refill(page)))
2744 			continue;
2745 
2746 		/*
2747 		 * Split buddy pages returned by expand() are received here in
2748 		 * physical page order. The page is added to the tail of
2749 		 * caller's list. From the callers perspective, the linked list
2750 		 * is ordered by page number under some conditions. This is
2751 		 * useful for IO devices that can forward direction from the
2752 		 * head, thus also in the physical page order. This is useful
2753 		 * for IO devices that can merge IO requests if the physical
2754 		 * pages are ordered properly.
2755 		 */
2756 		list_add_tail(&page->lru, list);
2757 		alloced++;
2758 		if (is_migrate_cma(get_pcppage_migratetype(page)))
2759 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2760 					      -(1 << order));
2761 	}
2762 
2763 	/*
2764 	 * i pages were removed from the buddy list even if some leak due
2765 	 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2766 	 * on i. Do not confuse with 'alloced' which is the number of
2767 	 * pages added to the pcp list.
2768 	 */
2769 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2770 	spin_unlock(&zone->lock);
2771 	return alloced;
2772 }
2773 
2774 #ifdef CONFIG_NUMA
2775 /*
2776  * Called from the vmstat counter updater to drain pagesets of this
2777  * currently executing processor on remote nodes after they have
2778  * expired.
2779  *
2780  * Note that this function must be called with the thread pinned to
2781  * a single processor.
2782  */
2783 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2784 {
2785 	unsigned long flags;
2786 	int to_drain, batch;
2787 
2788 	local_irq_save(flags);
2789 	batch = READ_ONCE(pcp->batch);
2790 	to_drain = min(pcp->count, batch);
2791 	if (to_drain > 0)
2792 		free_pcppages_bulk(zone, to_drain, pcp);
2793 	local_irq_restore(flags);
2794 }
2795 #endif
2796 
2797 /*
2798  * Drain pcplists of the indicated processor and zone.
2799  *
2800  * The processor must either be the current processor and the
2801  * thread pinned to the current processor or a processor that
2802  * is not online.
2803  */
2804 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2805 {
2806 	unsigned long flags;
2807 	struct per_cpu_pageset *pset;
2808 	struct per_cpu_pages *pcp;
2809 
2810 	local_irq_save(flags);
2811 	pset = per_cpu_ptr(zone->pageset, cpu);
2812 
2813 	pcp = &pset->pcp;
2814 	if (pcp->count)
2815 		free_pcppages_bulk(zone, pcp->count, pcp);
2816 	local_irq_restore(flags);
2817 }
2818 
2819 /*
2820  * Drain pcplists of all zones on the indicated processor.
2821  *
2822  * The processor must either be the current processor and the
2823  * thread pinned to the current processor or a processor that
2824  * is not online.
2825  */
2826 static void drain_pages(unsigned int cpu)
2827 {
2828 	struct zone *zone;
2829 
2830 	for_each_populated_zone(zone) {
2831 		drain_pages_zone(cpu, zone);
2832 	}
2833 }
2834 
2835 /*
2836  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2837  *
2838  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2839  * the single zone's pages.
2840  */
2841 void drain_local_pages(struct zone *zone)
2842 {
2843 	int cpu = smp_processor_id();
2844 
2845 	if (zone)
2846 		drain_pages_zone(cpu, zone);
2847 	else
2848 		drain_pages(cpu);
2849 }
2850 
2851 static void drain_local_pages_wq(struct work_struct *work)
2852 {
2853 	struct pcpu_drain *drain;
2854 
2855 	drain = container_of(work, struct pcpu_drain, work);
2856 
2857 	/*
2858 	 * drain_all_pages doesn't use proper cpu hotplug protection so
2859 	 * we can race with cpu offline when the WQ can move this from
2860 	 * a cpu pinned worker to an unbound one. We can operate on a different
2861 	 * cpu which is allright but we also have to make sure to not move to
2862 	 * a different one.
2863 	 */
2864 	preempt_disable();
2865 	drain_local_pages(drain->zone);
2866 	preempt_enable();
2867 }
2868 
2869 /*
2870  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2871  *
2872  * When zone parameter is non-NULL, spill just the single zone's pages.
2873  *
2874  * Note that this can be extremely slow as the draining happens in a workqueue.
2875  */
2876 void drain_all_pages(struct zone *zone)
2877 {
2878 	int cpu;
2879 
2880 	/*
2881 	 * Allocate in the BSS so we wont require allocation in
2882 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2883 	 */
2884 	static cpumask_t cpus_with_pcps;
2885 
2886 	/*
2887 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
2888 	 * initialized.
2889 	 */
2890 	if (WARN_ON_ONCE(!mm_percpu_wq))
2891 		return;
2892 
2893 	/*
2894 	 * Do not drain if one is already in progress unless it's specific to
2895 	 * a zone. Such callers are primarily CMA and memory hotplug and need
2896 	 * the drain to be complete when the call returns.
2897 	 */
2898 	if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2899 		if (!zone)
2900 			return;
2901 		mutex_lock(&pcpu_drain_mutex);
2902 	}
2903 
2904 	/*
2905 	 * We don't care about racing with CPU hotplug event
2906 	 * as offline notification will cause the notified
2907 	 * cpu to drain that CPU pcps and on_each_cpu_mask
2908 	 * disables preemption as part of its processing
2909 	 */
2910 	for_each_online_cpu(cpu) {
2911 		struct per_cpu_pageset *pcp;
2912 		struct zone *z;
2913 		bool has_pcps = false;
2914 
2915 		if (zone) {
2916 			pcp = per_cpu_ptr(zone->pageset, cpu);
2917 			if (pcp->pcp.count)
2918 				has_pcps = true;
2919 		} else {
2920 			for_each_populated_zone(z) {
2921 				pcp = per_cpu_ptr(z->pageset, cpu);
2922 				if (pcp->pcp.count) {
2923 					has_pcps = true;
2924 					break;
2925 				}
2926 			}
2927 		}
2928 
2929 		if (has_pcps)
2930 			cpumask_set_cpu(cpu, &cpus_with_pcps);
2931 		else
2932 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
2933 	}
2934 
2935 	for_each_cpu(cpu, &cpus_with_pcps) {
2936 		struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
2937 
2938 		drain->zone = zone;
2939 		INIT_WORK(&drain->work, drain_local_pages_wq);
2940 		queue_work_on(cpu, mm_percpu_wq, &drain->work);
2941 	}
2942 	for_each_cpu(cpu, &cpus_with_pcps)
2943 		flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
2944 
2945 	mutex_unlock(&pcpu_drain_mutex);
2946 }
2947 
2948 #ifdef CONFIG_HIBERNATION
2949 
2950 /*
2951  * Touch the watchdog for every WD_PAGE_COUNT pages.
2952  */
2953 #define WD_PAGE_COUNT	(128*1024)
2954 
2955 void mark_free_pages(struct zone *zone)
2956 {
2957 	unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
2958 	unsigned long flags;
2959 	unsigned int order, t;
2960 	struct page *page;
2961 
2962 	if (zone_is_empty(zone))
2963 		return;
2964 
2965 	spin_lock_irqsave(&zone->lock, flags);
2966 
2967 	max_zone_pfn = zone_end_pfn(zone);
2968 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2969 		if (pfn_valid(pfn)) {
2970 			page = pfn_to_page(pfn);
2971 
2972 			if (!--page_count) {
2973 				touch_nmi_watchdog();
2974 				page_count = WD_PAGE_COUNT;
2975 			}
2976 
2977 			if (page_zone(page) != zone)
2978 				continue;
2979 
2980 			if (!swsusp_page_is_forbidden(page))
2981 				swsusp_unset_page_free(page);
2982 		}
2983 
2984 	for_each_migratetype_order(order, t) {
2985 		list_for_each_entry(page,
2986 				&zone->free_area[order].free_list[t], lru) {
2987 			unsigned long i;
2988 
2989 			pfn = page_to_pfn(page);
2990 			for (i = 0; i < (1UL << order); i++) {
2991 				if (!--page_count) {
2992 					touch_nmi_watchdog();
2993 					page_count = WD_PAGE_COUNT;
2994 				}
2995 				swsusp_set_page_free(pfn_to_page(pfn + i));
2996 			}
2997 		}
2998 	}
2999 	spin_unlock_irqrestore(&zone->lock, flags);
3000 }
3001 #endif /* CONFIG_PM */
3002 
3003 static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
3004 {
3005 	int migratetype;
3006 
3007 	if (!free_pcp_prepare(page))
3008 		return false;
3009 
3010 	migratetype = get_pfnblock_migratetype(page, pfn);
3011 	set_pcppage_migratetype(page, migratetype);
3012 	return true;
3013 }
3014 
3015 static void free_unref_page_commit(struct page *page, unsigned long pfn)
3016 {
3017 	struct zone *zone = page_zone(page);
3018 	struct per_cpu_pages *pcp;
3019 	int migratetype;
3020 
3021 	migratetype = get_pcppage_migratetype(page);
3022 	__count_vm_event(PGFREE);
3023 
3024 	/*
3025 	 * We only track unmovable, reclaimable and movable on pcp lists.
3026 	 * Free ISOLATE pages back to the allocator because they are being
3027 	 * offlined but treat HIGHATOMIC as movable pages so we can get those
3028 	 * areas back if necessary. Otherwise, we may have to free
3029 	 * excessively into the page allocator
3030 	 */
3031 	if (migratetype >= MIGRATE_PCPTYPES) {
3032 		if (unlikely(is_migrate_isolate(migratetype))) {
3033 			free_one_page(zone, page, pfn, 0, migratetype);
3034 			return;
3035 		}
3036 		migratetype = MIGRATE_MOVABLE;
3037 	}
3038 
3039 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
3040 	list_add(&page->lru, &pcp->lists[migratetype]);
3041 	pcp->count++;
3042 	if (pcp->count >= pcp->high) {
3043 		unsigned long batch = READ_ONCE(pcp->batch);
3044 		free_pcppages_bulk(zone, batch, pcp);
3045 	}
3046 }
3047 
3048 /*
3049  * Free a 0-order page
3050  */
3051 void free_unref_page(struct page *page)
3052 {
3053 	unsigned long flags;
3054 	unsigned long pfn = page_to_pfn(page);
3055 
3056 	if (!free_unref_page_prepare(page, pfn))
3057 		return;
3058 
3059 	local_irq_save(flags);
3060 	free_unref_page_commit(page, pfn);
3061 	local_irq_restore(flags);
3062 }
3063 
3064 /*
3065  * Free a list of 0-order pages
3066  */
3067 void free_unref_page_list(struct list_head *list)
3068 {
3069 	struct page *page, *next;
3070 	unsigned long flags, pfn;
3071 	int batch_count = 0;
3072 
3073 	/* Prepare pages for freeing */
3074 	list_for_each_entry_safe(page, next, list, lru) {
3075 		pfn = page_to_pfn(page);
3076 		if (!free_unref_page_prepare(page, pfn))
3077 			list_del(&page->lru);
3078 		set_page_private(page, pfn);
3079 	}
3080 
3081 	local_irq_save(flags);
3082 	list_for_each_entry_safe(page, next, list, lru) {
3083 		unsigned long pfn = page_private(page);
3084 
3085 		set_page_private(page, 0);
3086 		trace_mm_page_free_batched(page);
3087 		free_unref_page_commit(page, pfn);
3088 
3089 		/*
3090 		 * Guard against excessive IRQ disabled times when we get
3091 		 * a large list of pages to free.
3092 		 */
3093 		if (++batch_count == SWAP_CLUSTER_MAX) {
3094 			local_irq_restore(flags);
3095 			batch_count = 0;
3096 			local_irq_save(flags);
3097 		}
3098 	}
3099 	local_irq_restore(flags);
3100 }
3101 
3102 /*
3103  * split_page takes a non-compound higher-order page, and splits it into
3104  * n (1<<order) sub-pages: page[0..n]
3105  * Each sub-page must be freed individually.
3106  *
3107  * Note: this is probably too low level an operation for use in drivers.
3108  * Please consult with lkml before using this in your driver.
3109  */
3110 void split_page(struct page *page, unsigned int order)
3111 {
3112 	int i;
3113 
3114 	VM_BUG_ON_PAGE(PageCompound(page), page);
3115 	VM_BUG_ON_PAGE(!page_count(page), page);
3116 
3117 	for (i = 1; i < (1 << order); i++)
3118 		set_page_refcounted(page + i);
3119 	split_page_owner(page, order);
3120 }
3121 EXPORT_SYMBOL_GPL(split_page);
3122 
3123 int __isolate_free_page(struct page *page, unsigned int order)
3124 {
3125 	struct free_area *area = &page_zone(page)->free_area[order];
3126 	unsigned long watermark;
3127 	struct zone *zone;
3128 	int mt;
3129 
3130 	BUG_ON(!PageBuddy(page));
3131 
3132 	zone = page_zone(page);
3133 	mt = get_pageblock_migratetype(page);
3134 
3135 	if (!is_migrate_isolate(mt)) {
3136 		/*
3137 		 * Obey watermarks as if the page was being allocated. We can
3138 		 * emulate a high-order watermark check with a raised order-0
3139 		 * watermark, because we already know our high-order page
3140 		 * exists.
3141 		 */
3142 		watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3143 		if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3144 			return 0;
3145 
3146 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
3147 	}
3148 
3149 	/* Remove page from free list */
3150 
3151 	del_page_from_free_area(page, area);
3152 
3153 	/*
3154 	 * Set the pageblock if the isolated page is at least half of a
3155 	 * pageblock
3156 	 */
3157 	if (order >= pageblock_order - 1) {
3158 		struct page *endpage = page + (1 << order) - 1;
3159 		for (; page < endpage; page += pageblock_nr_pages) {
3160 			int mt = get_pageblock_migratetype(page);
3161 			if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3162 			    && !is_migrate_highatomic(mt))
3163 				set_pageblock_migratetype(page,
3164 							  MIGRATE_MOVABLE);
3165 		}
3166 	}
3167 
3168 
3169 	return 1UL << order;
3170 }
3171 
3172 /*
3173  * Update NUMA hit/miss statistics
3174  *
3175  * Must be called with interrupts disabled.
3176  */
3177 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
3178 {
3179 #ifdef CONFIG_NUMA
3180 	enum numa_stat_item local_stat = NUMA_LOCAL;
3181 
3182 	/* skip numa counters update if numa stats is disabled */
3183 	if (!static_branch_likely(&vm_numa_stat_key))
3184 		return;
3185 
3186 	if (zone_to_nid(z) != numa_node_id())
3187 		local_stat = NUMA_OTHER;
3188 
3189 	if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3190 		__inc_numa_state(z, NUMA_HIT);
3191 	else {
3192 		__inc_numa_state(z, NUMA_MISS);
3193 		__inc_numa_state(preferred_zone, NUMA_FOREIGN);
3194 	}
3195 	__inc_numa_state(z, local_stat);
3196 #endif
3197 }
3198 
3199 /* Remove page from the per-cpu list, caller must protect the list */
3200 static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
3201 			unsigned int alloc_flags,
3202 			struct per_cpu_pages *pcp,
3203 			struct list_head *list)
3204 {
3205 	struct page *page;
3206 
3207 	do {
3208 		if (list_empty(list)) {
3209 			pcp->count += rmqueue_bulk(zone, 0,
3210 					pcp->batch, list,
3211 					migratetype, alloc_flags);
3212 			if (unlikely(list_empty(list)))
3213 				return NULL;
3214 		}
3215 
3216 		page = list_first_entry(list, struct page, lru);
3217 		list_del(&page->lru);
3218 		pcp->count--;
3219 	} while (check_new_pcp(page));
3220 
3221 	return page;
3222 }
3223 
3224 /* Lock and remove page from the per-cpu list */
3225 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3226 			struct zone *zone, gfp_t gfp_flags,
3227 			int migratetype, unsigned int alloc_flags)
3228 {
3229 	struct per_cpu_pages *pcp;
3230 	struct list_head *list;
3231 	struct page *page;
3232 	unsigned long flags;
3233 
3234 	local_irq_save(flags);
3235 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
3236 	list = &pcp->lists[migratetype];
3237 	page = __rmqueue_pcplist(zone,  migratetype, alloc_flags, pcp, list);
3238 	if (page) {
3239 		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3240 		zone_statistics(preferred_zone, zone);
3241 	}
3242 	local_irq_restore(flags);
3243 	return page;
3244 }
3245 
3246 /*
3247  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
3248  */
3249 static inline
3250 struct page *rmqueue(struct zone *preferred_zone,
3251 			struct zone *zone, unsigned int order,
3252 			gfp_t gfp_flags, unsigned int alloc_flags,
3253 			int migratetype)
3254 {
3255 	unsigned long flags;
3256 	struct page *page;
3257 
3258 	if (likely(order == 0)) {
3259 		page = rmqueue_pcplist(preferred_zone, zone, gfp_flags,
3260 					migratetype, alloc_flags);
3261 		goto out;
3262 	}
3263 
3264 	/*
3265 	 * We most definitely don't want callers attempting to
3266 	 * allocate greater than order-1 page units with __GFP_NOFAIL.
3267 	 */
3268 	WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3269 	spin_lock_irqsave(&zone->lock, flags);
3270 
3271 	do {
3272 		page = NULL;
3273 		if (alloc_flags & ALLOC_HARDER) {
3274 			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3275 			if (page)
3276 				trace_mm_page_alloc_zone_locked(page, order, migratetype);
3277 		}
3278 		if (!page)
3279 			page = __rmqueue(zone, order, migratetype, alloc_flags);
3280 	} while (page && check_new_pages(page, order));
3281 	spin_unlock(&zone->lock);
3282 	if (!page)
3283 		goto failed;
3284 	__mod_zone_freepage_state(zone, -(1 << order),
3285 				  get_pcppage_migratetype(page));
3286 
3287 	__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3288 	zone_statistics(preferred_zone, zone);
3289 	local_irq_restore(flags);
3290 
3291 out:
3292 	/* Separate test+clear to avoid unnecessary atomics */
3293 	if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3294 		clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3295 		wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3296 	}
3297 
3298 	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3299 	return page;
3300 
3301 failed:
3302 	local_irq_restore(flags);
3303 	return NULL;
3304 }
3305 
3306 #ifdef CONFIG_FAIL_PAGE_ALLOC
3307 
3308 static struct {
3309 	struct fault_attr attr;
3310 
3311 	bool ignore_gfp_highmem;
3312 	bool ignore_gfp_reclaim;
3313 	u32 min_order;
3314 } fail_page_alloc = {
3315 	.attr = FAULT_ATTR_INITIALIZER,
3316 	.ignore_gfp_reclaim = true,
3317 	.ignore_gfp_highmem = true,
3318 	.min_order = 1,
3319 };
3320 
3321 static int __init setup_fail_page_alloc(char *str)
3322 {
3323 	return setup_fault_attr(&fail_page_alloc.attr, str);
3324 }
3325 __setup("fail_page_alloc=", setup_fail_page_alloc);
3326 
3327 static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3328 {
3329 	if (order < fail_page_alloc.min_order)
3330 		return false;
3331 	if (gfp_mask & __GFP_NOFAIL)
3332 		return false;
3333 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3334 		return false;
3335 	if (fail_page_alloc.ignore_gfp_reclaim &&
3336 			(gfp_mask & __GFP_DIRECT_RECLAIM))
3337 		return false;
3338 
3339 	return should_fail(&fail_page_alloc.attr, 1 << order);
3340 }
3341 
3342 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3343 
3344 static int __init fail_page_alloc_debugfs(void)
3345 {
3346 	umode_t mode = S_IFREG | 0600;
3347 	struct dentry *dir;
3348 
3349 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3350 					&fail_page_alloc.attr);
3351 
3352 	debugfs_create_bool("ignore-gfp-wait", mode, dir,
3353 			    &fail_page_alloc.ignore_gfp_reclaim);
3354 	debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3355 			    &fail_page_alloc.ignore_gfp_highmem);
3356 	debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3357 
3358 	return 0;
3359 }
3360 
3361 late_initcall(fail_page_alloc_debugfs);
3362 
3363 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3364 
3365 #else /* CONFIG_FAIL_PAGE_ALLOC */
3366 
3367 static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3368 {
3369 	return false;
3370 }
3371 
3372 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3373 
3374 static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3375 {
3376 	return __should_fail_alloc_page(gfp_mask, order);
3377 }
3378 ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3379 
3380 /*
3381  * Return true if free base pages are above 'mark'. For high-order checks it
3382  * will return true of the order-0 watermark is reached and there is at least
3383  * one free page of a suitable size. Checking now avoids taking the zone lock
3384  * to check in the allocation paths if no pages are free.
3385  */
3386 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3387 			 int classzone_idx, unsigned int alloc_flags,
3388 			 long free_pages)
3389 {
3390 	long min = mark;
3391 	int o;
3392 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3393 
3394 	/* free_pages may go negative - that's OK */
3395 	free_pages -= (1 << order) - 1;
3396 
3397 	if (alloc_flags & ALLOC_HIGH)
3398 		min -= min / 2;
3399 
3400 	/*
3401 	 * If the caller does not have rights to ALLOC_HARDER then subtract
3402 	 * the high-atomic reserves. This will over-estimate the size of the
3403 	 * atomic reserve but it avoids a search.
3404 	 */
3405 	if (likely(!alloc_harder)) {
3406 		free_pages -= z->nr_reserved_highatomic;
3407 	} else {
3408 		/*
3409 		 * OOM victims can try even harder than normal ALLOC_HARDER
3410 		 * users on the grounds that it's definitely going to be in
3411 		 * the exit path shortly and free memory. Any allocation it
3412 		 * makes during the free path will be small and short-lived.
3413 		 */
3414 		if (alloc_flags & ALLOC_OOM)
3415 			min -= min / 2;
3416 		else
3417 			min -= min / 4;
3418 	}
3419 
3420 
3421 #ifdef CONFIG_CMA
3422 	/* If allocation can't use CMA areas don't use free CMA pages */
3423 	if (!(alloc_flags & ALLOC_CMA))
3424 		free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
3425 #endif
3426 
3427 	/*
3428 	 * Check watermarks for an order-0 allocation request. If these
3429 	 * are not met, then a high-order request also cannot go ahead
3430 	 * even if a suitable page happened to be free.
3431 	 */
3432 	if (free_pages <= min + z->lowmem_reserve[classzone_idx])
3433 		return false;
3434 
3435 	/* If this is an order-0 request then the watermark is fine */
3436 	if (!order)
3437 		return true;
3438 
3439 	/* For a high-order request, check at least one suitable page is free */
3440 	for (o = order; o < MAX_ORDER; o++) {
3441 		struct free_area *area = &z->free_area[o];
3442 		int mt;
3443 
3444 		if (!area->nr_free)
3445 			continue;
3446 
3447 		for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3448 			if (!free_area_empty(area, mt))
3449 				return true;
3450 		}
3451 
3452 #ifdef CONFIG_CMA
3453 		if ((alloc_flags & ALLOC_CMA) &&
3454 		    !free_area_empty(area, MIGRATE_CMA)) {
3455 			return true;
3456 		}
3457 #endif
3458 		if (alloc_harder &&
3459 			!list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
3460 			return true;
3461 	}
3462 	return false;
3463 }
3464 
3465 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3466 		      int classzone_idx, unsigned int alloc_flags)
3467 {
3468 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3469 					zone_page_state(z, NR_FREE_PAGES));
3470 }
3471 
3472 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3473 		unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3474 {
3475 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3476 	long cma_pages = 0;
3477 
3478 #ifdef CONFIG_CMA
3479 	/* If allocation can't use CMA areas don't use free CMA pages */
3480 	if (!(alloc_flags & ALLOC_CMA))
3481 		cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3482 #endif
3483 
3484 	/*
3485 	 * Fast check for order-0 only. If this fails then the reserves
3486 	 * need to be calculated. There is a corner case where the check
3487 	 * passes but only the high-order atomic reserve are free. If
3488 	 * the caller is !atomic then it'll uselessly search the free
3489 	 * list. That corner case is then slower but it is harmless.
3490 	 */
3491 	if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3492 		return true;
3493 
3494 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3495 					free_pages);
3496 }
3497 
3498 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3499 			unsigned long mark, int classzone_idx)
3500 {
3501 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3502 
3503 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3504 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3505 
3506 	return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
3507 								free_pages);
3508 }
3509 
3510 #ifdef CONFIG_NUMA
3511 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3512 {
3513 	return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3514 				RECLAIM_DISTANCE;
3515 }
3516 #else	/* CONFIG_NUMA */
3517 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3518 {
3519 	return true;
3520 }
3521 #endif	/* CONFIG_NUMA */
3522 
3523 /*
3524  * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid
3525  * fragmentation is subtle. If the preferred zone was HIGHMEM then
3526  * premature use of a lower zone may cause lowmem pressure problems that
3527  * are worse than fragmentation. If the next zone is ZONE_DMA then it is
3528  * probably too small. It only makes sense to spread allocations to avoid
3529  * fragmentation between the Normal and DMA32 zones.
3530  */
3531 static inline unsigned int
3532 alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3533 {
3534 	unsigned int alloc_flags = 0;
3535 
3536 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3537 		alloc_flags |= ALLOC_KSWAPD;
3538 
3539 #ifdef CONFIG_ZONE_DMA32
3540 	if (!zone)
3541 		return alloc_flags;
3542 
3543 	if (zone_idx(zone) != ZONE_NORMAL)
3544 		return alloc_flags;
3545 
3546 	/*
3547 	 * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and
3548 	 * the pointer is within zone->zone_pgdat->node_zones[]. Also assume
3549 	 * on UMA that if Normal is populated then so is DMA32.
3550 	 */
3551 	BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
3552 	if (nr_online_nodes > 1 && !populated_zone(--zone))
3553 		return alloc_flags;
3554 
3555 	alloc_flags |= ALLOC_NOFRAGMENT;
3556 #endif /* CONFIG_ZONE_DMA32 */
3557 	return alloc_flags;
3558 }
3559 
3560 /*
3561  * get_page_from_freelist goes through the zonelist trying to allocate
3562  * a page.
3563  */
3564 static struct page *
3565 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3566 						const struct alloc_context *ac)
3567 {
3568 	struct zoneref *z;
3569 	struct zone *zone;
3570 	struct pglist_data *last_pgdat_dirty_limit = NULL;
3571 	bool no_fallback;
3572 
3573 retry:
3574 	/*
3575 	 * Scan zonelist, looking for a zone with enough free.
3576 	 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3577 	 */
3578 	no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
3579 	z = ac->preferred_zoneref;
3580 	for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3581 								ac->nodemask) {
3582 		struct page *page;
3583 		unsigned long mark;
3584 
3585 		if (cpusets_enabled() &&
3586 			(alloc_flags & ALLOC_CPUSET) &&
3587 			!__cpuset_zone_allowed(zone, gfp_mask))
3588 				continue;
3589 		/*
3590 		 * When allocating a page cache page for writing, we
3591 		 * want to get it from a node that is within its dirty
3592 		 * limit, such that no single node holds more than its
3593 		 * proportional share of globally allowed dirty pages.
3594 		 * The dirty limits take into account the node's
3595 		 * lowmem reserves and high watermark so that kswapd
3596 		 * should be able to balance it without having to
3597 		 * write pages from its LRU list.
3598 		 *
3599 		 * XXX: For now, allow allocations to potentially
3600 		 * exceed the per-node dirty limit in the slowpath
3601 		 * (spread_dirty_pages unset) before going into reclaim,
3602 		 * which is important when on a NUMA setup the allowed
3603 		 * nodes are together not big enough to reach the
3604 		 * global limit.  The proper fix for these situations
3605 		 * will require awareness of nodes in the
3606 		 * dirty-throttling and the flusher threads.
3607 		 */
3608 		if (ac->spread_dirty_pages) {
3609 			if (last_pgdat_dirty_limit == zone->zone_pgdat)
3610 				continue;
3611 
3612 			if (!node_dirty_ok(zone->zone_pgdat)) {
3613 				last_pgdat_dirty_limit = zone->zone_pgdat;
3614 				continue;
3615 			}
3616 		}
3617 
3618 		if (no_fallback && nr_online_nodes > 1 &&
3619 		    zone != ac->preferred_zoneref->zone) {
3620 			int local_nid;
3621 
3622 			/*
3623 			 * If moving to a remote node, retry but allow
3624 			 * fragmenting fallbacks. Locality is more important
3625 			 * than fragmentation avoidance.
3626 			 */
3627 			local_nid = zone_to_nid(ac->preferred_zoneref->zone);
3628 			if (zone_to_nid(zone) != local_nid) {
3629 				alloc_flags &= ~ALLOC_NOFRAGMENT;
3630 				goto retry;
3631 			}
3632 		}
3633 
3634 		mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
3635 		if (!zone_watermark_fast(zone, order, mark,
3636 				       ac_classzone_idx(ac), alloc_flags)) {
3637 			int ret;
3638 
3639 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3640 			/*
3641 			 * Watermark failed for this zone, but see if we can
3642 			 * grow this zone if it contains deferred pages.
3643 			 */
3644 			if (static_branch_unlikely(&deferred_pages)) {
3645 				if (_deferred_grow_zone(zone, order))
3646 					goto try_this_zone;
3647 			}
3648 #endif
3649 			/* Checked here to keep the fast path fast */
3650 			BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3651 			if (alloc_flags & ALLOC_NO_WATERMARKS)
3652 				goto try_this_zone;
3653 
3654 			if (node_reclaim_mode == 0 ||
3655 			    !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3656 				continue;
3657 
3658 			ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3659 			switch (ret) {
3660 			case NODE_RECLAIM_NOSCAN:
3661 				/* did not scan */
3662 				continue;
3663 			case NODE_RECLAIM_FULL:
3664 				/* scanned but unreclaimable */
3665 				continue;
3666 			default:
3667 				/* did we reclaim enough */
3668 				if (zone_watermark_ok(zone, order, mark,
3669 						ac_classzone_idx(ac), alloc_flags))
3670 					goto try_this_zone;
3671 
3672 				continue;
3673 			}
3674 		}
3675 
3676 try_this_zone:
3677 		page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3678 				gfp_mask, alloc_flags, ac->migratetype);
3679 		if (page) {
3680 			prep_new_page(page, order, gfp_mask, alloc_flags);
3681 
3682 			/*
3683 			 * If this is a high-order atomic allocation then check
3684 			 * if the pageblock should be reserved for the future
3685 			 */
3686 			if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3687 				reserve_highatomic_pageblock(page, zone, order);
3688 
3689 			return page;
3690 		} else {
3691 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3692 			/* Try again if zone has deferred pages */
3693 			if (static_branch_unlikely(&deferred_pages)) {
3694 				if (_deferred_grow_zone(zone, order))
3695 					goto try_this_zone;
3696 			}
3697 #endif
3698 		}
3699 	}
3700 
3701 	/*
3702 	 * It's possible on a UMA machine to get through all zones that are
3703 	 * fragmented. If avoiding fragmentation, reset and try again.
3704 	 */
3705 	if (no_fallback) {
3706 		alloc_flags &= ~ALLOC_NOFRAGMENT;
3707 		goto retry;
3708 	}
3709 
3710 	return NULL;
3711 }
3712 
3713 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3714 {
3715 	unsigned int filter = SHOW_MEM_FILTER_NODES;
3716 	static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
3717 
3718 	if (!__ratelimit(&show_mem_rs))
3719 		return;
3720 
3721 	/*
3722 	 * This documents exceptions given to allocations in certain
3723 	 * contexts that are allowed to allocate outside current's set
3724 	 * of allowed nodes.
3725 	 */
3726 	if (!(gfp_mask & __GFP_NOMEMALLOC))
3727 		if (tsk_is_oom_victim(current) ||
3728 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
3729 			filter &= ~SHOW_MEM_FILTER_NODES;
3730 	if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3731 		filter &= ~SHOW_MEM_FILTER_NODES;
3732 
3733 	show_mem(filter, nodemask);
3734 }
3735 
3736 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3737 {
3738 	struct va_format vaf;
3739 	va_list args;
3740 	static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3741 				      DEFAULT_RATELIMIT_BURST);
3742 
3743 	if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3744 		return;
3745 
3746 	va_start(args, fmt);
3747 	vaf.fmt = fmt;
3748 	vaf.va = &args;
3749 	pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
3750 			current->comm, &vaf, gfp_mask, &gfp_mask,
3751 			nodemask_pr_args(nodemask));
3752 	va_end(args);
3753 
3754 	cpuset_print_current_mems_allowed();
3755 	pr_cont("\n");
3756 	dump_stack();
3757 	warn_alloc_show_mem(gfp_mask, nodemask);
3758 }
3759 
3760 static inline struct page *
3761 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3762 			      unsigned int alloc_flags,
3763 			      const struct alloc_context *ac)
3764 {
3765 	struct page *page;
3766 
3767 	page = get_page_from_freelist(gfp_mask, order,
3768 			alloc_flags|ALLOC_CPUSET, ac);
3769 	/*
3770 	 * fallback to ignore cpuset restriction if our nodes
3771 	 * are depleted
3772 	 */
3773 	if (!page)
3774 		page = get_page_from_freelist(gfp_mask, order,
3775 				alloc_flags, ac);
3776 
3777 	return page;
3778 }
3779 
3780 static inline struct page *
3781 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3782 	const struct alloc_context *ac, unsigned long *did_some_progress)
3783 {
3784 	struct oom_control oc = {
3785 		.zonelist = ac->zonelist,
3786 		.nodemask = ac->nodemask,
3787 		.memcg = NULL,
3788 		.gfp_mask = gfp_mask,
3789 		.order = order,
3790 	};
3791 	struct page *page;
3792 
3793 	*did_some_progress = 0;
3794 
3795 	/*
3796 	 * Acquire the oom lock.  If that fails, somebody else is
3797 	 * making progress for us.
3798 	 */
3799 	if (!mutex_trylock(&oom_lock)) {
3800 		*did_some_progress = 1;
3801 		schedule_timeout_uninterruptible(1);
3802 		return NULL;
3803 	}
3804 
3805 	/*
3806 	 * Go through the zonelist yet one more time, keep very high watermark
3807 	 * here, this is only to catch a parallel oom killing, we must fail if
3808 	 * we're still under heavy pressure. But make sure that this reclaim
3809 	 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3810 	 * allocation which will never fail due to oom_lock already held.
3811 	 */
3812 	page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3813 				      ~__GFP_DIRECT_RECLAIM, order,
3814 				      ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3815 	if (page)
3816 		goto out;
3817 
3818 	/* Coredumps can quickly deplete all memory reserves */
3819 	if (current->flags & PF_DUMPCORE)
3820 		goto out;
3821 	/* The OOM killer will not help higher order allocs */
3822 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3823 		goto out;
3824 	/*
3825 	 * We have already exhausted all our reclaim opportunities without any
3826 	 * success so it is time to admit defeat. We will skip the OOM killer
3827 	 * because it is very likely that the caller has a more reasonable
3828 	 * fallback than shooting a random task.
3829 	 */
3830 	if (gfp_mask & __GFP_RETRY_MAYFAIL)
3831 		goto out;
3832 	/* The OOM killer does not needlessly kill tasks for lowmem */
3833 	if (ac->high_zoneidx < ZONE_NORMAL)
3834 		goto out;
3835 	if (pm_suspended_storage())
3836 		goto out;
3837 	/*
3838 	 * XXX: GFP_NOFS allocations should rather fail than rely on
3839 	 * other request to make a forward progress.
3840 	 * We are in an unfortunate situation where out_of_memory cannot
3841 	 * do much for this context but let's try it to at least get
3842 	 * access to memory reserved if the current task is killed (see
3843 	 * out_of_memory). Once filesystems are ready to handle allocation
3844 	 * failures more gracefully we should just bail out here.
3845 	 */
3846 
3847 	/* The OOM killer may not free memory on a specific node */
3848 	if (gfp_mask & __GFP_THISNODE)
3849 		goto out;
3850 
3851 	/* Exhausted what can be done so it's blame time */
3852 	if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3853 		*did_some_progress = 1;
3854 
3855 		/*
3856 		 * Help non-failing allocations by giving them access to memory
3857 		 * reserves
3858 		 */
3859 		if (gfp_mask & __GFP_NOFAIL)
3860 			page = __alloc_pages_cpuset_fallback(gfp_mask, order,
3861 					ALLOC_NO_WATERMARKS, ac);
3862 	}
3863 out:
3864 	mutex_unlock(&oom_lock);
3865 	return page;
3866 }
3867 
3868 /*
3869  * Maximum number of compaction retries wit a progress before OOM
3870  * killer is consider as the only way to move forward.
3871  */
3872 #define MAX_COMPACT_RETRIES 16
3873 
3874 #ifdef CONFIG_COMPACTION
3875 /* Try memory compaction for high-order allocations before reclaim */
3876 static struct page *
3877 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3878 		unsigned int alloc_flags, const struct alloc_context *ac,
3879 		enum compact_priority prio, enum compact_result *compact_result)
3880 {
3881 	struct page *page = NULL;
3882 	unsigned long pflags;
3883 	unsigned int noreclaim_flag;
3884 
3885 	if (!order)
3886 		return NULL;
3887 
3888 	psi_memstall_enter(&pflags);
3889 	noreclaim_flag = memalloc_noreclaim_save();
3890 
3891 	*compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3892 								prio, &page);
3893 
3894 	memalloc_noreclaim_restore(noreclaim_flag);
3895 	psi_memstall_leave(&pflags);
3896 
3897 	/*
3898 	 * At least in one zone compaction wasn't deferred or skipped, so let's
3899 	 * count a compaction stall
3900 	 */
3901 	count_vm_event(COMPACTSTALL);
3902 
3903 	/* Prep a captured page if available */
3904 	if (page)
3905 		prep_new_page(page, order, gfp_mask, alloc_flags);
3906 
3907 	/* Try get a page from the freelist if available */
3908 	if (!page)
3909 		page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3910 
3911 	if (page) {
3912 		struct zone *zone = page_zone(page);
3913 
3914 		zone->compact_blockskip_flush = false;
3915 		compaction_defer_reset(zone, order, true);
3916 		count_vm_event(COMPACTSUCCESS);
3917 		return page;
3918 	}
3919 
3920 	/*
3921 	 * It's bad if compaction run occurs and fails. The most likely reason
3922 	 * is that pages exist, but not enough to satisfy watermarks.
3923 	 */
3924 	count_vm_event(COMPACTFAIL);
3925 
3926 	cond_resched();
3927 
3928 	return NULL;
3929 }
3930 
3931 static inline bool
3932 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3933 		     enum compact_result compact_result,
3934 		     enum compact_priority *compact_priority,
3935 		     int *compaction_retries)
3936 {
3937 	int max_retries = MAX_COMPACT_RETRIES;
3938 	int min_priority;
3939 	bool ret = false;
3940 	int retries = *compaction_retries;
3941 	enum compact_priority priority = *compact_priority;
3942 
3943 	if (!order)
3944 		return false;
3945 
3946 	if (compaction_made_progress(compact_result))
3947 		(*compaction_retries)++;
3948 
3949 	/*
3950 	 * compaction considers all the zone as desperately out of memory
3951 	 * so it doesn't really make much sense to retry except when the
3952 	 * failure could be caused by insufficient priority
3953 	 */
3954 	if (compaction_failed(compact_result))
3955 		goto check_priority;
3956 
3957 	/*
3958 	 * make sure the compaction wasn't deferred or didn't bail out early
3959 	 * due to locks contention before we declare that we should give up.
3960 	 * But do not retry if the given zonelist is not suitable for
3961 	 * compaction.
3962 	 */
3963 	if (compaction_withdrawn(compact_result)) {
3964 		ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3965 		goto out;
3966 	}
3967 
3968 	/*
3969 	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
3970 	 * costly ones because they are de facto nofail and invoke OOM
3971 	 * killer to move on while costly can fail and users are ready
3972 	 * to cope with that. 1/4 retries is rather arbitrary but we
3973 	 * would need much more detailed feedback from compaction to
3974 	 * make a better decision.
3975 	 */
3976 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3977 		max_retries /= 4;
3978 	if (*compaction_retries <= max_retries) {
3979 		ret = true;
3980 		goto out;
3981 	}
3982 
3983 	/*
3984 	 * Make sure there are attempts at the highest priority if we exhausted
3985 	 * all retries or failed at the lower priorities.
3986 	 */
3987 check_priority:
3988 	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3989 			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3990 
3991 	if (*compact_priority > min_priority) {
3992 		(*compact_priority)--;
3993 		*compaction_retries = 0;
3994 		ret = true;
3995 	}
3996 out:
3997 	trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3998 	return ret;
3999 }
4000 #else
4001 static inline struct page *
4002 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4003 		unsigned int alloc_flags, const struct alloc_context *ac,
4004 		enum compact_priority prio, enum compact_result *compact_result)
4005 {
4006 	*compact_result = COMPACT_SKIPPED;
4007 	return NULL;
4008 }
4009 
4010 static inline bool
4011 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4012 		     enum compact_result compact_result,
4013 		     enum compact_priority *compact_priority,
4014 		     int *compaction_retries)
4015 {
4016 	struct zone *zone;
4017 	struct zoneref *z;
4018 
4019 	if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4020 		return false;
4021 
4022 	/*
4023 	 * There are setups with compaction disabled which would prefer to loop
4024 	 * inside the allocator rather than hit the oom killer prematurely.
4025 	 * Let's give them a good hope and keep retrying while the order-0
4026 	 * watermarks are OK.
4027 	 */
4028 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
4029 					ac->nodemask) {
4030 		if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4031 					ac_classzone_idx(ac), alloc_flags))
4032 			return true;
4033 	}
4034 	return false;
4035 }
4036 #endif /* CONFIG_COMPACTION */
4037 
4038 #ifdef CONFIG_LOCKDEP
4039 static struct lockdep_map __fs_reclaim_map =
4040 	STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4041 
4042 static bool __need_fs_reclaim(gfp_t gfp_mask)
4043 {
4044 	gfp_mask = current_gfp_context(gfp_mask);
4045 
4046 	/* no reclaim without waiting on it */
4047 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4048 		return false;
4049 
4050 	/* this guy won't enter reclaim */
4051 	if (current->flags & PF_MEMALLOC)
4052 		return false;
4053 
4054 	/* We're only interested __GFP_FS allocations for now */
4055 	if (!(gfp_mask & __GFP_FS))
4056 		return false;
4057 
4058 	if (gfp_mask & __GFP_NOLOCKDEP)
4059 		return false;
4060 
4061 	return true;
4062 }
4063 
4064 void __fs_reclaim_acquire(void)
4065 {
4066 	lock_map_acquire(&__fs_reclaim_map);
4067 }
4068 
4069 void __fs_reclaim_release(void)
4070 {
4071 	lock_map_release(&__fs_reclaim_map);
4072 }
4073 
4074 void fs_reclaim_acquire(gfp_t gfp_mask)
4075 {
4076 	if (__need_fs_reclaim(gfp_mask))
4077 		__fs_reclaim_acquire();
4078 }
4079 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4080 
4081 void fs_reclaim_release(gfp_t gfp_mask)
4082 {
4083 	if (__need_fs_reclaim(gfp_mask))
4084 		__fs_reclaim_release();
4085 }
4086 EXPORT_SYMBOL_GPL(fs_reclaim_release);
4087 #endif
4088 
4089 /* Perform direct synchronous page reclaim */
4090 static int
4091 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
4092 					const struct alloc_context *ac)
4093 {
4094 	int progress;
4095 	unsigned int noreclaim_flag;
4096 	unsigned long pflags;
4097 
4098 	cond_resched();
4099 
4100 	/* We now go into synchronous reclaim */
4101 	cpuset_memory_pressure_bump();
4102 	psi_memstall_enter(&pflags);
4103 	fs_reclaim_acquire(gfp_mask);
4104 	noreclaim_flag = memalloc_noreclaim_save();
4105 
4106 	progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4107 								ac->nodemask);
4108 
4109 	memalloc_noreclaim_restore(noreclaim_flag);
4110 	fs_reclaim_release(gfp_mask);
4111 	psi_memstall_leave(&pflags);
4112 
4113 	cond_resched();
4114 
4115 	return progress;
4116 }
4117 
4118 /* The really slow allocator path where we enter direct reclaim */
4119 static inline struct page *
4120 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4121 		unsigned int alloc_flags, const struct alloc_context *ac,
4122 		unsigned long *did_some_progress)
4123 {
4124 	struct page *page = NULL;
4125 	bool drained = false;
4126 
4127 	*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4128 	if (unlikely(!(*did_some_progress)))
4129 		return NULL;
4130 
4131 retry:
4132 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4133 
4134 	/*
4135 	 * If an allocation failed after direct reclaim, it could be because
4136 	 * pages are pinned on the per-cpu lists or in high alloc reserves.
4137 	 * Shrink them them and try again
4138 	 */
4139 	if (!page && !drained) {
4140 		unreserve_highatomic_pageblock(ac, false);
4141 		drain_all_pages(NULL);
4142 		drained = true;
4143 		goto retry;
4144 	}
4145 
4146 	return page;
4147 }
4148 
4149 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4150 			     const struct alloc_context *ac)
4151 {
4152 	struct zoneref *z;
4153 	struct zone *zone;
4154 	pg_data_t *last_pgdat = NULL;
4155 	enum zone_type high_zoneidx = ac->high_zoneidx;
4156 
4157 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx,
4158 					ac->nodemask) {
4159 		if (last_pgdat != zone->zone_pgdat)
4160 			wakeup_kswapd(zone, gfp_mask, order, high_zoneidx);
4161 		last_pgdat = zone->zone_pgdat;
4162 	}
4163 }
4164 
4165 static inline unsigned int
4166 gfp_to_alloc_flags(gfp_t gfp_mask)
4167 {
4168 	unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4169 
4170 	/* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
4171 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4172 
4173 	/*
4174 	 * The caller may dip into page reserves a bit more if the caller
4175 	 * cannot run direct reclaim, or if the caller has realtime scheduling
4176 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
4177 	 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
4178 	 */
4179 	alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
4180 
4181 	if (gfp_mask & __GFP_ATOMIC) {
4182 		/*
4183 		 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
4184 		 * if it can't schedule.
4185 		 */
4186 		if (!(gfp_mask & __GFP_NOMEMALLOC))
4187 			alloc_flags |= ALLOC_HARDER;
4188 		/*
4189 		 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
4190 		 * comment for __cpuset_node_allowed().
4191 		 */
4192 		alloc_flags &= ~ALLOC_CPUSET;
4193 	} else if (unlikely(rt_task(current)) && !in_interrupt())
4194 		alloc_flags |= ALLOC_HARDER;
4195 
4196 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4197 		alloc_flags |= ALLOC_KSWAPD;
4198 
4199 #ifdef CONFIG_CMA
4200 	if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4201 		alloc_flags |= ALLOC_CMA;
4202 #endif
4203 	return alloc_flags;
4204 }
4205 
4206 static bool oom_reserves_allowed(struct task_struct *tsk)
4207 {
4208 	if (!tsk_is_oom_victim(tsk))
4209 		return false;
4210 
4211 	/*
4212 	 * !MMU doesn't have oom reaper so give access to memory reserves
4213 	 * only to the thread with TIF_MEMDIE set
4214 	 */
4215 	if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4216 		return false;
4217 
4218 	return true;
4219 }
4220 
4221 /*
4222  * Distinguish requests which really need access to full memory
4223  * reserves from oom victims which can live with a portion of it
4224  */
4225 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4226 {
4227 	if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4228 		return 0;
4229 	if (gfp_mask & __GFP_MEMALLOC)
4230 		return ALLOC_NO_WATERMARKS;
4231 	if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4232 		return ALLOC_NO_WATERMARKS;
4233 	if (!in_interrupt()) {
4234 		if (current->flags & PF_MEMALLOC)
4235 			return ALLOC_NO_WATERMARKS;
4236 		else if (oom_reserves_allowed(current))
4237 			return ALLOC_OOM;
4238 	}
4239 
4240 	return 0;
4241 }
4242 
4243 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4244 {
4245 	return !!__gfp_pfmemalloc_flags(gfp_mask);
4246 }
4247 
4248 /*
4249  * Checks whether it makes sense to retry the reclaim to make a forward progress
4250  * for the given allocation request.
4251  *
4252  * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
4253  * without success, or when we couldn't even meet the watermark if we
4254  * reclaimed all remaining pages on the LRU lists.
4255  *
4256  * Returns true if a retry is viable or false to enter the oom path.
4257  */
4258 static inline bool
4259 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4260 		     struct alloc_context *ac, int alloc_flags,
4261 		     bool did_some_progress, int *no_progress_loops)
4262 {
4263 	struct zone *zone;
4264 	struct zoneref *z;
4265 	bool ret = false;
4266 
4267 	/*
4268 	 * Costly allocations might have made a progress but this doesn't mean
4269 	 * their order will become available due to high fragmentation so
4270 	 * always increment the no progress counter for them
4271 	 */
4272 	if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4273 		*no_progress_loops = 0;
4274 	else
4275 		(*no_progress_loops)++;
4276 
4277 	/*
4278 	 * Make sure we converge to OOM if we cannot make any progress
4279 	 * several times in the row.
4280 	 */
4281 	if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4282 		/* Before OOM, exhaust highatomic_reserve */
4283 		return unreserve_highatomic_pageblock(ac, true);
4284 	}
4285 
4286 	/*
4287 	 * Keep reclaiming pages while there is a chance this will lead
4288 	 * somewhere.  If none of the target zones can satisfy our allocation
4289 	 * request even if all reclaimable pages are considered then we are
4290 	 * screwed and have to go OOM.
4291 	 */
4292 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
4293 					ac->nodemask) {
4294 		unsigned long available;
4295 		unsigned long reclaimable;
4296 		unsigned long min_wmark = min_wmark_pages(zone);
4297 		bool wmark;
4298 
4299 		available = reclaimable = zone_reclaimable_pages(zone);
4300 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4301 
4302 		/*
4303 		 * Would the allocation succeed if we reclaimed all
4304 		 * reclaimable pages?
4305 		 */
4306 		wmark = __zone_watermark_ok(zone, order, min_wmark,
4307 				ac_classzone_idx(ac), alloc_flags, available);
4308 		trace_reclaim_retry_zone(z, order, reclaimable,
4309 				available, min_wmark, *no_progress_loops, wmark);
4310 		if (wmark) {
4311 			/*
4312 			 * If we didn't make any progress and have a lot of
4313 			 * dirty + writeback pages then we should wait for
4314 			 * an IO to complete to slow down the reclaim and
4315 			 * prevent from pre mature OOM
4316 			 */
4317 			if (!did_some_progress) {
4318 				unsigned long write_pending;
4319 
4320 				write_pending = zone_page_state_snapshot(zone,
4321 							NR_ZONE_WRITE_PENDING);
4322 
4323 				if (2 * write_pending > reclaimable) {
4324 					congestion_wait(BLK_RW_ASYNC, HZ/10);
4325 					return true;
4326 				}
4327 			}
4328 
4329 			ret = true;
4330 			goto out;
4331 		}
4332 	}
4333 
4334 out:
4335 	/*
4336 	 * Memory allocation/reclaim might be called from a WQ context and the
4337 	 * current implementation of the WQ concurrency control doesn't
4338 	 * recognize that a particular WQ is congested if the worker thread is
4339 	 * looping without ever sleeping. Therefore we have to do a short sleep
4340 	 * here rather than calling cond_resched().
4341 	 */
4342 	if (current->flags & PF_WQ_WORKER)
4343 		schedule_timeout_uninterruptible(1);
4344 	else
4345 		cond_resched();
4346 	return ret;
4347 }
4348 
4349 static inline bool
4350 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4351 {
4352 	/*
4353 	 * It's possible that cpuset's mems_allowed and the nodemask from
4354 	 * mempolicy don't intersect. This should be normally dealt with by
4355 	 * policy_nodemask(), but it's possible to race with cpuset update in
4356 	 * such a way the check therein was true, and then it became false
4357 	 * before we got our cpuset_mems_cookie here.
4358 	 * This assumes that for all allocations, ac->nodemask can come only
4359 	 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4360 	 * when it does not intersect with the cpuset restrictions) or the
4361 	 * caller can deal with a violated nodemask.
4362 	 */
4363 	if (cpusets_enabled() && ac->nodemask &&
4364 			!cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4365 		ac->nodemask = NULL;
4366 		return true;
4367 	}
4368 
4369 	/*
4370 	 * When updating a task's mems_allowed or mempolicy nodemask, it is
4371 	 * possible to race with parallel threads in such a way that our
4372 	 * allocation can fail while the mask is being updated. If we are about
4373 	 * to fail, check if the cpuset changed during allocation and if so,
4374 	 * retry.
4375 	 */
4376 	if (read_mems_allowed_retry(cpuset_mems_cookie))
4377 		return true;
4378 
4379 	return false;
4380 }
4381 
4382 static inline struct page *
4383 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4384 						struct alloc_context *ac)
4385 {
4386 	bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4387 	const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4388 	struct page *page = NULL;
4389 	unsigned int alloc_flags;
4390 	unsigned long did_some_progress;
4391 	enum compact_priority compact_priority;
4392 	enum compact_result compact_result;
4393 	int compaction_retries;
4394 	int no_progress_loops;
4395 	unsigned int cpuset_mems_cookie;
4396 	int reserve_flags;
4397 
4398 	/*
4399 	 * We also sanity check to catch abuse of atomic reserves being used by
4400 	 * callers that are not in atomic context.
4401 	 */
4402 	if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4403 				(__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4404 		gfp_mask &= ~__GFP_ATOMIC;
4405 
4406 retry_cpuset:
4407 	compaction_retries = 0;
4408 	no_progress_loops = 0;
4409 	compact_priority = DEF_COMPACT_PRIORITY;
4410 	cpuset_mems_cookie = read_mems_allowed_begin();
4411 
4412 	/*
4413 	 * The fast path uses conservative alloc_flags to succeed only until
4414 	 * kswapd needs to be woken up, and to avoid the cost of setting up
4415 	 * alloc_flags precisely. So we do that now.
4416 	 */
4417 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
4418 
4419 	/*
4420 	 * We need to recalculate the starting point for the zonelist iterator
4421 	 * because we might have used different nodemask in the fast path, or
4422 	 * there was a cpuset modification and we are retrying - otherwise we
4423 	 * could end up iterating over non-eligible zones endlessly.
4424 	 */
4425 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4426 					ac->high_zoneidx, ac->nodemask);
4427 	if (!ac->preferred_zoneref->zone)
4428 		goto nopage;
4429 
4430 	if (alloc_flags & ALLOC_KSWAPD)
4431 		wake_all_kswapds(order, gfp_mask, ac);
4432 
4433 	/*
4434 	 * The adjusted alloc_flags might result in immediate success, so try
4435 	 * that first
4436 	 */
4437 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4438 	if (page)
4439 		goto got_pg;
4440 
4441 	/*
4442 	 * For costly allocations, try direct compaction first, as it's likely
4443 	 * that we have enough base pages and don't need to reclaim. For non-
4444 	 * movable high-order allocations, do that as well, as compaction will
4445 	 * try prevent permanent fragmentation by migrating from blocks of the
4446 	 * same migratetype.
4447 	 * Don't try this for allocations that are allowed to ignore
4448 	 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4449 	 */
4450 	if (can_direct_reclaim &&
4451 			(costly_order ||
4452 			   (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4453 			&& !gfp_pfmemalloc_allowed(gfp_mask)) {
4454 		page = __alloc_pages_direct_compact(gfp_mask, order,
4455 						alloc_flags, ac,
4456 						INIT_COMPACT_PRIORITY,
4457 						&compact_result);
4458 		if (page)
4459 			goto got_pg;
4460 
4461 		/*
4462 		 * Checks for costly allocations with __GFP_NORETRY, which
4463 		 * includes THP page fault allocations
4464 		 */
4465 		if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4466 			/*
4467 			 * If compaction is deferred for high-order allocations,
4468 			 * it is because sync compaction recently failed. If
4469 			 * this is the case and the caller requested a THP
4470 			 * allocation, we do not want to heavily disrupt the
4471 			 * system, so we fail the allocation instead of entering
4472 			 * direct reclaim.
4473 			 */
4474 			if (compact_result == COMPACT_DEFERRED)
4475 				goto nopage;
4476 
4477 			/*
4478 			 * Looks like reclaim/compaction is worth trying, but
4479 			 * sync compaction could be very expensive, so keep
4480 			 * using async compaction.
4481 			 */
4482 			compact_priority = INIT_COMPACT_PRIORITY;
4483 		}
4484 	}
4485 
4486 retry:
4487 	/* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4488 	if (alloc_flags & ALLOC_KSWAPD)
4489 		wake_all_kswapds(order, gfp_mask, ac);
4490 
4491 	reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4492 	if (reserve_flags)
4493 		alloc_flags = reserve_flags;
4494 
4495 	/*
4496 	 * Reset the nodemask and zonelist iterators if memory policies can be
4497 	 * ignored. These allocations are high priority and system rather than
4498 	 * user oriented.
4499 	 */
4500 	if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4501 		ac->nodemask = NULL;
4502 		ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4503 					ac->high_zoneidx, ac->nodemask);
4504 	}
4505 
4506 	/* Attempt with potentially adjusted zonelist and alloc_flags */
4507 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4508 	if (page)
4509 		goto got_pg;
4510 
4511 	/* Caller is not willing to reclaim, we can't balance anything */
4512 	if (!can_direct_reclaim)
4513 		goto nopage;
4514 
4515 	/* Avoid recursion of direct reclaim */
4516 	if (current->flags & PF_MEMALLOC)
4517 		goto nopage;
4518 
4519 	/* Try direct reclaim and then allocating */
4520 	page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4521 							&did_some_progress);
4522 	if (page)
4523 		goto got_pg;
4524 
4525 	/* Try direct compaction and then allocating */
4526 	page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4527 					compact_priority, &compact_result);
4528 	if (page)
4529 		goto got_pg;
4530 
4531 	/* Do not loop if specifically requested */
4532 	if (gfp_mask & __GFP_NORETRY)
4533 		goto nopage;
4534 
4535 	/*
4536 	 * Do not retry costly high order allocations unless they are
4537 	 * __GFP_RETRY_MAYFAIL
4538 	 */
4539 	if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4540 		goto nopage;
4541 
4542 	if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4543 				 did_some_progress > 0, &no_progress_loops))
4544 		goto retry;
4545 
4546 	/*
4547 	 * It doesn't make any sense to retry for the compaction if the order-0
4548 	 * reclaim is not able to make any progress because the current
4549 	 * implementation of the compaction depends on the sufficient amount
4550 	 * of free memory (see __compaction_suitable)
4551 	 */
4552 	if (did_some_progress > 0 &&
4553 			should_compact_retry(ac, order, alloc_flags,
4554 				compact_result, &compact_priority,
4555 				&compaction_retries))
4556 		goto retry;
4557 
4558 
4559 	/* Deal with possible cpuset update races before we start OOM killing */
4560 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4561 		goto retry_cpuset;
4562 
4563 	/* Reclaim has failed us, start killing things */
4564 	page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4565 	if (page)
4566 		goto got_pg;
4567 
4568 	/* Avoid allocations with no watermarks from looping endlessly */
4569 	if (tsk_is_oom_victim(current) &&
4570 	    (alloc_flags == ALLOC_OOM ||
4571 	     (gfp_mask & __GFP_NOMEMALLOC)))
4572 		goto nopage;
4573 
4574 	/* Retry as long as the OOM killer is making progress */
4575 	if (did_some_progress) {
4576 		no_progress_loops = 0;
4577 		goto retry;
4578 	}
4579 
4580 nopage:
4581 	/* Deal with possible cpuset update races before we fail */
4582 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4583 		goto retry_cpuset;
4584 
4585 	/*
4586 	 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4587 	 * we always retry
4588 	 */
4589 	if (gfp_mask & __GFP_NOFAIL) {
4590 		/*
4591 		 * All existing users of the __GFP_NOFAIL are blockable, so warn
4592 		 * of any new users that actually require GFP_NOWAIT
4593 		 */
4594 		if (WARN_ON_ONCE(!can_direct_reclaim))
4595 			goto fail;
4596 
4597 		/*
4598 		 * PF_MEMALLOC request from this context is rather bizarre
4599 		 * because we cannot reclaim anything and only can loop waiting
4600 		 * for somebody to do a work for us
4601 		 */
4602 		WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4603 
4604 		/*
4605 		 * non failing costly orders are a hard requirement which we
4606 		 * are not prepared for much so let's warn about these users
4607 		 * so that we can identify them and convert them to something
4608 		 * else.
4609 		 */
4610 		WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4611 
4612 		/*
4613 		 * Help non-failing allocations by giving them access to memory
4614 		 * reserves but do not use ALLOC_NO_WATERMARKS because this
4615 		 * could deplete whole memory reserves which would just make
4616 		 * the situation worse
4617 		 */
4618 		page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4619 		if (page)
4620 			goto got_pg;
4621 
4622 		cond_resched();
4623 		goto retry;
4624 	}
4625 fail:
4626 	warn_alloc(gfp_mask, ac->nodemask,
4627 			"page allocation failure: order:%u", order);
4628 got_pg:
4629 	return page;
4630 }
4631 
4632 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4633 		int preferred_nid, nodemask_t *nodemask,
4634 		struct alloc_context *ac, gfp_t *alloc_mask,
4635 		unsigned int *alloc_flags)
4636 {
4637 	ac->high_zoneidx = gfp_zone(gfp_mask);
4638 	ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4639 	ac->nodemask = nodemask;
4640 	ac->migratetype = gfpflags_to_migratetype(gfp_mask);
4641 
4642 	if (cpusets_enabled()) {
4643 		*alloc_mask |= __GFP_HARDWALL;
4644 		if (!ac->nodemask)
4645 			ac->nodemask = &cpuset_current_mems_allowed;
4646 		else
4647 			*alloc_flags |= ALLOC_CPUSET;
4648 	}
4649 
4650 	fs_reclaim_acquire(gfp_mask);
4651 	fs_reclaim_release(gfp_mask);
4652 
4653 	might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4654 
4655 	if (should_fail_alloc_page(gfp_mask, order))
4656 		return false;
4657 
4658 	if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4659 		*alloc_flags |= ALLOC_CMA;
4660 
4661 	return true;
4662 }
4663 
4664 /* Determine whether to spread dirty pages and what the first usable zone */
4665 static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac)
4666 {
4667 	/* Dirty zone balancing only done in the fast path */
4668 	ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4669 
4670 	/*
4671 	 * The preferred zone is used for statistics but crucially it is
4672 	 * also used as the starting point for the zonelist iterator. It
4673 	 * may get reset for allocations that ignore memory policies.
4674 	 */
4675 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4676 					ac->high_zoneidx, ac->nodemask);
4677 }
4678 
4679 /*
4680  * This is the 'heart' of the zoned buddy allocator.
4681  */
4682 struct page *
4683 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4684 							nodemask_t *nodemask)
4685 {
4686 	struct page *page;
4687 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
4688 	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4689 	struct alloc_context ac = { };
4690 
4691 	/*
4692 	 * There are several places where we assume that the order value is sane
4693 	 * so bail out early if the request is out of bound.
4694 	 */
4695 	if (unlikely(order >= MAX_ORDER)) {
4696 		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4697 		return NULL;
4698 	}
4699 
4700 	gfp_mask &= gfp_allowed_mask;
4701 	alloc_mask = gfp_mask;
4702 	if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4703 		return NULL;
4704 
4705 	finalise_ac(gfp_mask, &ac);
4706 
4707 	/*
4708 	 * Forbid the first pass from falling back to types that fragment
4709 	 * memory until all local zones are considered.
4710 	 */
4711 	alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask);
4712 
4713 	/* First allocation attempt */
4714 	page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4715 	if (likely(page))
4716 		goto out;
4717 
4718 	/*
4719 	 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4720 	 * resp. GFP_NOIO which has to be inherited for all allocation requests
4721 	 * from a particular context which has been marked by
4722 	 * memalloc_no{fs,io}_{save,restore}.
4723 	 */
4724 	alloc_mask = current_gfp_context(gfp_mask);
4725 	ac.spread_dirty_pages = false;
4726 
4727 	/*
4728 	 * Restore the original nodemask if it was potentially replaced with
4729 	 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4730 	 */
4731 	if (unlikely(ac.nodemask != nodemask))
4732 		ac.nodemask = nodemask;
4733 
4734 	page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4735 
4736 out:
4737 	if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4738 	    unlikely(__memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4739 		__free_pages(page, order);
4740 		page = NULL;
4741 	}
4742 
4743 	trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4744 
4745 	return page;
4746 }
4747 EXPORT_SYMBOL(__alloc_pages_nodemask);
4748 
4749 /*
4750  * Common helper functions. Never use with __GFP_HIGHMEM because the returned
4751  * address cannot represent highmem pages. Use alloc_pages and then kmap if
4752  * you need to access high mem.
4753  */
4754 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4755 {
4756 	struct page *page;
4757 
4758 	page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
4759 	if (!page)
4760 		return 0;
4761 	return (unsigned long) page_address(page);
4762 }
4763 EXPORT_SYMBOL(__get_free_pages);
4764 
4765 unsigned long get_zeroed_page(gfp_t gfp_mask)
4766 {
4767 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4768 }
4769 EXPORT_SYMBOL(get_zeroed_page);
4770 
4771 static inline void free_the_page(struct page *page, unsigned int order)
4772 {
4773 	if (order == 0)		/* Via pcp? */
4774 		free_unref_page(page);
4775 	else
4776 		__free_pages_ok(page, order);
4777 }
4778 
4779 void __free_pages(struct page *page, unsigned int order)
4780 {
4781 	if (put_page_testzero(page))
4782 		free_the_page(page, order);
4783 }
4784 EXPORT_SYMBOL(__free_pages);
4785 
4786 void free_pages(unsigned long addr, unsigned int order)
4787 {
4788 	if (addr != 0) {
4789 		VM_BUG_ON(!virt_addr_valid((void *)addr));
4790 		__free_pages(virt_to_page((void *)addr), order);
4791 	}
4792 }
4793 
4794 EXPORT_SYMBOL(free_pages);
4795 
4796 /*
4797  * Page Fragment:
4798  *  An arbitrary-length arbitrary-offset area of memory which resides
4799  *  within a 0 or higher order page.  Multiple fragments within that page
4800  *  are individually refcounted, in the page's reference counter.
4801  *
4802  * The page_frag functions below provide a simple allocation framework for
4803  * page fragments.  This is used by the network stack and network device
4804  * drivers to provide a backing region of memory for use as either an
4805  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4806  */
4807 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4808 					     gfp_t gfp_mask)
4809 {
4810 	struct page *page = NULL;
4811 	gfp_t gfp = gfp_mask;
4812 
4813 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4814 	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4815 		    __GFP_NOMEMALLOC;
4816 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4817 				PAGE_FRAG_CACHE_MAX_ORDER);
4818 	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4819 #endif
4820 	if (unlikely(!page))
4821 		page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4822 
4823 	nc->va = page ? page_address(page) : NULL;
4824 
4825 	return page;
4826 }
4827 
4828 void __page_frag_cache_drain(struct page *page, unsigned int count)
4829 {
4830 	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4831 
4832 	if (page_ref_sub_and_test(page, count))
4833 		free_the_page(page, compound_order(page));
4834 }
4835 EXPORT_SYMBOL(__page_frag_cache_drain);
4836 
4837 void *page_frag_alloc(struct page_frag_cache *nc,
4838 		      unsigned int fragsz, gfp_t gfp_mask)
4839 {
4840 	unsigned int size = PAGE_SIZE;
4841 	struct page *page;
4842 	int offset;
4843 
4844 	if (unlikely(!nc->va)) {
4845 refill:
4846 		page = __page_frag_cache_refill(nc, gfp_mask);
4847 		if (!page)
4848 			return NULL;
4849 
4850 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4851 		/* if size can vary use size else just use PAGE_SIZE */
4852 		size = nc->size;
4853 #endif
4854 		/* Even if we own the page, we do not use atomic_set().
4855 		 * This would break get_page_unless_zero() users.
4856 		 */
4857 		page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
4858 
4859 		/* reset page count bias and offset to start of new frag */
4860 		nc->pfmemalloc = page_is_pfmemalloc(page);
4861 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4862 		nc->offset = size;
4863 	}
4864 
4865 	offset = nc->offset - fragsz;
4866 	if (unlikely(offset < 0)) {
4867 		page = virt_to_page(nc->va);
4868 
4869 		if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
4870 			goto refill;
4871 
4872 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4873 		/* if size can vary use size else just use PAGE_SIZE */
4874 		size = nc->size;
4875 #endif
4876 		/* OK, page count is 0, we can safely set it */
4877 		set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
4878 
4879 		/* reset page count bias and offset to start of new frag */
4880 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4881 		offset = size - fragsz;
4882 	}
4883 
4884 	nc->pagecnt_bias--;
4885 	nc->offset = offset;
4886 
4887 	return nc->va + offset;
4888 }
4889 EXPORT_SYMBOL(page_frag_alloc);
4890 
4891 /*
4892  * Frees a page fragment allocated out of either a compound or order 0 page.
4893  */
4894 void page_frag_free(void *addr)
4895 {
4896 	struct page *page = virt_to_head_page(addr);
4897 
4898 	if (unlikely(put_page_testzero(page)))
4899 		free_the_page(page, compound_order(page));
4900 }
4901 EXPORT_SYMBOL(page_frag_free);
4902 
4903 static void *make_alloc_exact(unsigned long addr, unsigned int order,
4904 		size_t size)
4905 {
4906 	if (addr) {
4907 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
4908 		unsigned long used = addr + PAGE_ALIGN(size);
4909 
4910 		split_page(virt_to_page((void *)addr), order);
4911 		while (used < alloc_end) {
4912 			free_page(used);
4913 			used += PAGE_SIZE;
4914 		}
4915 	}
4916 	return (void *)addr;
4917 }
4918 
4919 /**
4920  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4921  * @size: the number of bytes to allocate
4922  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
4923  *
4924  * This function is similar to alloc_pages(), except that it allocates the
4925  * minimum number of pages to satisfy the request.  alloc_pages() can only
4926  * allocate memory in power-of-two pages.
4927  *
4928  * This function is also limited by MAX_ORDER.
4929  *
4930  * Memory allocated by this function must be released by free_pages_exact().
4931  *
4932  * Return: pointer to the allocated area or %NULL in case of error.
4933  */
4934 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4935 {
4936 	unsigned int order = get_order(size);
4937 	unsigned long addr;
4938 
4939 	if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
4940 		gfp_mask &= ~__GFP_COMP;
4941 
4942 	addr = __get_free_pages(gfp_mask, order);
4943 	return make_alloc_exact(addr, order, size);
4944 }
4945 EXPORT_SYMBOL(alloc_pages_exact);
4946 
4947 /**
4948  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4949  *			   pages on a node.
4950  * @nid: the preferred node ID where memory should be allocated
4951  * @size: the number of bytes to allocate
4952  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
4953  *
4954  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4955  * back.
4956  *
4957  * Return: pointer to the allocated area or %NULL in case of error.
4958  */
4959 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4960 {
4961 	unsigned int order = get_order(size);
4962 	struct page *p;
4963 
4964 	if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
4965 		gfp_mask &= ~__GFP_COMP;
4966 
4967 	p = alloc_pages_node(nid, gfp_mask, order);
4968 	if (!p)
4969 		return NULL;
4970 	return make_alloc_exact((unsigned long)page_address(p), order, size);
4971 }
4972 
4973 /**
4974  * free_pages_exact - release memory allocated via alloc_pages_exact()
4975  * @virt: the value returned by alloc_pages_exact.
4976  * @size: size of allocation, same value as passed to alloc_pages_exact().
4977  *
4978  * Release the memory allocated by a previous call to alloc_pages_exact.
4979  */
4980 void free_pages_exact(void *virt, size_t size)
4981 {
4982 	unsigned long addr = (unsigned long)virt;
4983 	unsigned long end = addr + PAGE_ALIGN(size);
4984 
4985 	while (addr < end) {
4986 		free_page(addr);
4987 		addr += PAGE_SIZE;
4988 	}
4989 }
4990 EXPORT_SYMBOL(free_pages_exact);
4991 
4992 /**
4993  * nr_free_zone_pages - count number of pages beyond high watermark
4994  * @offset: The zone index of the highest zone
4995  *
4996  * nr_free_zone_pages() counts the number of pages which are beyond the
4997  * high watermark within all zones at or below a given zone index.  For each
4998  * zone, the number of pages is calculated as:
4999  *
5000  *     nr_free_zone_pages = managed_pages - high_pages
5001  *
5002  * Return: number of pages beyond high watermark.
5003  */
5004 static unsigned long nr_free_zone_pages(int offset)
5005 {
5006 	struct zoneref *z;
5007 	struct zone *zone;
5008 
5009 	/* Just pick one node, since fallback list is circular */
5010 	unsigned long sum = 0;
5011 
5012 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5013 
5014 	for_each_zone_zonelist(zone, z, zonelist, offset) {
5015 		unsigned long size = zone_managed_pages(zone);
5016 		unsigned long high = high_wmark_pages(zone);
5017 		if (size > high)
5018 			sum += size - high;
5019 	}
5020 
5021 	return sum;
5022 }
5023 
5024 /**
5025  * nr_free_buffer_pages - count number of pages beyond high watermark
5026  *
5027  * nr_free_buffer_pages() counts the number of pages which are beyond the high
5028  * watermark within ZONE_DMA and ZONE_NORMAL.
5029  *
5030  * Return: number of pages beyond high watermark within ZONE_DMA and
5031  * ZONE_NORMAL.
5032  */
5033 unsigned long nr_free_buffer_pages(void)
5034 {
5035 	return nr_free_zone_pages(gfp_zone(GFP_USER));
5036 }
5037 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5038 
5039 /**
5040  * nr_free_pagecache_pages - count number of pages beyond high watermark
5041  *
5042  * nr_free_pagecache_pages() counts the number of pages which are beyond the
5043  * high watermark within all zones.
5044  *
5045  * Return: number of pages beyond high watermark within all zones.
5046  */
5047 unsigned long nr_free_pagecache_pages(void)
5048 {
5049 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
5050 }
5051 
5052 static inline void show_node(struct zone *zone)
5053 {
5054 	if (IS_ENABLED(CONFIG_NUMA))
5055 		printk("Node %d ", zone_to_nid(zone));
5056 }
5057 
5058 long si_mem_available(void)
5059 {
5060 	long available;
5061 	unsigned long pagecache;
5062 	unsigned long wmark_low = 0;
5063 	unsigned long pages[NR_LRU_LISTS];
5064 	unsigned long reclaimable;
5065 	struct zone *zone;
5066 	int lru;
5067 
5068 	for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5069 		pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5070 
5071 	for_each_zone(zone)
5072 		wmark_low += low_wmark_pages(zone);
5073 
5074 	/*
5075 	 * Estimate the amount of memory available for userspace allocations,
5076 	 * without causing swapping.
5077 	 */
5078 	available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5079 
5080 	/*
5081 	 * Not all the page cache can be freed, otherwise the system will
5082 	 * start swapping. Assume at least half of the page cache, or the
5083 	 * low watermark worth of cache, needs to stay.
5084 	 */
5085 	pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5086 	pagecache -= min(pagecache / 2, wmark_low);
5087 	available += pagecache;
5088 
5089 	/*
5090 	 * Part of the reclaimable slab and other kernel memory consists of
5091 	 * items that are in use, and cannot be freed. Cap this estimate at the
5092 	 * low watermark.
5093 	 */
5094 	reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) +
5095 			global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5096 	available += reclaimable - min(reclaimable / 2, wmark_low);
5097 
5098 	if (available < 0)
5099 		available = 0;
5100 	return available;
5101 }
5102 EXPORT_SYMBOL_GPL(si_mem_available);
5103 
5104 void si_meminfo(struct sysinfo *val)
5105 {
5106 	val->totalram = totalram_pages();
5107 	val->sharedram = global_node_page_state(NR_SHMEM);
5108 	val->freeram = global_zone_page_state(NR_FREE_PAGES);
5109 	val->bufferram = nr_blockdev_pages();
5110 	val->totalhigh = totalhigh_pages();
5111 	val->freehigh = nr_free_highpages();
5112 	val->mem_unit = PAGE_SIZE;
5113 }
5114 
5115 EXPORT_SYMBOL(si_meminfo);
5116 
5117 #ifdef CONFIG_NUMA
5118 void si_meminfo_node(struct sysinfo *val, int nid)
5119 {
5120 	int zone_type;		/* needs to be signed */
5121 	unsigned long managed_pages = 0;
5122 	unsigned long managed_highpages = 0;
5123 	unsigned long free_highpages = 0;
5124 	pg_data_t *pgdat = NODE_DATA(nid);
5125 
5126 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5127 		managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5128 	val->totalram = managed_pages;
5129 	val->sharedram = node_page_state(pgdat, NR_SHMEM);
5130 	val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5131 #ifdef CONFIG_HIGHMEM
5132 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5133 		struct zone *zone = &pgdat->node_zones[zone_type];
5134 
5135 		if (is_highmem(zone)) {
5136 			managed_highpages += zone_managed_pages(zone);
5137 			free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5138 		}
5139 	}
5140 	val->totalhigh = managed_highpages;
5141 	val->freehigh = free_highpages;
5142 #else
5143 	val->totalhigh = managed_highpages;
5144 	val->freehigh = free_highpages;
5145 #endif
5146 	val->mem_unit = PAGE_SIZE;
5147 }
5148 #endif
5149 
5150 /*
5151  * Determine whether the node should be displayed or not, depending on whether
5152  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
5153  */
5154 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5155 {
5156 	if (!(flags & SHOW_MEM_FILTER_NODES))
5157 		return false;
5158 
5159 	/*
5160 	 * no node mask - aka implicit memory numa policy. Do not bother with
5161 	 * the synchronization - read_mems_allowed_begin - because we do not
5162 	 * have to be precise here.
5163 	 */
5164 	if (!nodemask)
5165 		nodemask = &cpuset_current_mems_allowed;
5166 
5167 	return !node_isset(nid, *nodemask);
5168 }
5169 
5170 #define K(x) ((x) << (PAGE_SHIFT-10))
5171 
5172 static void show_migration_types(unsigned char type)
5173 {
5174 	static const char types[MIGRATE_TYPES] = {
5175 		[MIGRATE_UNMOVABLE]	= 'U',
5176 		[MIGRATE_MOVABLE]	= 'M',
5177 		[MIGRATE_RECLAIMABLE]	= 'E',
5178 		[MIGRATE_HIGHATOMIC]	= 'H',
5179 #ifdef CONFIG_CMA
5180 		[MIGRATE_CMA]		= 'C',
5181 #endif
5182 #ifdef CONFIG_MEMORY_ISOLATION
5183 		[MIGRATE_ISOLATE]	= 'I',
5184 #endif
5185 	};
5186 	char tmp[MIGRATE_TYPES + 1];
5187 	char *p = tmp;
5188 	int i;
5189 
5190 	for (i = 0; i < MIGRATE_TYPES; i++) {
5191 		if (type & (1 << i))
5192 			*p++ = types[i];
5193 	}
5194 
5195 	*p = '\0';
5196 	printk(KERN_CONT "(%s) ", tmp);
5197 }
5198 
5199 /*
5200  * Show free area list (used inside shift_scroll-lock stuff)
5201  * We also calculate the percentage fragmentation. We do this by counting the
5202  * memory on each free list with the exception of the first item on the list.
5203  *
5204  * Bits in @filter:
5205  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
5206  *   cpuset.
5207  */
5208 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5209 {
5210 	unsigned long free_pcp = 0;
5211 	int cpu;
5212 	struct zone *zone;
5213 	pg_data_t *pgdat;
5214 
5215 	for_each_populated_zone(zone) {
5216 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5217 			continue;
5218 
5219 		for_each_online_cpu(cpu)
5220 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5221 	}
5222 
5223 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5224 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5225 		" unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
5226 		" slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5227 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5228 		" free:%lu free_pcp:%lu free_cma:%lu\n",
5229 		global_node_page_state(NR_ACTIVE_ANON),
5230 		global_node_page_state(NR_INACTIVE_ANON),
5231 		global_node_page_state(NR_ISOLATED_ANON),
5232 		global_node_page_state(NR_ACTIVE_FILE),
5233 		global_node_page_state(NR_INACTIVE_FILE),
5234 		global_node_page_state(NR_ISOLATED_FILE),
5235 		global_node_page_state(NR_UNEVICTABLE),
5236 		global_node_page_state(NR_FILE_DIRTY),
5237 		global_node_page_state(NR_WRITEBACK),
5238 		global_node_page_state(NR_UNSTABLE_NFS),
5239 		global_node_page_state(NR_SLAB_RECLAIMABLE),
5240 		global_node_page_state(NR_SLAB_UNRECLAIMABLE),
5241 		global_node_page_state(NR_FILE_MAPPED),
5242 		global_node_page_state(NR_SHMEM),
5243 		global_zone_page_state(NR_PAGETABLE),
5244 		global_zone_page_state(NR_BOUNCE),
5245 		global_zone_page_state(NR_FREE_PAGES),
5246 		free_pcp,
5247 		global_zone_page_state(NR_FREE_CMA_PAGES));
5248 
5249 	for_each_online_pgdat(pgdat) {
5250 		if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5251 			continue;
5252 
5253 		printk("Node %d"
5254 			" active_anon:%lukB"
5255 			" inactive_anon:%lukB"
5256 			" active_file:%lukB"
5257 			" inactive_file:%lukB"
5258 			" unevictable:%lukB"
5259 			" isolated(anon):%lukB"
5260 			" isolated(file):%lukB"
5261 			" mapped:%lukB"
5262 			" dirty:%lukB"
5263 			" writeback:%lukB"
5264 			" shmem:%lukB"
5265 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5266 			" shmem_thp: %lukB"
5267 			" shmem_pmdmapped: %lukB"
5268 			" anon_thp: %lukB"
5269 #endif
5270 			" writeback_tmp:%lukB"
5271 			" unstable:%lukB"
5272 			" all_unreclaimable? %s"
5273 			"\n",
5274 			pgdat->node_id,
5275 			K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5276 			K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5277 			K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5278 			K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5279 			K(node_page_state(pgdat, NR_UNEVICTABLE)),
5280 			K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5281 			K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5282 			K(node_page_state(pgdat, NR_FILE_MAPPED)),
5283 			K(node_page_state(pgdat, NR_FILE_DIRTY)),
5284 			K(node_page_state(pgdat, NR_WRITEBACK)),
5285 			K(node_page_state(pgdat, NR_SHMEM)),
5286 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5287 			K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
5288 			K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
5289 					* HPAGE_PMD_NR),
5290 			K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
5291 #endif
5292 			K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5293 			K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
5294 			pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5295 				"yes" : "no");
5296 	}
5297 
5298 	for_each_populated_zone(zone) {
5299 		int i;
5300 
5301 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5302 			continue;
5303 
5304 		free_pcp = 0;
5305 		for_each_online_cpu(cpu)
5306 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5307 
5308 		show_node(zone);
5309 		printk(KERN_CONT
5310 			"%s"
5311 			" free:%lukB"
5312 			" min:%lukB"
5313 			" low:%lukB"
5314 			" high:%lukB"
5315 			" active_anon:%lukB"
5316 			" inactive_anon:%lukB"
5317 			" active_file:%lukB"
5318 			" inactive_file:%lukB"
5319 			" unevictable:%lukB"
5320 			" writepending:%lukB"
5321 			" present:%lukB"
5322 			" managed:%lukB"
5323 			" mlocked:%lukB"
5324 			" kernel_stack:%lukB"
5325 			" pagetables:%lukB"
5326 			" bounce:%lukB"
5327 			" free_pcp:%lukB"
5328 			" local_pcp:%ukB"
5329 			" free_cma:%lukB"
5330 			"\n",
5331 			zone->name,
5332 			K(zone_page_state(zone, NR_FREE_PAGES)),
5333 			K(min_wmark_pages(zone)),
5334 			K(low_wmark_pages(zone)),
5335 			K(high_wmark_pages(zone)),
5336 			K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
5337 			K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
5338 			K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
5339 			K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
5340 			K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5341 			K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
5342 			K(zone->present_pages),
5343 			K(zone_managed_pages(zone)),
5344 			K(zone_page_state(zone, NR_MLOCK)),
5345 			zone_page_state(zone, NR_KERNEL_STACK_KB),
5346 			K(zone_page_state(zone, NR_PAGETABLE)),
5347 			K(zone_page_state(zone, NR_BOUNCE)),
5348 			K(free_pcp),
5349 			K(this_cpu_read(zone->pageset->pcp.count)),
5350 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
5351 		printk("lowmem_reserve[]:");
5352 		for (i = 0; i < MAX_NR_ZONES; i++)
5353 			printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
5354 		printk(KERN_CONT "\n");
5355 	}
5356 
5357 	for_each_populated_zone(zone) {
5358 		unsigned int order;
5359 		unsigned long nr[MAX_ORDER], flags, total = 0;
5360 		unsigned char types[MAX_ORDER];
5361 
5362 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5363 			continue;
5364 		show_node(zone);
5365 		printk(KERN_CONT "%s: ", zone->name);
5366 
5367 		spin_lock_irqsave(&zone->lock, flags);
5368 		for (order = 0; order < MAX_ORDER; order++) {
5369 			struct free_area *area = &zone->free_area[order];
5370 			int type;
5371 
5372 			nr[order] = area->nr_free;
5373 			total += nr[order] << order;
5374 
5375 			types[order] = 0;
5376 			for (type = 0; type < MIGRATE_TYPES; type++) {
5377 				if (!free_area_empty(area, type))
5378 					types[order] |= 1 << type;
5379 			}
5380 		}
5381 		spin_unlock_irqrestore(&zone->lock, flags);
5382 		for (order = 0; order < MAX_ORDER; order++) {
5383 			printk(KERN_CONT "%lu*%lukB ",
5384 			       nr[order], K(1UL) << order);
5385 			if (nr[order])
5386 				show_migration_types(types[order]);
5387 		}
5388 		printk(KERN_CONT "= %lukB\n", K(total));
5389 	}
5390 
5391 	hugetlb_show_meminfo();
5392 
5393 	printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5394 
5395 	show_swap_cache_info();
5396 }
5397 
5398 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5399 {
5400 	zoneref->zone = zone;
5401 	zoneref->zone_idx = zone_idx(zone);
5402 }
5403 
5404 /*
5405  * Builds allocation fallback zone lists.
5406  *
5407  * Add all populated zones of a node to the zonelist.
5408  */
5409 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5410 {
5411 	struct zone *zone;
5412 	enum zone_type zone_type = MAX_NR_ZONES;
5413 	int nr_zones = 0;
5414 
5415 	do {
5416 		zone_type--;
5417 		zone = pgdat->node_zones + zone_type;
5418 		if (managed_zone(zone)) {
5419 			zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5420 			check_highest_zone(zone_type);
5421 		}
5422 	} while (zone_type);
5423 
5424 	return nr_zones;
5425 }
5426 
5427 #ifdef CONFIG_NUMA
5428 
5429 static int __parse_numa_zonelist_order(char *s)
5430 {
5431 	/*
5432 	 * We used to support different zonlists modes but they turned
5433 	 * out to be just not useful. Let's keep the warning in place
5434 	 * if somebody still use the cmd line parameter so that we do
5435 	 * not fail it silently
5436 	 */
5437 	if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5438 		pr_warn("Ignoring unsupported numa_zonelist_order value:  %s\n", s);
5439 		return -EINVAL;
5440 	}
5441 	return 0;
5442 }
5443 
5444 static __init int setup_numa_zonelist_order(char *s)
5445 {
5446 	if (!s)
5447 		return 0;
5448 
5449 	return __parse_numa_zonelist_order(s);
5450 }
5451 early_param("numa_zonelist_order", setup_numa_zonelist_order);
5452 
5453 char numa_zonelist_order[] = "Node";
5454 
5455 /*
5456  * sysctl handler for numa_zonelist_order
5457  */
5458 int numa_zonelist_order_handler(struct ctl_table *table, int write,
5459 		void __user *buffer, size_t *length,
5460 		loff_t *ppos)
5461 {
5462 	char *str;
5463 	int ret;
5464 
5465 	if (!write)
5466 		return proc_dostring(table, write, buffer, length, ppos);
5467 	str = memdup_user_nul(buffer, 16);
5468 	if (IS_ERR(str))
5469 		return PTR_ERR(str);
5470 
5471 	ret = __parse_numa_zonelist_order(str);
5472 	kfree(str);
5473 	return ret;
5474 }
5475 
5476 
5477 #define MAX_NODE_LOAD (nr_online_nodes)
5478 static int node_load[MAX_NUMNODES];
5479 
5480 /**
5481  * find_next_best_node - find the next node that should appear in a given node's fallback list
5482  * @node: node whose fallback list we're appending
5483  * @used_node_mask: nodemask_t of already used nodes
5484  *
5485  * We use a number of factors to determine which is the next node that should
5486  * appear on a given node's fallback list.  The node should not have appeared
5487  * already in @node's fallback list, and it should be the next closest node
5488  * according to the distance array (which contains arbitrary distance values
5489  * from each node to each node in the system), and should also prefer nodes
5490  * with no CPUs, since presumably they'll have very little allocation pressure
5491  * on them otherwise.
5492  *
5493  * Return: node id of the found node or %NUMA_NO_NODE if no node is found.
5494  */
5495 static int find_next_best_node(int node, nodemask_t *used_node_mask)
5496 {
5497 	int n, val;
5498 	int min_val = INT_MAX;
5499 	int best_node = NUMA_NO_NODE;
5500 	const struct cpumask *tmp = cpumask_of_node(0);
5501 
5502 	/* Use the local node if we haven't already */
5503 	if (!node_isset(node, *used_node_mask)) {
5504 		node_set(node, *used_node_mask);
5505 		return node;
5506 	}
5507 
5508 	for_each_node_state(n, N_MEMORY) {
5509 
5510 		/* Don't want a node to appear more than once */
5511 		if (node_isset(n, *used_node_mask))
5512 			continue;
5513 
5514 		/* Use the distance array to find the distance */
5515 		val = node_distance(node, n);
5516 
5517 		/* Penalize nodes under us ("prefer the next node") */
5518 		val += (n < node);
5519 
5520 		/* Give preference to headless and unused nodes */
5521 		tmp = cpumask_of_node(n);
5522 		if (!cpumask_empty(tmp))
5523 			val += PENALTY_FOR_NODE_WITH_CPUS;
5524 
5525 		/* Slight preference for less loaded node */
5526 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5527 		val += node_load[n];
5528 
5529 		if (val < min_val) {
5530 			min_val = val;
5531 			best_node = n;
5532 		}
5533 	}
5534 
5535 	if (best_node >= 0)
5536 		node_set(best_node, *used_node_mask);
5537 
5538 	return best_node;
5539 }
5540 
5541 
5542 /*
5543  * Build zonelists ordered by node and zones within node.
5544  * This results in maximum locality--normal zone overflows into local
5545  * DMA zone, if any--but risks exhausting DMA zone.
5546  */
5547 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5548 		unsigned nr_nodes)
5549 {
5550 	struct zoneref *zonerefs;
5551 	int i;
5552 
5553 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5554 
5555 	for (i = 0; i < nr_nodes; i++) {
5556 		int nr_zones;
5557 
5558 		pg_data_t *node = NODE_DATA(node_order[i]);
5559 
5560 		nr_zones = build_zonerefs_node(node, zonerefs);
5561 		zonerefs += nr_zones;
5562 	}
5563 	zonerefs->zone = NULL;
5564 	zonerefs->zone_idx = 0;
5565 }
5566 
5567 /*
5568  * Build gfp_thisnode zonelists
5569  */
5570 static void build_thisnode_zonelists(pg_data_t *pgdat)
5571 {
5572 	struct zoneref *zonerefs;
5573 	int nr_zones;
5574 
5575 	zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5576 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5577 	zonerefs += nr_zones;
5578 	zonerefs->zone = NULL;
5579 	zonerefs->zone_idx = 0;
5580 }
5581 
5582 /*
5583  * Build zonelists ordered by zone and nodes within zones.
5584  * This results in conserving DMA zone[s] until all Normal memory is
5585  * exhausted, but results in overflowing to remote node while memory
5586  * may still exist in local DMA zone.
5587  */
5588 
5589 static void build_zonelists(pg_data_t *pgdat)
5590 {
5591 	static int node_order[MAX_NUMNODES];
5592 	int node, load, nr_nodes = 0;
5593 	nodemask_t used_mask;
5594 	int local_node, prev_node;
5595 
5596 	/* NUMA-aware ordering of nodes */
5597 	local_node = pgdat->node_id;
5598 	load = nr_online_nodes;
5599 	prev_node = local_node;
5600 	nodes_clear(used_mask);
5601 
5602 	memset(node_order, 0, sizeof(node_order));
5603 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5604 		/*
5605 		 * We don't want to pressure a particular node.
5606 		 * So adding penalty to the first node in same
5607 		 * distance group to make it round-robin.
5608 		 */
5609 		if (node_distance(local_node, node) !=
5610 		    node_distance(local_node, prev_node))
5611 			node_load[node] = load;
5612 
5613 		node_order[nr_nodes++] = node;
5614 		prev_node = node;
5615 		load--;
5616 	}
5617 
5618 	build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5619 	build_thisnode_zonelists(pgdat);
5620 }
5621 
5622 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5623 /*
5624  * Return node id of node used for "local" allocations.
5625  * I.e., first node id of first zone in arg node's generic zonelist.
5626  * Used for initializing percpu 'numa_mem', which is used primarily
5627  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5628  */
5629 int local_memory_node(int node)
5630 {
5631 	struct zoneref *z;
5632 
5633 	z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5634 				   gfp_zone(GFP_KERNEL),
5635 				   NULL);
5636 	return zone_to_nid(z->zone);
5637 }
5638 #endif
5639 
5640 static void setup_min_unmapped_ratio(void);
5641 static void setup_min_slab_ratio(void);
5642 #else	/* CONFIG_NUMA */
5643 
5644 static void build_zonelists(pg_data_t *pgdat)
5645 {
5646 	int node, local_node;
5647 	struct zoneref *zonerefs;
5648 	int nr_zones;
5649 
5650 	local_node = pgdat->node_id;
5651 
5652 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5653 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5654 	zonerefs += nr_zones;
5655 
5656 	/*
5657 	 * Now we build the zonelist so that it contains the zones
5658 	 * of all the other nodes.
5659 	 * We don't want to pressure a particular node, so when
5660 	 * building the zones for node N, we make sure that the
5661 	 * zones coming right after the local ones are those from
5662 	 * node N+1 (modulo N)
5663 	 */
5664 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5665 		if (!node_online(node))
5666 			continue;
5667 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5668 		zonerefs += nr_zones;
5669 	}
5670 	for (node = 0; node < local_node; node++) {
5671 		if (!node_online(node))
5672 			continue;
5673 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5674 		zonerefs += nr_zones;
5675 	}
5676 
5677 	zonerefs->zone = NULL;
5678 	zonerefs->zone_idx = 0;
5679 }
5680 
5681 #endif	/* CONFIG_NUMA */
5682 
5683 /*
5684  * Boot pageset table. One per cpu which is going to be used for all
5685  * zones and all nodes. The parameters will be set in such a way
5686  * that an item put on a list will immediately be handed over to
5687  * the buddy list. This is safe since pageset manipulation is done
5688  * with interrupts disabled.
5689  *
5690  * The boot_pagesets must be kept even after bootup is complete for
5691  * unused processors and/or zones. They do play a role for bootstrapping
5692  * hotplugged processors.
5693  *
5694  * zoneinfo_show() and maybe other functions do
5695  * not check if the processor is online before following the pageset pointer.
5696  * Other parts of the kernel may not check if the zone is available.
5697  */
5698 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5699 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5700 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5701 
5702 static void __build_all_zonelists(void *data)
5703 {
5704 	int nid;
5705 	int __maybe_unused cpu;
5706 	pg_data_t *self = data;
5707 	static DEFINE_SPINLOCK(lock);
5708 
5709 	spin_lock(&lock);
5710 
5711 #ifdef CONFIG_NUMA
5712 	memset(node_load, 0, sizeof(node_load));
5713 #endif
5714 
5715 	/*
5716 	 * This node is hotadded and no memory is yet present.   So just
5717 	 * building zonelists is fine - no need to touch other nodes.
5718 	 */
5719 	if (self && !node_online(self->node_id)) {
5720 		build_zonelists(self);
5721 	} else {
5722 		for_each_online_node(nid) {
5723 			pg_data_t *pgdat = NODE_DATA(nid);
5724 
5725 			build_zonelists(pgdat);
5726 		}
5727 
5728 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5729 		/*
5730 		 * We now know the "local memory node" for each node--
5731 		 * i.e., the node of the first zone in the generic zonelist.
5732 		 * Set up numa_mem percpu variable for on-line cpus.  During
5733 		 * boot, only the boot cpu should be on-line;  we'll init the
5734 		 * secondary cpus' numa_mem as they come on-line.  During
5735 		 * node/memory hotplug, we'll fixup all on-line cpus.
5736 		 */
5737 		for_each_online_cpu(cpu)
5738 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5739 #endif
5740 	}
5741 
5742 	spin_unlock(&lock);
5743 }
5744 
5745 static noinline void __init
5746 build_all_zonelists_init(void)
5747 {
5748 	int cpu;
5749 
5750 	__build_all_zonelists(NULL);
5751 
5752 	/*
5753 	 * Initialize the boot_pagesets that are going to be used
5754 	 * for bootstrapping processors. The real pagesets for
5755 	 * each zone will be allocated later when the per cpu
5756 	 * allocator is available.
5757 	 *
5758 	 * boot_pagesets are used also for bootstrapping offline
5759 	 * cpus if the system is already booted because the pagesets
5760 	 * are needed to initialize allocators on a specific cpu too.
5761 	 * F.e. the percpu allocator needs the page allocator which
5762 	 * needs the percpu allocator in order to allocate its pagesets
5763 	 * (a chicken-egg dilemma).
5764 	 */
5765 	for_each_possible_cpu(cpu)
5766 		setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5767 
5768 	mminit_verify_zonelist();
5769 	cpuset_init_current_mems_allowed();
5770 }
5771 
5772 /*
5773  * unless system_state == SYSTEM_BOOTING.
5774  *
5775  * __ref due to call of __init annotated helper build_all_zonelists_init
5776  * [protected by SYSTEM_BOOTING].
5777  */
5778 void __ref build_all_zonelists(pg_data_t *pgdat)
5779 {
5780 	if (system_state == SYSTEM_BOOTING) {
5781 		build_all_zonelists_init();
5782 	} else {
5783 		__build_all_zonelists(pgdat);
5784 		/* cpuset refresh routine should be here */
5785 	}
5786 	vm_total_pages = nr_free_pagecache_pages();
5787 	/*
5788 	 * Disable grouping by mobility if the number of pages in the
5789 	 * system is too low to allow the mechanism to work. It would be
5790 	 * more accurate, but expensive to check per-zone. This check is
5791 	 * made on memory-hotadd so a system can start with mobility
5792 	 * disabled and enable it later
5793 	 */
5794 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5795 		page_group_by_mobility_disabled = 1;
5796 	else
5797 		page_group_by_mobility_disabled = 0;
5798 
5799 	pr_info("Built %u zonelists, mobility grouping %s.  Total pages: %ld\n",
5800 		nr_online_nodes,
5801 		page_group_by_mobility_disabled ? "off" : "on",
5802 		vm_total_pages);
5803 #ifdef CONFIG_NUMA
5804 	pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5805 #endif
5806 }
5807 
5808 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
5809 static bool __meminit
5810 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
5811 {
5812 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5813 	static struct memblock_region *r;
5814 
5815 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5816 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
5817 			for_each_memblock(memory, r) {
5818 				if (*pfn < memblock_region_memory_end_pfn(r))
5819 					break;
5820 			}
5821 		}
5822 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
5823 		    memblock_is_mirror(r)) {
5824 			*pfn = memblock_region_memory_end_pfn(r);
5825 			return true;
5826 		}
5827 	}
5828 #endif
5829 	return false;
5830 }
5831 
5832 /*
5833  * Initially all pages are reserved - free ones are freed
5834  * up by memblock_free_all() once the early boot process is
5835  * done. Non-atomic initialization, single-pass.
5836  */
5837 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5838 		unsigned long start_pfn, enum memmap_context context,
5839 		struct vmem_altmap *altmap)
5840 {
5841 	unsigned long pfn, end_pfn = start_pfn + size;
5842 	struct page *page;
5843 
5844 	if (highest_memmap_pfn < end_pfn - 1)
5845 		highest_memmap_pfn = end_pfn - 1;
5846 
5847 #ifdef CONFIG_ZONE_DEVICE
5848 	/*
5849 	 * Honor reservation requested by the driver for this ZONE_DEVICE
5850 	 * memory. We limit the total number of pages to initialize to just
5851 	 * those that might contain the memory mapping. We will defer the
5852 	 * ZONE_DEVICE page initialization until after we have released
5853 	 * the hotplug lock.
5854 	 */
5855 	if (zone == ZONE_DEVICE) {
5856 		if (!altmap)
5857 			return;
5858 
5859 		if (start_pfn == altmap->base_pfn)
5860 			start_pfn += altmap->reserve;
5861 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5862 	}
5863 #endif
5864 
5865 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5866 		/*
5867 		 * There can be holes in boot-time mem_map[]s handed to this
5868 		 * function.  They do not exist on hotplugged memory.
5869 		 */
5870 		if (context == MEMMAP_EARLY) {
5871 			if (!early_pfn_valid(pfn))
5872 				continue;
5873 			if (!early_pfn_in_nid(pfn, nid))
5874 				continue;
5875 			if (overlap_memmap_init(zone, &pfn))
5876 				continue;
5877 			if (defer_init(nid, pfn, end_pfn))
5878 				break;
5879 		}
5880 
5881 		page = pfn_to_page(pfn);
5882 		__init_single_page(page, pfn, zone, nid);
5883 		if (context == MEMMAP_HOTPLUG)
5884 			__SetPageReserved(page);
5885 
5886 		/*
5887 		 * Mark the block movable so that blocks are reserved for
5888 		 * movable at startup. This will force kernel allocations
5889 		 * to reserve their blocks rather than leaking throughout
5890 		 * the address space during boot when many long-lived
5891 		 * kernel allocations are made.
5892 		 *
5893 		 * bitmap is created for zone's valid pfn range. but memmap
5894 		 * can be created for invalid pages (for alignment)
5895 		 * check here not to call set_pageblock_migratetype() against
5896 		 * pfn out of zone.
5897 		 */
5898 		if (!(pfn & (pageblock_nr_pages - 1))) {
5899 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5900 			cond_resched();
5901 		}
5902 	}
5903 }
5904 
5905 #ifdef CONFIG_ZONE_DEVICE
5906 void __ref memmap_init_zone_device(struct zone *zone,
5907 				   unsigned long start_pfn,
5908 				   unsigned long size,
5909 				   struct dev_pagemap *pgmap)
5910 {
5911 	unsigned long pfn, end_pfn = start_pfn + size;
5912 	struct pglist_data *pgdat = zone->zone_pgdat;
5913 	struct vmem_altmap *altmap = pgmap_altmap(pgmap);
5914 	unsigned long zone_idx = zone_idx(zone);
5915 	unsigned long start = jiffies;
5916 	int nid = pgdat->node_id;
5917 
5918 	if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
5919 		return;
5920 
5921 	/*
5922 	 * The call to memmap_init_zone should have already taken care
5923 	 * of the pages reserved for the memmap, so we can just jump to
5924 	 * the end of that region and start processing the device pages.
5925 	 */
5926 	if (altmap) {
5927 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5928 		size = end_pfn - start_pfn;
5929 	}
5930 
5931 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5932 		struct page *page = pfn_to_page(pfn);
5933 
5934 		__init_single_page(page, pfn, zone_idx, nid);
5935 
5936 		/*
5937 		 * Mark page reserved as it will need to wait for onlining
5938 		 * phase for it to be fully associated with a zone.
5939 		 *
5940 		 * We can use the non-atomic __set_bit operation for setting
5941 		 * the flag as we are still initializing the pages.
5942 		 */
5943 		__SetPageReserved(page);
5944 
5945 		/*
5946 		 * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
5947 		 * and zone_device_data.  It is a bug if a ZONE_DEVICE page is
5948 		 * ever freed or placed on a driver-private list.
5949 		 */
5950 		page->pgmap = pgmap;
5951 		page->zone_device_data = NULL;
5952 
5953 		/*
5954 		 * Mark the block movable so that blocks are reserved for
5955 		 * movable at startup. This will force kernel allocations
5956 		 * to reserve their blocks rather than leaking throughout
5957 		 * the address space during boot when many long-lived
5958 		 * kernel allocations are made.
5959 		 *
5960 		 * bitmap is created for zone's valid pfn range. but memmap
5961 		 * can be created for invalid pages (for alignment)
5962 		 * check here not to call set_pageblock_migratetype() against
5963 		 * pfn out of zone.
5964 		 *
5965 		 * Please note that MEMMAP_HOTPLUG path doesn't clear memmap
5966 		 * because this is done early in section_activate()
5967 		 */
5968 		if (!(pfn & (pageblock_nr_pages - 1))) {
5969 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5970 			cond_resched();
5971 		}
5972 	}
5973 
5974 	pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev),
5975 		size, jiffies_to_msecs(jiffies - start));
5976 }
5977 
5978 #endif
5979 static void __meminit zone_init_free_lists(struct zone *zone)
5980 {
5981 	unsigned int order, t;
5982 	for_each_migratetype_order(order, t) {
5983 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5984 		zone->free_area[order].nr_free = 0;
5985 	}
5986 }
5987 
5988 void __meminit __weak memmap_init(unsigned long size, int nid,
5989 				  unsigned long zone, unsigned long start_pfn)
5990 {
5991 	memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL);
5992 }
5993 
5994 static int zone_batchsize(struct zone *zone)
5995 {
5996 #ifdef CONFIG_MMU
5997 	int batch;
5998 
5999 	/*
6000 	 * The per-cpu-pages pools are set to around 1000th of the
6001 	 * size of the zone.
6002 	 */
6003 	batch = zone_managed_pages(zone) / 1024;
6004 	/* But no more than a meg. */
6005 	if (batch * PAGE_SIZE > 1024 * 1024)
6006 		batch = (1024 * 1024) / PAGE_SIZE;
6007 	batch /= 4;		/* We effectively *= 4 below */
6008 	if (batch < 1)
6009 		batch = 1;
6010 
6011 	/*
6012 	 * Clamp the batch to a 2^n - 1 value. Having a power
6013 	 * of 2 value was found to be more likely to have
6014 	 * suboptimal cache aliasing properties in some cases.
6015 	 *
6016 	 * For example if 2 tasks are alternately allocating
6017 	 * batches of pages, one task can end up with a lot
6018 	 * of pages of one half of the possible page colors
6019 	 * and the other with pages of the other colors.
6020 	 */
6021 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
6022 
6023 	return batch;
6024 
6025 #else
6026 	/* The deferral and batching of frees should be suppressed under NOMMU
6027 	 * conditions.
6028 	 *
6029 	 * The problem is that NOMMU needs to be able to allocate large chunks
6030 	 * of contiguous memory as there's no hardware page translation to
6031 	 * assemble apparent contiguous memory from discontiguous pages.
6032 	 *
6033 	 * Queueing large contiguous runs of pages for batching, however,
6034 	 * causes the pages to actually be freed in smaller chunks.  As there
6035 	 * can be a significant delay between the individual batches being
6036 	 * recycled, this leads to the once large chunks of space being
6037 	 * fragmented and becoming unavailable for high-order allocations.
6038 	 */
6039 	return 0;
6040 #endif
6041 }
6042 
6043 /*
6044  * pcp->high and pcp->batch values are related and dependent on one another:
6045  * ->batch must never be higher then ->high.
6046  * The following function updates them in a safe manner without read side
6047  * locking.
6048  *
6049  * Any new users of pcp->batch and pcp->high should ensure they can cope with
6050  * those fields changing asynchronously (acording the the above rule).
6051  *
6052  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
6053  * outside of boot time (or some other assurance that no concurrent updaters
6054  * exist).
6055  */
6056 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6057 		unsigned long batch)
6058 {
6059        /* start with a fail safe value for batch */
6060 	pcp->batch = 1;
6061 	smp_wmb();
6062 
6063        /* Update high, then batch, in order */
6064 	pcp->high = high;
6065 	smp_wmb();
6066 
6067 	pcp->batch = batch;
6068 }
6069 
6070 /* a companion to pageset_set_high() */
6071 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
6072 {
6073 	pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
6074 }
6075 
6076 static void pageset_init(struct per_cpu_pageset *p)
6077 {
6078 	struct per_cpu_pages *pcp;
6079 	int migratetype;
6080 
6081 	memset(p, 0, sizeof(*p));
6082 
6083 	pcp = &p->pcp;
6084 	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
6085 		INIT_LIST_HEAD(&pcp->lists[migratetype]);
6086 }
6087 
6088 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
6089 {
6090 	pageset_init(p);
6091 	pageset_set_batch(p, batch);
6092 }
6093 
6094 /*
6095  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
6096  * to the value high for the pageset p.
6097  */
6098 static void pageset_set_high(struct per_cpu_pageset *p,
6099 				unsigned long high)
6100 {
6101 	unsigned long batch = max(1UL, high / 4);
6102 	if ((high / 4) > (PAGE_SHIFT * 8))
6103 		batch = PAGE_SHIFT * 8;
6104 
6105 	pageset_update(&p->pcp, high, batch);
6106 }
6107 
6108 static void pageset_set_high_and_batch(struct zone *zone,
6109 				       struct per_cpu_pageset *pcp)
6110 {
6111 	if (percpu_pagelist_fraction)
6112 		pageset_set_high(pcp,
6113 			(zone_managed_pages(zone) /
6114 				percpu_pagelist_fraction));
6115 	else
6116 		pageset_set_batch(pcp, zone_batchsize(zone));
6117 }
6118 
6119 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
6120 {
6121 	struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
6122 
6123 	pageset_init(pcp);
6124 	pageset_set_high_and_batch(zone, pcp);
6125 }
6126 
6127 void __meminit setup_zone_pageset(struct zone *zone)
6128 {
6129 	int cpu;
6130 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
6131 	for_each_possible_cpu(cpu)
6132 		zone_pageset_init(zone, cpu);
6133 }
6134 
6135 /*
6136  * Allocate per cpu pagesets and initialize them.
6137  * Before this call only boot pagesets were available.
6138  */
6139 void __init setup_per_cpu_pageset(void)
6140 {
6141 	struct pglist_data *pgdat;
6142 	struct zone *zone;
6143 
6144 	for_each_populated_zone(zone)
6145 		setup_zone_pageset(zone);
6146 
6147 	for_each_online_pgdat(pgdat)
6148 		pgdat->per_cpu_nodestats =
6149 			alloc_percpu(struct per_cpu_nodestat);
6150 }
6151 
6152 static __meminit void zone_pcp_init(struct zone *zone)
6153 {
6154 	/*
6155 	 * per cpu subsystem is not up at this point. The following code
6156 	 * relies on the ability of the linker to provide the
6157 	 * offset of a (static) per cpu variable into the per cpu area.
6158 	 */
6159 	zone->pageset = &boot_pageset;
6160 
6161 	if (populated_zone(zone))
6162 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
6163 			zone->name, zone->present_pages,
6164 					 zone_batchsize(zone));
6165 }
6166 
6167 void __meminit init_currently_empty_zone(struct zone *zone,
6168 					unsigned long zone_start_pfn,
6169 					unsigned long size)
6170 {
6171 	struct pglist_data *pgdat = zone->zone_pgdat;
6172 	int zone_idx = zone_idx(zone) + 1;
6173 
6174 	if (zone_idx > pgdat->nr_zones)
6175 		pgdat->nr_zones = zone_idx;
6176 
6177 	zone->zone_start_pfn = zone_start_pfn;
6178 
6179 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
6180 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
6181 			pgdat->node_id,
6182 			(unsigned long)zone_idx(zone),
6183 			zone_start_pfn, (zone_start_pfn + size));
6184 
6185 	zone_init_free_lists(zone);
6186 	zone->initialized = 1;
6187 }
6188 
6189 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6190 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
6191 
6192 /*
6193  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
6194  */
6195 int __meminit __early_pfn_to_nid(unsigned long pfn,
6196 					struct mminit_pfnnid_cache *state)
6197 {
6198 	unsigned long start_pfn, end_pfn;
6199 	int nid;
6200 
6201 	if (state->last_start <= pfn && pfn < state->last_end)
6202 		return state->last_nid;
6203 
6204 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
6205 	if (nid != NUMA_NO_NODE) {
6206 		state->last_start = start_pfn;
6207 		state->last_end = end_pfn;
6208 		state->last_nid = nid;
6209 	}
6210 
6211 	return nid;
6212 }
6213 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
6214 
6215 /**
6216  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
6217  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
6218  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
6219  *
6220  * If an architecture guarantees that all ranges registered contain no holes
6221  * and may be freed, this this function may be used instead of calling
6222  * memblock_free_early_nid() manually.
6223  */
6224 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
6225 {
6226 	unsigned long start_pfn, end_pfn;
6227 	int i, this_nid;
6228 
6229 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
6230 		start_pfn = min(start_pfn, max_low_pfn);
6231 		end_pfn = min(end_pfn, max_low_pfn);
6232 
6233 		if (start_pfn < end_pfn)
6234 			memblock_free_early_nid(PFN_PHYS(start_pfn),
6235 					(end_pfn - start_pfn) << PAGE_SHIFT,
6236 					this_nid);
6237 	}
6238 }
6239 
6240 /**
6241  * sparse_memory_present_with_active_regions - Call memory_present for each active range
6242  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
6243  *
6244  * If an architecture guarantees that all ranges registered contain no holes and may
6245  * be freed, this function may be used instead of calling memory_present() manually.
6246  */
6247 void __init sparse_memory_present_with_active_regions(int nid)
6248 {
6249 	unsigned long start_pfn, end_pfn;
6250 	int i, this_nid;
6251 
6252 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
6253 		memory_present(this_nid, start_pfn, end_pfn);
6254 }
6255 
6256 /**
6257  * get_pfn_range_for_nid - Return the start and end page frames for a node
6258  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
6259  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
6260  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
6261  *
6262  * It returns the start and end page frame of a node based on information
6263  * provided by memblock_set_node(). If called for a node
6264  * with no available memory, a warning is printed and the start and end
6265  * PFNs will be 0.
6266  */
6267 void __init get_pfn_range_for_nid(unsigned int nid,
6268 			unsigned long *start_pfn, unsigned long *end_pfn)
6269 {
6270 	unsigned long this_start_pfn, this_end_pfn;
6271 	int i;
6272 
6273 	*start_pfn = -1UL;
6274 	*end_pfn = 0;
6275 
6276 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
6277 		*start_pfn = min(*start_pfn, this_start_pfn);
6278 		*end_pfn = max(*end_pfn, this_end_pfn);
6279 	}
6280 
6281 	if (*start_pfn == -1UL)
6282 		*start_pfn = 0;
6283 }
6284 
6285 /*
6286  * This finds a zone that can be used for ZONE_MOVABLE pages. The
6287  * assumption is made that zones within a node are ordered in monotonic
6288  * increasing memory addresses so that the "highest" populated zone is used
6289  */
6290 static void __init find_usable_zone_for_movable(void)
6291 {
6292 	int zone_index;
6293 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
6294 		if (zone_index == ZONE_MOVABLE)
6295 			continue;
6296 
6297 		if (arch_zone_highest_possible_pfn[zone_index] >
6298 				arch_zone_lowest_possible_pfn[zone_index])
6299 			break;
6300 	}
6301 
6302 	VM_BUG_ON(zone_index == -1);
6303 	movable_zone = zone_index;
6304 }
6305 
6306 /*
6307  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
6308  * because it is sized independent of architecture. Unlike the other zones,
6309  * the starting point for ZONE_MOVABLE is not fixed. It may be different
6310  * in each node depending on the size of each node and how evenly kernelcore
6311  * is distributed. This helper function adjusts the zone ranges
6312  * provided by the architecture for a given node by using the end of the
6313  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
6314  * zones within a node are in order of monotonic increases memory addresses
6315  */
6316 static void __init adjust_zone_range_for_zone_movable(int nid,
6317 					unsigned long zone_type,
6318 					unsigned long node_start_pfn,
6319 					unsigned long node_end_pfn,
6320 					unsigned long *zone_start_pfn,
6321 					unsigned long *zone_end_pfn)
6322 {
6323 	/* Only adjust if ZONE_MOVABLE is on this node */
6324 	if (zone_movable_pfn[nid]) {
6325 		/* Size ZONE_MOVABLE */
6326 		if (zone_type == ZONE_MOVABLE) {
6327 			*zone_start_pfn = zone_movable_pfn[nid];
6328 			*zone_end_pfn = min(node_end_pfn,
6329 				arch_zone_highest_possible_pfn[movable_zone]);
6330 
6331 		/* Adjust for ZONE_MOVABLE starting within this range */
6332 		} else if (!mirrored_kernelcore &&
6333 			*zone_start_pfn < zone_movable_pfn[nid] &&
6334 			*zone_end_pfn > zone_movable_pfn[nid]) {
6335 			*zone_end_pfn = zone_movable_pfn[nid];
6336 
6337 		/* Check if this whole range is within ZONE_MOVABLE */
6338 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
6339 			*zone_start_pfn = *zone_end_pfn;
6340 	}
6341 }
6342 
6343 /*
6344  * Return the number of pages a zone spans in a node, including holes
6345  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
6346  */
6347 static unsigned long __init zone_spanned_pages_in_node(int nid,
6348 					unsigned long zone_type,
6349 					unsigned long node_start_pfn,
6350 					unsigned long node_end_pfn,
6351 					unsigned long *zone_start_pfn,
6352 					unsigned long *zone_end_pfn,
6353 					unsigned long *ignored)
6354 {
6355 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6356 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6357 	/* When hotadd a new node from cpu_up(), the node should be empty */
6358 	if (!node_start_pfn && !node_end_pfn)
6359 		return 0;
6360 
6361 	/* Get the start and end of the zone */
6362 	*zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6363 	*zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6364 	adjust_zone_range_for_zone_movable(nid, zone_type,
6365 				node_start_pfn, node_end_pfn,
6366 				zone_start_pfn, zone_end_pfn);
6367 
6368 	/* Check that this node has pages within the zone's required range */
6369 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6370 		return 0;
6371 
6372 	/* Move the zone boundaries inside the node if necessary */
6373 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6374 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6375 
6376 	/* Return the spanned pages */
6377 	return *zone_end_pfn - *zone_start_pfn;
6378 }
6379 
6380 /*
6381  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
6382  * then all holes in the requested range will be accounted for.
6383  */
6384 unsigned long __init __absent_pages_in_range(int nid,
6385 				unsigned long range_start_pfn,
6386 				unsigned long range_end_pfn)
6387 {
6388 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
6389 	unsigned long start_pfn, end_pfn;
6390 	int i;
6391 
6392 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6393 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6394 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6395 		nr_absent -= end_pfn - start_pfn;
6396 	}
6397 	return nr_absent;
6398 }
6399 
6400 /**
6401  * absent_pages_in_range - Return number of page frames in holes within a range
6402  * @start_pfn: The start PFN to start searching for holes
6403  * @end_pfn: The end PFN to stop searching for holes
6404  *
6405  * Return: the number of pages frames in memory holes within a range.
6406  */
6407 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6408 							unsigned long end_pfn)
6409 {
6410 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6411 }
6412 
6413 /* Return the number of page frames in holes in a zone on a node */
6414 static unsigned long __init zone_absent_pages_in_node(int nid,
6415 					unsigned long zone_type,
6416 					unsigned long node_start_pfn,
6417 					unsigned long node_end_pfn,
6418 					unsigned long *ignored)
6419 {
6420 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6421 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6422 	unsigned long zone_start_pfn, zone_end_pfn;
6423 	unsigned long nr_absent;
6424 
6425 	/* When hotadd a new node from cpu_up(), the node should be empty */
6426 	if (!node_start_pfn && !node_end_pfn)
6427 		return 0;
6428 
6429 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6430 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6431 
6432 	adjust_zone_range_for_zone_movable(nid, zone_type,
6433 			node_start_pfn, node_end_pfn,
6434 			&zone_start_pfn, &zone_end_pfn);
6435 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6436 
6437 	/*
6438 	 * ZONE_MOVABLE handling.
6439 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6440 	 * and vice versa.
6441 	 */
6442 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6443 		unsigned long start_pfn, end_pfn;
6444 		struct memblock_region *r;
6445 
6446 		for_each_memblock(memory, r) {
6447 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
6448 					  zone_start_pfn, zone_end_pfn);
6449 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
6450 					zone_start_pfn, zone_end_pfn);
6451 
6452 			if (zone_type == ZONE_MOVABLE &&
6453 			    memblock_is_mirror(r))
6454 				nr_absent += end_pfn - start_pfn;
6455 
6456 			if (zone_type == ZONE_NORMAL &&
6457 			    !memblock_is_mirror(r))
6458 				nr_absent += end_pfn - start_pfn;
6459 		}
6460 	}
6461 
6462 	return nr_absent;
6463 }
6464 
6465 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6466 static inline unsigned long __init zone_spanned_pages_in_node(int nid,
6467 					unsigned long zone_type,
6468 					unsigned long node_start_pfn,
6469 					unsigned long node_end_pfn,
6470 					unsigned long *zone_start_pfn,
6471 					unsigned long *zone_end_pfn,
6472 					unsigned long *zones_size)
6473 {
6474 	unsigned int zone;
6475 
6476 	*zone_start_pfn = node_start_pfn;
6477 	for (zone = 0; zone < zone_type; zone++)
6478 		*zone_start_pfn += zones_size[zone];
6479 
6480 	*zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
6481 
6482 	return zones_size[zone_type];
6483 }
6484 
6485 static inline unsigned long __init zone_absent_pages_in_node(int nid,
6486 						unsigned long zone_type,
6487 						unsigned long node_start_pfn,
6488 						unsigned long node_end_pfn,
6489 						unsigned long *zholes_size)
6490 {
6491 	if (!zholes_size)
6492 		return 0;
6493 
6494 	return zholes_size[zone_type];
6495 }
6496 
6497 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6498 
6499 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
6500 						unsigned long node_start_pfn,
6501 						unsigned long node_end_pfn,
6502 						unsigned long *zones_size,
6503 						unsigned long *zholes_size)
6504 {
6505 	unsigned long realtotalpages = 0, totalpages = 0;
6506 	enum zone_type i;
6507 
6508 	for (i = 0; i < MAX_NR_ZONES; i++) {
6509 		struct zone *zone = pgdat->node_zones + i;
6510 		unsigned long zone_start_pfn, zone_end_pfn;
6511 		unsigned long size, real_size;
6512 
6513 		size = zone_spanned_pages_in_node(pgdat->node_id, i,
6514 						  node_start_pfn,
6515 						  node_end_pfn,
6516 						  &zone_start_pfn,
6517 						  &zone_end_pfn,
6518 						  zones_size);
6519 		real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
6520 						  node_start_pfn, node_end_pfn,
6521 						  zholes_size);
6522 		if (size)
6523 			zone->zone_start_pfn = zone_start_pfn;
6524 		else
6525 			zone->zone_start_pfn = 0;
6526 		zone->spanned_pages = size;
6527 		zone->present_pages = real_size;
6528 
6529 		totalpages += size;
6530 		realtotalpages += real_size;
6531 	}
6532 
6533 	pgdat->node_spanned_pages = totalpages;
6534 	pgdat->node_present_pages = realtotalpages;
6535 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6536 							realtotalpages);
6537 }
6538 
6539 #ifndef CONFIG_SPARSEMEM
6540 /*
6541  * Calculate the size of the zone->blockflags rounded to an unsigned long
6542  * Start by making sure zonesize is a multiple of pageblock_order by rounding
6543  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6544  * round what is now in bits to nearest long in bits, then return it in
6545  * bytes.
6546  */
6547 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6548 {
6549 	unsigned long usemapsize;
6550 
6551 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6552 	usemapsize = roundup(zonesize, pageblock_nr_pages);
6553 	usemapsize = usemapsize >> pageblock_order;
6554 	usemapsize *= NR_PAGEBLOCK_BITS;
6555 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6556 
6557 	return usemapsize / 8;
6558 }
6559 
6560 static void __ref setup_usemap(struct pglist_data *pgdat,
6561 				struct zone *zone,
6562 				unsigned long zone_start_pfn,
6563 				unsigned long zonesize)
6564 {
6565 	unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6566 	zone->pageblock_flags = NULL;
6567 	if (usemapsize) {
6568 		zone->pageblock_flags =
6569 			memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
6570 					    pgdat->node_id);
6571 		if (!zone->pageblock_flags)
6572 			panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
6573 			      usemapsize, zone->name, pgdat->node_id);
6574 	}
6575 }
6576 #else
6577 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6578 				unsigned long zone_start_pfn, unsigned long zonesize) {}
6579 #endif /* CONFIG_SPARSEMEM */
6580 
6581 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6582 
6583 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6584 void __init set_pageblock_order(void)
6585 {
6586 	unsigned int order;
6587 
6588 	/* Check that pageblock_nr_pages has not already been setup */
6589 	if (pageblock_order)
6590 		return;
6591 
6592 	if (HPAGE_SHIFT > PAGE_SHIFT)
6593 		order = HUGETLB_PAGE_ORDER;
6594 	else
6595 		order = MAX_ORDER - 1;
6596 
6597 	/*
6598 	 * Assume the largest contiguous order of interest is a huge page.
6599 	 * This value may be variable depending on boot parameters on IA64 and
6600 	 * powerpc.
6601 	 */
6602 	pageblock_order = order;
6603 }
6604 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6605 
6606 /*
6607  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6608  * is unused as pageblock_order is set at compile-time. See
6609  * include/linux/pageblock-flags.h for the values of pageblock_order based on
6610  * the kernel config
6611  */
6612 void __init set_pageblock_order(void)
6613 {
6614 }
6615 
6616 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6617 
6618 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6619 						unsigned long present_pages)
6620 {
6621 	unsigned long pages = spanned_pages;
6622 
6623 	/*
6624 	 * Provide a more accurate estimation if there are holes within
6625 	 * the zone and SPARSEMEM is in use. If there are holes within the
6626 	 * zone, each populated memory region may cost us one or two extra
6627 	 * memmap pages due to alignment because memmap pages for each
6628 	 * populated regions may not be naturally aligned on page boundary.
6629 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6630 	 */
6631 	if (spanned_pages > present_pages + (present_pages >> 4) &&
6632 	    IS_ENABLED(CONFIG_SPARSEMEM))
6633 		pages = present_pages;
6634 
6635 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6636 }
6637 
6638 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6639 static void pgdat_init_split_queue(struct pglist_data *pgdat)
6640 {
6641 	spin_lock_init(&pgdat->split_queue_lock);
6642 	INIT_LIST_HEAD(&pgdat->split_queue);
6643 	pgdat->split_queue_len = 0;
6644 }
6645 #else
6646 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6647 #endif
6648 
6649 #ifdef CONFIG_COMPACTION
6650 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6651 {
6652 	init_waitqueue_head(&pgdat->kcompactd_wait);
6653 }
6654 #else
6655 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6656 #endif
6657 
6658 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6659 {
6660 	pgdat_resize_init(pgdat);
6661 
6662 	pgdat_init_split_queue(pgdat);
6663 	pgdat_init_kcompactd(pgdat);
6664 
6665 	init_waitqueue_head(&pgdat->kswapd_wait);
6666 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
6667 
6668 	pgdat_page_ext_init(pgdat);
6669 	spin_lock_init(&pgdat->lru_lock);
6670 	lruvec_init(node_lruvec(pgdat));
6671 }
6672 
6673 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6674 							unsigned long remaining_pages)
6675 {
6676 	atomic_long_set(&zone->managed_pages, remaining_pages);
6677 	zone_set_nid(zone, nid);
6678 	zone->name = zone_names[idx];
6679 	zone->zone_pgdat = NODE_DATA(nid);
6680 	spin_lock_init(&zone->lock);
6681 	zone_seqlock_init(zone);
6682 	zone_pcp_init(zone);
6683 }
6684 
6685 /*
6686  * Set up the zone data structures
6687  * - init pgdat internals
6688  * - init all zones belonging to this node
6689  *
6690  * NOTE: this function is only called during memory hotplug
6691  */
6692 #ifdef CONFIG_MEMORY_HOTPLUG
6693 void __ref free_area_init_core_hotplug(int nid)
6694 {
6695 	enum zone_type z;
6696 	pg_data_t *pgdat = NODE_DATA(nid);
6697 
6698 	pgdat_init_internals(pgdat);
6699 	for (z = 0; z < MAX_NR_ZONES; z++)
6700 		zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6701 }
6702 #endif
6703 
6704 /*
6705  * Set up the zone data structures:
6706  *   - mark all pages reserved
6707  *   - mark all memory queues empty
6708  *   - clear the memory bitmaps
6709  *
6710  * NOTE: pgdat should get zeroed by caller.
6711  * NOTE: this function is only called during early init.
6712  */
6713 static void __init free_area_init_core(struct pglist_data *pgdat)
6714 {
6715 	enum zone_type j;
6716 	int nid = pgdat->node_id;
6717 
6718 	pgdat_init_internals(pgdat);
6719 	pgdat->per_cpu_nodestats = &boot_nodestats;
6720 
6721 	for (j = 0; j < MAX_NR_ZONES; j++) {
6722 		struct zone *zone = pgdat->node_zones + j;
6723 		unsigned long size, freesize, memmap_pages;
6724 		unsigned long zone_start_pfn = zone->zone_start_pfn;
6725 
6726 		size = zone->spanned_pages;
6727 		freesize = zone->present_pages;
6728 
6729 		/*
6730 		 * Adjust freesize so that it accounts for how much memory
6731 		 * is used by this zone for memmap. This affects the watermark
6732 		 * and per-cpu initialisations
6733 		 */
6734 		memmap_pages = calc_memmap_size(size, freesize);
6735 		if (!is_highmem_idx(j)) {
6736 			if (freesize >= memmap_pages) {
6737 				freesize -= memmap_pages;
6738 				if (memmap_pages)
6739 					printk(KERN_DEBUG
6740 					       "  %s zone: %lu pages used for memmap\n",
6741 					       zone_names[j], memmap_pages);
6742 			} else
6743 				pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
6744 					zone_names[j], memmap_pages, freesize);
6745 		}
6746 
6747 		/* Account for reserved pages */
6748 		if (j == 0 && freesize > dma_reserve) {
6749 			freesize -= dma_reserve;
6750 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
6751 					zone_names[0], dma_reserve);
6752 		}
6753 
6754 		if (!is_highmem_idx(j))
6755 			nr_kernel_pages += freesize;
6756 		/* Charge for highmem memmap if there are enough kernel pages */
6757 		else if (nr_kernel_pages > memmap_pages * 2)
6758 			nr_kernel_pages -= memmap_pages;
6759 		nr_all_pages += freesize;
6760 
6761 		/*
6762 		 * Set an approximate value for lowmem here, it will be adjusted
6763 		 * when the bootmem allocator frees pages into the buddy system.
6764 		 * And all highmem pages will be managed by the buddy system.
6765 		 */
6766 		zone_init_internals(zone, j, nid, freesize);
6767 
6768 		if (!size)
6769 			continue;
6770 
6771 		set_pageblock_order();
6772 		setup_usemap(pgdat, zone, zone_start_pfn, size);
6773 		init_currently_empty_zone(zone, zone_start_pfn, size);
6774 		memmap_init(size, nid, j, zone_start_pfn);
6775 	}
6776 }
6777 
6778 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6779 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6780 {
6781 	unsigned long __maybe_unused start = 0;
6782 	unsigned long __maybe_unused offset = 0;
6783 
6784 	/* Skip empty nodes */
6785 	if (!pgdat->node_spanned_pages)
6786 		return;
6787 
6788 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6789 	offset = pgdat->node_start_pfn - start;
6790 	/* ia64 gets its own node_mem_map, before this, without bootmem */
6791 	if (!pgdat->node_mem_map) {
6792 		unsigned long size, end;
6793 		struct page *map;
6794 
6795 		/*
6796 		 * The zone's endpoints aren't required to be MAX_ORDER
6797 		 * aligned but the node_mem_map endpoints must be in order
6798 		 * for the buddy allocator to function correctly.
6799 		 */
6800 		end = pgdat_end_pfn(pgdat);
6801 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
6802 		size =  (end - start) * sizeof(struct page);
6803 		map = memblock_alloc_node(size, SMP_CACHE_BYTES,
6804 					  pgdat->node_id);
6805 		if (!map)
6806 			panic("Failed to allocate %ld bytes for node %d memory map\n",
6807 			      size, pgdat->node_id);
6808 		pgdat->node_mem_map = map + offset;
6809 	}
6810 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6811 				__func__, pgdat->node_id, (unsigned long)pgdat,
6812 				(unsigned long)pgdat->node_mem_map);
6813 #ifndef CONFIG_NEED_MULTIPLE_NODES
6814 	/*
6815 	 * With no DISCONTIG, the global mem_map is just set as node 0's
6816 	 */
6817 	if (pgdat == NODE_DATA(0)) {
6818 		mem_map = NODE_DATA(0)->node_mem_map;
6819 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6820 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6821 			mem_map -= offset;
6822 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6823 	}
6824 #endif
6825 }
6826 #else
6827 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6828 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6829 
6830 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6831 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
6832 {
6833 	pgdat->first_deferred_pfn = ULONG_MAX;
6834 }
6835 #else
6836 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
6837 #endif
6838 
6839 void __init free_area_init_node(int nid, unsigned long *zones_size,
6840 				   unsigned long node_start_pfn,
6841 				   unsigned long *zholes_size)
6842 {
6843 	pg_data_t *pgdat = NODE_DATA(nid);
6844 	unsigned long start_pfn = 0;
6845 	unsigned long end_pfn = 0;
6846 
6847 	/* pg_data_t should be reset to zero when it's allocated */
6848 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
6849 
6850 	pgdat->node_id = nid;
6851 	pgdat->node_start_pfn = node_start_pfn;
6852 	pgdat->per_cpu_nodestats = NULL;
6853 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6854 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6855 	pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6856 		(u64)start_pfn << PAGE_SHIFT,
6857 		end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6858 #else
6859 	start_pfn = node_start_pfn;
6860 #endif
6861 	calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6862 				  zones_size, zholes_size);
6863 
6864 	alloc_node_mem_map(pgdat);
6865 	pgdat_set_deferred_range(pgdat);
6866 
6867 	free_area_init_core(pgdat);
6868 }
6869 
6870 #if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6871 /*
6872  * Zero all valid struct pages in range [spfn, epfn), return number of struct
6873  * pages zeroed
6874  */
6875 static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
6876 {
6877 	unsigned long pfn;
6878 	u64 pgcnt = 0;
6879 
6880 	for (pfn = spfn; pfn < epfn; pfn++) {
6881 		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6882 			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6883 				+ pageblock_nr_pages - 1;
6884 			continue;
6885 		}
6886 		mm_zero_struct_page(pfn_to_page(pfn));
6887 		pgcnt++;
6888 	}
6889 
6890 	return pgcnt;
6891 }
6892 
6893 /*
6894  * Only struct pages that are backed by physical memory are zeroed and
6895  * initialized by going through __init_single_page(). But, there are some
6896  * struct pages which are reserved in memblock allocator and their fields
6897  * may be accessed (for example page_to_pfn() on some configuration accesses
6898  * flags). We must explicitly zero those struct pages.
6899  *
6900  * This function also addresses a similar issue where struct pages are left
6901  * uninitialized because the physical address range is not covered by
6902  * memblock.memory or memblock.reserved. That could happen when memblock
6903  * layout is manually configured via memmap=.
6904  */
6905 void __init zero_resv_unavail(void)
6906 {
6907 	phys_addr_t start, end;
6908 	u64 i, pgcnt;
6909 	phys_addr_t next = 0;
6910 
6911 	/*
6912 	 * Loop through unavailable ranges not covered by memblock.memory.
6913 	 */
6914 	pgcnt = 0;
6915 	for_each_mem_range(i, &memblock.memory, NULL,
6916 			NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
6917 		if (next < start)
6918 			pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
6919 		next = end;
6920 	}
6921 	pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
6922 
6923 	/*
6924 	 * Struct pages that do not have backing memory. This could be because
6925 	 * firmware is using some of this memory, or for some other reasons.
6926 	 */
6927 	if (pgcnt)
6928 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
6929 }
6930 #endif /* !CONFIG_FLAT_NODE_MEM_MAP */
6931 
6932 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6933 
6934 #if MAX_NUMNODES > 1
6935 /*
6936  * Figure out the number of possible node ids.
6937  */
6938 void __init setup_nr_node_ids(void)
6939 {
6940 	unsigned int highest;
6941 
6942 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
6943 	nr_node_ids = highest + 1;
6944 }
6945 #endif
6946 
6947 /**
6948  * node_map_pfn_alignment - determine the maximum internode alignment
6949  *
6950  * This function should be called after node map is populated and sorted.
6951  * It calculates the maximum power of two alignment which can distinguish
6952  * all the nodes.
6953  *
6954  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6955  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
6956  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
6957  * shifted, 1GiB is enough and this function will indicate so.
6958  *
6959  * This is used to test whether pfn -> nid mapping of the chosen memory
6960  * model has fine enough granularity to avoid incorrect mapping for the
6961  * populated node map.
6962  *
6963  * Return: the determined alignment in pfn's.  0 if there is no alignment
6964  * requirement (single node).
6965  */
6966 unsigned long __init node_map_pfn_alignment(void)
6967 {
6968 	unsigned long accl_mask = 0, last_end = 0;
6969 	unsigned long start, end, mask;
6970 	int last_nid = NUMA_NO_NODE;
6971 	int i, nid;
6972 
6973 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
6974 		if (!start || last_nid < 0 || last_nid == nid) {
6975 			last_nid = nid;
6976 			last_end = end;
6977 			continue;
6978 		}
6979 
6980 		/*
6981 		 * Start with a mask granular enough to pin-point to the
6982 		 * start pfn and tick off bits one-by-one until it becomes
6983 		 * too coarse to separate the current node from the last.
6984 		 */
6985 		mask = ~((1 << __ffs(start)) - 1);
6986 		while (mask && last_end <= (start & (mask << 1)))
6987 			mask <<= 1;
6988 
6989 		/* accumulate all internode masks */
6990 		accl_mask |= mask;
6991 	}
6992 
6993 	/* convert mask to number of pages */
6994 	return ~accl_mask + 1;
6995 }
6996 
6997 /* Find the lowest pfn for a node */
6998 static unsigned long __init find_min_pfn_for_node(int nid)
6999 {
7000 	unsigned long min_pfn = ULONG_MAX;
7001 	unsigned long start_pfn;
7002 	int i;
7003 
7004 	for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
7005 		min_pfn = min(min_pfn, start_pfn);
7006 
7007 	if (min_pfn == ULONG_MAX) {
7008 		pr_warn("Could not find start_pfn for node %d\n", nid);
7009 		return 0;
7010 	}
7011 
7012 	return min_pfn;
7013 }
7014 
7015 /**
7016  * find_min_pfn_with_active_regions - Find the minimum PFN registered
7017  *
7018  * Return: the minimum PFN based on information provided via
7019  * memblock_set_node().
7020  */
7021 unsigned long __init find_min_pfn_with_active_regions(void)
7022 {
7023 	return find_min_pfn_for_node(MAX_NUMNODES);
7024 }
7025 
7026 /*
7027  * early_calculate_totalpages()
7028  * Sum pages in active regions for movable zone.
7029  * Populate N_MEMORY for calculating usable_nodes.
7030  */
7031 static unsigned long __init early_calculate_totalpages(void)
7032 {
7033 	unsigned long totalpages = 0;
7034 	unsigned long start_pfn, end_pfn;
7035 	int i, nid;
7036 
7037 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7038 		unsigned long pages = end_pfn - start_pfn;
7039 
7040 		totalpages += pages;
7041 		if (pages)
7042 			node_set_state(nid, N_MEMORY);
7043 	}
7044 	return totalpages;
7045 }
7046 
7047 /*
7048  * Find the PFN the Movable zone begins in each node. Kernel memory
7049  * is spread evenly between nodes as long as the nodes have enough
7050  * memory. When they don't, some nodes will have more kernelcore than
7051  * others
7052  */
7053 static void __init find_zone_movable_pfns_for_nodes(void)
7054 {
7055 	int i, nid;
7056 	unsigned long usable_startpfn;
7057 	unsigned long kernelcore_node, kernelcore_remaining;
7058 	/* save the state before borrow the nodemask */
7059 	nodemask_t saved_node_state = node_states[N_MEMORY];
7060 	unsigned long totalpages = early_calculate_totalpages();
7061 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7062 	struct memblock_region *r;
7063 
7064 	/* Need to find movable_zone earlier when movable_node is specified. */
7065 	find_usable_zone_for_movable();
7066 
7067 	/*
7068 	 * If movable_node is specified, ignore kernelcore and movablecore
7069 	 * options.
7070 	 */
7071 	if (movable_node_is_enabled()) {
7072 		for_each_memblock(memory, r) {
7073 			if (!memblock_is_hotpluggable(r))
7074 				continue;
7075 
7076 			nid = r->nid;
7077 
7078 			usable_startpfn = PFN_DOWN(r->base);
7079 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7080 				min(usable_startpfn, zone_movable_pfn[nid]) :
7081 				usable_startpfn;
7082 		}
7083 
7084 		goto out2;
7085 	}
7086 
7087 	/*
7088 	 * If kernelcore=mirror is specified, ignore movablecore option
7089 	 */
7090 	if (mirrored_kernelcore) {
7091 		bool mem_below_4gb_not_mirrored = false;
7092 
7093 		for_each_memblock(memory, r) {
7094 			if (memblock_is_mirror(r))
7095 				continue;
7096 
7097 			nid = r->nid;
7098 
7099 			usable_startpfn = memblock_region_memory_base_pfn(r);
7100 
7101 			if (usable_startpfn < 0x100000) {
7102 				mem_below_4gb_not_mirrored = true;
7103 				continue;
7104 			}
7105 
7106 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7107 				min(usable_startpfn, zone_movable_pfn[nid]) :
7108 				usable_startpfn;
7109 		}
7110 
7111 		if (mem_below_4gb_not_mirrored)
7112 			pr_warn("This configuration results in unmirrored kernel memory.");
7113 
7114 		goto out2;
7115 	}
7116 
7117 	/*
7118 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
7119 	 * amount of necessary memory.
7120 	 */
7121 	if (required_kernelcore_percent)
7122 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7123 				       10000UL;
7124 	if (required_movablecore_percent)
7125 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7126 					10000UL;
7127 
7128 	/*
7129 	 * If movablecore= was specified, calculate what size of
7130 	 * kernelcore that corresponds so that memory usable for
7131 	 * any allocation type is evenly spread. If both kernelcore
7132 	 * and movablecore are specified, then the value of kernelcore
7133 	 * will be used for required_kernelcore if it's greater than
7134 	 * what movablecore would have allowed.
7135 	 */
7136 	if (required_movablecore) {
7137 		unsigned long corepages;
7138 
7139 		/*
7140 		 * Round-up so that ZONE_MOVABLE is at least as large as what
7141 		 * was requested by the user
7142 		 */
7143 		required_movablecore =
7144 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7145 		required_movablecore = min(totalpages, required_movablecore);
7146 		corepages = totalpages - required_movablecore;
7147 
7148 		required_kernelcore = max(required_kernelcore, corepages);
7149 	}
7150 
7151 	/*
7152 	 * If kernelcore was not specified or kernelcore size is larger
7153 	 * than totalpages, there is no ZONE_MOVABLE.
7154 	 */
7155 	if (!required_kernelcore || required_kernelcore >= totalpages)
7156 		goto out;
7157 
7158 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
7159 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7160 
7161 restart:
7162 	/* Spread kernelcore memory as evenly as possible throughout nodes */
7163 	kernelcore_node = required_kernelcore / usable_nodes;
7164 	for_each_node_state(nid, N_MEMORY) {
7165 		unsigned long start_pfn, end_pfn;
7166 
7167 		/*
7168 		 * Recalculate kernelcore_node if the division per node
7169 		 * now exceeds what is necessary to satisfy the requested
7170 		 * amount of memory for the kernel
7171 		 */
7172 		if (required_kernelcore < kernelcore_node)
7173 			kernelcore_node = required_kernelcore / usable_nodes;
7174 
7175 		/*
7176 		 * As the map is walked, we track how much memory is usable
7177 		 * by the kernel using kernelcore_remaining. When it is
7178 		 * 0, the rest of the node is usable by ZONE_MOVABLE
7179 		 */
7180 		kernelcore_remaining = kernelcore_node;
7181 
7182 		/* Go through each range of PFNs within this node */
7183 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7184 			unsigned long size_pages;
7185 
7186 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7187 			if (start_pfn >= end_pfn)
7188 				continue;
7189 
7190 			/* Account for what is only usable for kernelcore */
7191 			if (start_pfn < usable_startpfn) {
7192 				unsigned long kernel_pages;
7193 				kernel_pages = min(end_pfn, usable_startpfn)
7194 								- start_pfn;
7195 
7196 				kernelcore_remaining -= min(kernel_pages,
7197 							kernelcore_remaining);
7198 				required_kernelcore -= min(kernel_pages,
7199 							required_kernelcore);
7200 
7201 				/* Continue if range is now fully accounted */
7202 				if (end_pfn <= usable_startpfn) {
7203 
7204 					/*
7205 					 * Push zone_movable_pfn to the end so
7206 					 * that if we have to rebalance
7207 					 * kernelcore across nodes, we will
7208 					 * not double account here
7209 					 */
7210 					zone_movable_pfn[nid] = end_pfn;
7211 					continue;
7212 				}
7213 				start_pfn = usable_startpfn;
7214 			}
7215 
7216 			/*
7217 			 * The usable PFN range for ZONE_MOVABLE is from
7218 			 * start_pfn->end_pfn. Calculate size_pages as the
7219 			 * number of pages used as kernelcore
7220 			 */
7221 			size_pages = end_pfn - start_pfn;
7222 			if (size_pages > kernelcore_remaining)
7223 				size_pages = kernelcore_remaining;
7224 			zone_movable_pfn[nid] = start_pfn + size_pages;
7225 
7226 			/*
7227 			 * Some kernelcore has been met, update counts and
7228 			 * break if the kernelcore for this node has been
7229 			 * satisfied
7230 			 */
7231 			required_kernelcore -= min(required_kernelcore,
7232 								size_pages);
7233 			kernelcore_remaining -= size_pages;
7234 			if (!kernelcore_remaining)
7235 				break;
7236 		}
7237 	}
7238 
7239 	/*
7240 	 * If there is still required_kernelcore, we do another pass with one
7241 	 * less node in the count. This will push zone_movable_pfn[nid] further
7242 	 * along on the nodes that still have memory until kernelcore is
7243 	 * satisfied
7244 	 */
7245 	usable_nodes--;
7246 	if (usable_nodes && required_kernelcore > usable_nodes)
7247 		goto restart;
7248 
7249 out2:
7250 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
7251 	for (nid = 0; nid < MAX_NUMNODES; nid++)
7252 		zone_movable_pfn[nid] =
7253 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7254 
7255 out:
7256 	/* restore the node_state */
7257 	node_states[N_MEMORY] = saved_node_state;
7258 }
7259 
7260 /* Any regular or high memory on that node ? */
7261 static void check_for_memory(pg_data_t *pgdat, int nid)
7262 {
7263 	enum zone_type zone_type;
7264 
7265 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7266 		struct zone *zone = &pgdat->node_zones[zone_type];
7267 		if (populated_zone(zone)) {
7268 			if (IS_ENABLED(CONFIG_HIGHMEM))
7269 				node_set_state(nid, N_HIGH_MEMORY);
7270 			if (zone_type <= ZONE_NORMAL)
7271 				node_set_state(nid, N_NORMAL_MEMORY);
7272 			break;
7273 		}
7274 	}
7275 }
7276 
7277 /**
7278  * free_area_init_nodes - Initialise all pg_data_t and zone data
7279  * @max_zone_pfn: an array of max PFNs for each zone
7280  *
7281  * This will call free_area_init_node() for each active node in the system.
7282  * Using the page ranges provided by memblock_set_node(), the size of each
7283  * zone in each node and their holes is calculated. If the maximum PFN
7284  * between two adjacent zones match, it is assumed that the zone is empty.
7285  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
7286  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
7287  * starts where the previous one ended. For example, ZONE_DMA32 starts
7288  * at arch_max_dma_pfn.
7289  */
7290 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
7291 {
7292 	unsigned long start_pfn, end_pfn;
7293 	int i, nid;
7294 
7295 	/* Record where the zone boundaries are */
7296 	memset(arch_zone_lowest_possible_pfn, 0,
7297 				sizeof(arch_zone_lowest_possible_pfn));
7298 	memset(arch_zone_highest_possible_pfn, 0,
7299 				sizeof(arch_zone_highest_possible_pfn));
7300 
7301 	start_pfn = find_min_pfn_with_active_regions();
7302 
7303 	for (i = 0; i < MAX_NR_ZONES; i++) {
7304 		if (i == ZONE_MOVABLE)
7305 			continue;
7306 
7307 		end_pfn = max(max_zone_pfn[i], start_pfn);
7308 		arch_zone_lowest_possible_pfn[i] = start_pfn;
7309 		arch_zone_highest_possible_pfn[i] = end_pfn;
7310 
7311 		start_pfn = end_pfn;
7312 	}
7313 
7314 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
7315 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7316 	find_zone_movable_pfns_for_nodes();
7317 
7318 	/* Print out the zone ranges */
7319 	pr_info("Zone ranges:\n");
7320 	for (i = 0; i < MAX_NR_ZONES; i++) {
7321 		if (i == ZONE_MOVABLE)
7322 			continue;
7323 		pr_info("  %-8s ", zone_names[i]);
7324 		if (arch_zone_lowest_possible_pfn[i] ==
7325 				arch_zone_highest_possible_pfn[i])
7326 			pr_cont("empty\n");
7327 		else
7328 			pr_cont("[mem %#018Lx-%#018Lx]\n",
7329 				(u64)arch_zone_lowest_possible_pfn[i]
7330 					<< PAGE_SHIFT,
7331 				((u64)arch_zone_highest_possible_pfn[i]
7332 					<< PAGE_SHIFT) - 1);
7333 	}
7334 
7335 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
7336 	pr_info("Movable zone start for each node\n");
7337 	for (i = 0; i < MAX_NUMNODES; i++) {
7338 		if (zone_movable_pfn[i])
7339 			pr_info("  Node %d: %#018Lx\n", i,
7340 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7341 	}
7342 
7343 	/*
7344 	 * Print out the early node map, and initialize the
7345 	 * subsection-map relative to active online memory ranges to
7346 	 * enable future "sub-section" extensions of the memory map.
7347 	 */
7348 	pr_info("Early memory node ranges\n");
7349 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7350 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7351 			(u64)start_pfn << PAGE_SHIFT,
7352 			((u64)end_pfn << PAGE_SHIFT) - 1);
7353 		subsection_map_init(start_pfn, end_pfn - start_pfn);
7354 	}
7355 
7356 	/* Initialise every node */
7357 	mminit_verify_pageflags_layout();
7358 	setup_nr_node_ids();
7359 	zero_resv_unavail();
7360 	for_each_online_node(nid) {
7361 		pg_data_t *pgdat = NODE_DATA(nid);
7362 		free_area_init_node(nid, NULL,
7363 				find_min_pfn_for_node(nid), NULL);
7364 
7365 		/* Any memory on that node */
7366 		if (pgdat->node_present_pages)
7367 			node_set_state(nid, N_MEMORY);
7368 		check_for_memory(pgdat, nid);
7369 	}
7370 }
7371 
7372 static int __init cmdline_parse_core(char *p, unsigned long *core,
7373 				     unsigned long *percent)
7374 {
7375 	unsigned long long coremem;
7376 	char *endptr;
7377 
7378 	if (!p)
7379 		return -EINVAL;
7380 
7381 	/* Value may be a percentage of total memory, otherwise bytes */
7382 	coremem = simple_strtoull(p, &endptr, 0);
7383 	if (*endptr == '%') {
7384 		/* Paranoid check for percent values greater than 100 */
7385 		WARN_ON(coremem > 100);
7386 
7387 		*percent = coremem;
7388 	} else {
7389 		coremem = memparse(p, &p);
7390 		/* Paranoid check that UL is enough for the coremem value */
7391 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7392 
7393 		*core = coremem >> PAGE_SHIFT;
7394 		*percent = 0UL;
7395 	}
7396 	return 0;
7397 }
7398 
7399 /*
7400  * kernelcore=size sets the amount of memory for use for allocations that
7401  * cannot be reclaimed or migrated.
7402  */
7403 static int __init cmdline_parse_kernelcore(char *p)
7404 {
7405 	/* parse kernelcore=mirror */
7406 	if (parse_option_str(p, "mirror")) {
7407 		mirrored_kernelcore = true;
7408 		return 0;
7409 	}
7410 
7411 	return cmdline_parse_core(p, &required_kernelcore,
7412 				  &required_kernelcore_percent);
7413 }
7414 
7415 /*
7416  * movablecore=size sets the amount of memory for use for allocations that
7417  * can be reclaimed or migrated.
7418  */
7419 static int __init cmdline_parse_movablecore(char *p)
7420 {
7421 	return cmdline_parse_core(p, &required_movablecore,
7422 				  &required_movablecore_percent);
7423 }
7424 
7425 early_param("kernelcore", cmdline_parse_kernelcore);
7426 early_param("movablecore", cmdline_parse_movablecore);
7427 
7428 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
7429 
7430 void adjust_managed_page_count(struct page *page, long count)
7431 {
7432 	atomic_long_add(count, &page_zone(page)->managed_pages);
7433 	totalram_pages_add(count);
7434 #ifdef CONFIG_HIGHMEM
7435 	if (PageHighMem(page))
7436 		totalhigh_pages_add(count);
7437 #endif
7438 }
7439 EXPORT_SYMBOL(adjust_managed_page_count);
7440 
7441 unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
7442 {
7443 	void *pos;
7444 	unsigned long pages = 0;
7445 
7446 	start = (void *)PAGE_ALIGN((unsigned long)start);
7447 	end = (void *)((unsigned long)end & PAGE_MASK);
7448 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7449 		struct page *page = virt_to_page(pos);
7450 		void *direct_map_addr;
7451 
7452 		/*
7453 		 * 'direct_map_addr' might be different from 'pos'
7454 		 * because some architectures' virt_to_page()
7455 		 * work with aliases.  Getting the direct map
7456 		 * address ensures that we get a _writeable_
7457 		 * alias for the memset().
7458 		 */
7459 		direct_map_addr = page_address(page);
7460 		if ((unsigned int)poison <= 0xFF)
7461 			memset(direct_map_addr, poison, PAGE_SIZE);
7462 
7463 		free_reserved_page(page);
7464 	}
7465 
7466 	if (pages && s)
7467 		pr_info("Freeing %s memory: %ldK\n",
7468 			s, pages << (PAGE_SHIFT - 10));
7469 
7470 	return pages;
7471 }
7472 
7473 #ifdef	CONFIG_HIGHMEM
7474 void free_highmem_page(struct page *page)
7475 {
7476 	__free_reserved_page(page);
7477 	totalram_pages_inc();
7478 	atomic_long_inc(&page_zone(page)->managed_pages);
7479 	totalhigh_pages_inc();
7480 }
7481 #endif
7482 
7483 
7484 void __init mem_init_print_info(const char *str)
7485 {
7486 	unsigned long physpages, codesize, datasize, rosize, bss_size;
7487 	unsigned long init_code_size, init_data_size;
7488 
7489 	physpages = get_num_physpages();
7490 	codesize = _etext - _stext;
7491 	datasize = _edata - _sdata;
7492 	rosize = __end_rodata - __start_rodata;
7493 	bss_size = __bss_stop - __bss_start;
7494 	init_data_size = __init_end - __init_begin;
7495 	init_code_size = _einittext - _sinittext;
7496 
7497 	/*
7498 	 * Detect special cases and adjust section sizes accordingly:
7499 	 * 1) .init.* may be embedded into .data sections
7500 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
7501 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
7502 	 * 3) .rodata.* may be embedded into .text or .data sections.
7503 	 */
7504 #define adj_init_size(start, end, size, pos, adj) \
7505 	do { \
7506 		if (start <= pos && pos < end && size > adj) \
7507 			size -= adj; \
7508 	} while (0)
7509 
7510 	adj_init_size(__init_begin, __init_end, init_data_size,
7511 		     _sinittext, init_code_size);
7512 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7513 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7514 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7515 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7516 
7517 #undef	adj_init_size
7518 
7519 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7520 #ifdef	CONFIG_HIGHMEM
7521 		", %luK highmem"
7522 #endif
7523 		"%s%s)\n",
7524 		nr_free_pages() << (PAGE_SHIFT - 10),
7525 		physpages << (PAGE_SHIFT - 10),
7526 		codesize >> 10, datasize >> 10, rosize >> 10,
7527 		(init_data_size + init_code_size) >> 10, bss_size >> 10,
7528 		(physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
7529 		totalcma_pages << (PAGE_SHIFT - 10),
7530 #ifdef	CONFIG_HIGHMEM
7531 		totalhigh_pages() << (PAGE_SHIFT - 10),
7532 #endif
7533 		str ? ", " : "", str ? str : "");
7534 }
7535 
7536 /**
7537  * set_dma_reserve - set the specified number of pages reserved in the first zone
7538  * @new_dma_reserve: The number of pages to mark reserved
7539  *
7540  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7541  * In the DMA zone, a significant percentage may be consumed by kernel image
7542  * and other unfreeable allocations which can skew the watermarks badly. This
7543  * function may optionally be used to account for unfreeable pages in the
7544  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7545  * smaller per-cpu batchsize.
7546  */
7547 void __init set_dma_reserve(unsigned long new_dma_reserve)
7548 {
7549 	dma_reserve = new_dma_reserve;
7550 }
7551 
7552 void __init free_area_init(unsigned long *zones_size)
7553 {
7554 	zero_resv_unavail();
7555 	free_area_init_node(0, zones_size,
7556 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
7557 }
7558 
7559 static int page_alloc_cpu_dead(unsigned int cpu)
7560 {
7561 
7562 	lru_add_drain_cpu(cpu);
7563 	drain_pages(cpu);
7564 
7565 	/*
7566 	 * Spill the event counters of the dead processor
7567 	 * into the current processors event counters.
7568 	 * This artificially elevates the count of the current
7569 	 * processor.
7570 	 */
7571 	vm_events_fold_cpu(cpu);
7572 
7573 	/*
7574 	 * Zero the differential counters of the dead processor
7575 	 * so that the vm statistics are consistent.
7576 	 *
7577 	 * This is only okay since the processor is dead and cannot
7578 	 * race with what we are doing.
7579 	 */
7580 	cpu_vm_stats_fold(cpu);
7581 	return 0;
7582 }
7583 
7584 #ifdef CONFIG_NUMA
7585 int hashdist = HASHDIST_DEFAULT;
7586 
7587 static int __init set_hashdist(char *str)
7588 {
7589 	if (!str)
7590 		return 0;
7591 	hashdist = simple_strtoul(str, &str, 0);
7592 	return 1;
7593 }
7594 __setup("hashdist=", set_hashdist);
7595 #endif
7596 
7597 void __init page_alloc_init(void)
7598 {
7599 	int ret;
7600 
7601 #ifdef CONFIG_NUMA
7602 	if (num_node_state(N_MEMORY) == 1)
7603 		hashdist = 0;
7604 #endif
7605 
7606 	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7607 					"mm/page_alloc:dead", NULL,
7608 					page_alloc_cpu_dead);
7609 	WARN_ON(ret < 0);
7610 }
7611 
7612 /*
7613  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7614  *	or min_free_kbytes changes.
7615  */
7616 static void calculate_totalreserve_pages(void)
7617 {
7618 	struct pglist_data *pgdat;
7619 	unsigned long reserve_pages = 0;
7620 	enum zone_type i, j;
7621 
7622 	for_each_online_pgdat(pgdat) {
7623 
7624 		pgdat->totalreserve_pages = 0;
7625 
7626 		for (i = 0; i < MAX_NR_ZONES; i++) {
7627 			struct zone *zone = pgdat->node_zones + i;
7628 			long max = 0;
7629 			unsigned long managed_pages = zone_managed_pages(zone);
7630 
7631 			/* Find valid and maximum lowmem_reserve in the zone */
7632 			for (j = i; j < MAX_NR_ZONES; j++) {
7633 				if (zone->lowmem_reserve[j] > max)
7634 					max = zone->lowmem_reserve[j];
7635 			}
7636 
7637 			/* we treat the high watermark as reserved pages. */
7638 			max += high_wmark_pages(zone);
7639 
7640 			if (max > managed_pages)
7641 				max = managed_pages;
7642 
7643 			pgdat->totalreserve_pages += max;
7644 
7645 			reserve_pages += max;
7646 		}
7647 	}
7648 	totalreserve_pages = reserve_pages;
7649 }
7650 
7651 /*
7652  * setup_per_zone_lowmem_reserve - called whenever
7653  *	sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
7654  *	has a correct pages reserved value, so an adequate number of
7655  *	pages are left in the zone after a successful __alloc_pages().
7656  */
7657 static void setup_per_zone_lowmem_reserve(void)
7658 {
7659 	struct pglist_data *pgdat;
7660 	enum zone_type j, idx;
7661 
7662 	for_each_online_pgdat(pgdat) {
7663 		for (j = 0; j < MAX_NR_ZONES; j++) {
7664 			struct zone *zone = pgdat->node_zones + j;
7665 			unsigned long managed_pages = zone_managed_pages(zone);
7666 
7667 			zone->lowmem_reserve[j] = 0;
7668 
7669 			idx = j;
7670 			while (idx) {
7671 				struct zone *lower_zone;
7672 
7673 				idx--;
7674 				lower_zone = pgdat->node_zones + idx;
7675 
7676 				if (sysctl_lowmem_reserve_ratio[idx] < 1) {
7677 					sysctl_lowmem_reserve_ratio[idx] = 0;
7678 					lower_zone->lowmem_reserve[j] = 0;
7679 				} else {
7680 					lower_zone->lowmem_reserve[j] =
7681 						managed_pages / sysctl_lowmem_reserve_ratio[idx];
7682 				}
7683 				managed_pages += zone_managed_pages(lower_zone);
7684 			}
7685 		}
7686 	}
7687 
7688 	/* update totalreserve_pages */
7689 	calculate_totalreserve_pages();
7690 }
7691 
7692 static void __setup_per_zone_wmarks(void)
7693 {
7694 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7695 	unsigned long lowmem_pages = 0;
7696 	struct zone *zone;
7697 	unsigned long flags;
7698 
7699 	/* Calculate total number of !ZONE_HIGHMEM pages */
7700 	for_each_zone(zone) {
7701 		if (!is_highmem(zone))
7702 			lowmem_pages += zone_managed_pages(zone);
7703 	}
7704 
7705 	for_each_zone(zone) {
7706 		u64 tmp;
7707 
7708 		spin_lock_irqsave(&zone->lock, flags);
7709 		tmp = (u64)pages_min * zone_managed_pages(zone);
7710 		do_div(tmp, lowmem_pages);
7711 		if (is_highmem(zone)) {
7712 			/*
7713 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7714 			 * need highmem pages, so cap pages_min to a small
7715 			 * value here.
7716 			 *
7717 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7718 			 * deltas control async page reclaim, and so should
7719 			 * not be capped for highmem.
7720 			 */
7721 			unsigned long min_pages;
7722 
7723 			min_pages = zone_managed_pages(zone) / 1024;
7724 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7725 			zone->_watermark[WMARK_MIN] = min_pages;
7726 		} else {
7727 			/*
7728 			 * If it's a lowmem zone, reserve a number of pages
7729 			 * proportionate to the zone's size.
7730 			 */
7731 			zone->_watermark[WMARK_MIN] = tmp;
7732 		}
7733 
7734 		/*
7735 		 * Set the kswapd watermarks distance according to the
7736 		 * scale factor in proportion to available memory, but
7737 		 * ensure a minimum size on small systems.
7738 		 */
7739 		tmp = max_t(u64, tmp >> 2,
7740 			    mult_frac(zone_managed_pages(zone),
7741 				      watermark_scale_factor, 10000));
7742 
7743 		zone->_watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
7744 		zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7745 		zone->watermark_boost = 0;
7746 
7747 		spin_unlock_irqrestore(&zone->lock, flags);
7748 	}
7749 
7750 	/* update totalreserve_pages */
7751 	calculate_totalreserve_pages();
7752 }
7753 
7754 /**
7755  * setup_per_zone_wmarks - called when min_free_kbytes changes
7756  * or when memory is hot-{added|removed}
7757  *
7758  * Ensures that the watermark[min,low,high] values for each zone are set
7759  * correctly with respect to min_free_kbytes.
7760  */
7761 void setup_per_zone_wmarks(void)
7762 {
7763 	static DEFINE_SPINLOCK(lock);
7764 
7765 	spin_lock(&lock);
7766 	__setup_per_zone_wmarks();
7767 	spin_unlock(&lock);
7768 }
7769 
7770 /*
7771  * Initialise min_free_kbytes.
7772  *
7773  * For small machines we want it small (128k min).  For large machines
7774  * we want it large (64MB max).  But it is not linear, because network
7775  * bandwidth does not increase linearly with machine size.  We use
7776  *
7777  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7778  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
7779  *
7780  * which yields
7781  *
7782  * 16MB:	512k
7783  * 32MB:	724k
7784  * 64MB:	1024k
7785  * 128MB:	1448k
7786  * 256MB:	2048k
7787  * 512MB:	2896k
7788  * 1024MB:	4096k
7789  * 2048MB:	5792k
7790  * 4096MB:	8192k
7791  * 8192MB:	11584k
7792  * 16384MB:	16384k
7793  */
7794 int __meminit init_per_zone_wmark_min(void)
7795 {
7796 	unsigned long lowmem_kbytes;
7797 	int new_min_free_kbytes;
7798 
7799 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7800 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7801 
7802 	if (new_min_free_kbytes > user_min_free_kbytes) {
7803 		min_free_kbytes = new_min_free_kbytes;
7804 		if (min_free_kbytes < 128)
7805 			min_free_kbytes = 128;
7806 		if (min_free_kbytes > 65536)
7807 			min_free_kbytes = 65536;
7808 	} else {
7809 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7810 				new_min_free_kbytes, user_min_free_kbytes);
7811 	}
7812 	setup_per_zone_wmarks();
7813 	refresh_zone_stat_thresholds();
7814 	setup_per_zone_lowmem_reserve();
7815 
7816 #ifdef CONFIG_NUMA
7817 	setup_min_unmapped_ratio();
7818 	setup_min_slab_ratio();
7819 #endif
7820 
7821 	return 0;
7822 }
7823 core_initcall(init_per_zone_wmark_min)
7824 
7825 /*
7826  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
7827  *	that we can call two helper functions whenever min_free_kbytes
7828  *	changes.
7829  */
7830 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7831 	void __user *buffer, size_t *length, loff_t *ppos)
7832 {
7833 	int rc;
7834 
7835 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7836 	if (rc)
7837 		return rc;
7838 
7839 	if (write) {
7840 		user_min_free_kbytes = min_free_kbytes;
7841 		setup_per_zone_wmarks();
7842 	}
7843 	return 0;
7844 }
7845 
7846 int watermark_boost_factor_sysctl_handler(struct ctl_table *table, int write,
7847 	void __user *buffer, size_t *length, loff_t *ppos)
7848 {
7849 	int rc;
7850 
7851 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7852 	if (rc)
7853 		return rc;
7854 
7855 	return 0;
7856 }
7857 
7858 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7859 	void __user *buffer, size_t *length, loff_t *ppos)
7860 {
7861 	int rc;
7862 
7863 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7864 	if (rc)
7865 		return rc;
7866 
7867 	if (write)
7868 		setup_per_zone_wmarks();
7869 
7870 	return 0;
7871 }
7872 
7873 #ifdef CONFIG_NUMA
7874 static void setup_min_unmapped_ratio(void)
7875 {
7876 	pg_data_t *pgdat;
7877 	struct zone *zone;
7878 
7879 	for_each_online_pgdat(pgdat)
7880 		pgdat->min_unmapped_pages = 0;
7881 
7882 	for_each_zone(zone)
7883 		zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
7884 						         sysctl_min_unmapped_ratio) / 100;
7885 }
7886 
7887 
7888 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7889 	void __user *buffer, size_t *length, loff_t *ppos)
7890 {
7891 	int rc;
7892 
7893 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7894 	if (rc)
7895 		return rc;
7896 
7897 	setup_min_unmapped_ratio();
7898 
7899 	return 0;
7900 }
7901 
7902 static void setup_min_slab_ratio(void)
7903 {
7904 	pg_data_t *pgdat;
7905 	struct zone *zone;
7906 
7907 	for_each_online_pgdat(pgdat)
7908 		pgdat->min_slab_pages = 0;
7909 
7910 	for_each_zone(zone)
7911 		zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
7912 						     sysctl_min_slab_ratio) / 100;
7913 }
7914 
7915 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7916 	void __user *buffer, size_t *length, loff_t *ppos)
7917 {
7918 	int rc;
7919 
7920 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7921 	if (rc)
7922 		return rc;
7923 
7924 	setup_min_slab_ratio();
7925 
7926 	return 0;
7927 }
7928 #endif
7929 
7930 /*
7931  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7932  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7933  *	whenever sysctl_lowmem_reserve_ratio changes.
7934  *
7935  * The reserve ratio obviously has absolutely no relation with the
7936  * minimum watermarks. The lowmem reserve ratio can only make sense
7937  * if in function of the boot time zone sizes.
7938  */
7939 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
7940 	void __user *buffer, size_t *length, loff_t *ppos)
7941 {
7942 	proc_dointvec_minmax(table, write, buffer, length, ppos);
7943 	setup_per_zone_lowmem_reserve();
7944 	return 0;
7945 }
7946 
7947 /*
7948  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7949  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
7950  * pagelist can have before it gets flushed back to buddy allocator.
7951  */
7952 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
7953 	void __user *buffer, size_t *length, loff_t *ppos)
7954 {
7955 	struct zone *zone;
7956 	int old_percpu_pagelist_fraction;
7957 	int ret;
7958 
7959 	mutex_lock(&pcp_batch_high_lock);
7960 	old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7961 
7962 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7963 	if (!write || ret < 0)
7964 		goto out;
7965 
7966 	/* Sanity checking to avoid pcp imbalance */
7967 	if (percpu_pagelist_fraction &&
7968 	    percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7969 		percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7970 		ret = -EINVAL;
7971 		goto out;
7972 	}
7973 
7974 	/* No change? */
7975 	if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7976 		goto out;
7977 
7978 	for_each_populated_zone(zone) {
7979 		unsigned int cpu;
7980 
7981 		for_each_possible_cpu(cpu)
7982 			pageset_set_high_and_batch(zone,
7983 					per_cpu_ptr(zone->pageset, cpu));
7984 	}
7985 out:
7986 	mutex_unlock(&pcp_batch_high_lock);
7987 	return ret;
7988 }
7989 
7990 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7991 /*
7992  * Returns the number of pages that arch has reserved but
7993  * is not known to alloc_large_system_hash().
7994  */
7995 static unsigned long __init arch_reserved_kernel_pages(void)
7996 {
7997 	return 0;
7998 }
7999 #endif
8000 
8001 /*
8002  * Adaptive scale is meant to reduce sizes of hash tables on large memory
8003  * machines. As memory size is increased the scale is also increased but at
8004  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
8005  * quadruples the scale is increased by one, which means the size of hash table
8006  * only doubles, instead of quadrupling as well.
8007  * Because 32-bit systems cannot have large physical memory, where this scaling
8008  * makes sense, it is disabled on such platforms.
8009  */
8010 #if __BITS_PER_LONG > 32
8011 #define ADAPT_SCALE_BASE	(64ul << 30)
8012 #define ADAPT_SCALE_SHIFT	2
8013 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
8014 #endif
8015 
8016 /*
8017  * allocate a large system hash table from bootmem
8018  * - it is assumed that the hash table must contain an exact power-of-2
8019  *   quantity of entries
8020  * - limit is the number of hash buckets, not the total allocation size
8021  */
8022 void *__init alloc_large_system_hash(const char *tablename,
8023 				     unsigned long bucketsize,
8024 				     unsigned long numentries,
8025 				     int scale,
8026 				     int flags,
8027 				     unsigned int *_hash_shift,
8028 				     unsigned int *_hash_mask,
8029 				     unsigned long low_limit,
8030 				     unsigned long high_limit)
8031 {
8032 	unsigned long long max = high_limit;
8033 	unsigned long log2qty, size;
8034 	void *table = NULL;
8035 	gfp_t gfp_flags;
8036 	bool virt;
8037 
8038 	/* allow the kernel cmdline to have a say */
8039 	if (!numentries) {
8040 		/* round applicable memory size up to nearest megabyte */
8041 		numentries = nr_kernel_pages;
8042 		numentries -= arch_reserved_kernel_pages();
8043 
8044 		/* It isn't necessary when PAGE_SIZE >= 1MB */
8045 		if (PAGE_SHIFT < 20)
8046 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8047 
8048 #if __BITS_PER_LONG > 32
8049 		if (!high_limit) {
8050 			unsigned long adapt;
8051 
8052 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8053 			     adapt <<= ADAPT_SCALE_SHIFT)
8054 				scale++;
8055 		}
8056 #endif
8057 
8058 		/* limit to 1 bucket per 2^scale bytes of low memory */
8059 		if (scale > PAGE_SHIFT)
8060 			numentries >>= (scale - PAGE_SHIFT);
8061 		else
8062 			numentries <<= (PAGE_SHIFT - scale);
8063 
8064 		/* Make sure we've got at least a 0-order allocation.. */
8065 		if (unlikely(flags & HASH_SMALL)) {
8066 			/* Makes no sense without HASH_EARLY */
8067 			WARN_ON(!(flags & HASH_EARLY));
8068 			if (!(numentries >> *_hash_shift)) {
8069 				numentries = 1UL << *_hash_shift;
8070 				BUG_ON(!numentries);
8071 			}
8072 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8073 			numentries = PAGE_SIZE / bucketsize;
8074 	}
8075 	numentries = roundup_pow_of_two(numentries);
8076 
8077 	/* limit allocation size to 1/16 total memory by default */
8078 	if (max == 0) {
8079 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8080 		do_div(max, bucketsize);
8081 	}
8082 	max = min(max, 0x80000000ULL);
8083 
8084 	if (numentries < low_limit)
8085 		numentries = low_limit;
8086 	if (numentries > max)
8087 		numentries = max;
8088 
8089 	log2qty = ilog2(numentries);
8090 
8091 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8092 	do {
8093 		virt = false;
8094 		size = bucketsize << log2qty;
8095 		if (flags & HASH_EARLY) {
8096 			if (flags & HASH_ZERO)
8097 				table = memblock_alloc(size, SMP_CACHE_BYTES);
8098 			else
8099 				table = memblock_alloc_raw(size,
8100 							   SMP_CACHE_BYTES);
8101 		} else if (get_order(size) >= MAX_ORDER || hashdist) {
8102 			table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
8103 			virt = true;
8104 		} else {
8105 			/*
8106 			 * If bucketsize is not a power-of-two, we may free
8107 			 * some pages at the end of hash table which
8108 			 * alloc_pages_exact() automatically does
8109 			 */
8110 			table = alloc_pages_exact(size, gfp_flags);
8111 			kmemleak_alloc(table, size, 1, gfp_flags);
8112 		}
8113 	} while (!table && size > PAGE_SIZE && --log2qty);
8114 
8115 	if (!table)
8116 		panic("Failed to allocate %s hash table\n", tablename);
8117 
8118 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8119 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8120 		virt ? "vmalloc" : "linear");
8121 
8122 	if (_hash_shift)
8123 		*_hash_shift = log2qty;
8124 	if (_hash_mask)
8125 		*_hash_mask = (1 << log2qty) - 1;
8126 
8127 	return table;
8128 }
8129 
8130 /*
8131  * This function checks whether pageblock includes unmovable pages or not.
8132  * If @count is not zero, it is okay to include less @count unmovable pages
8133  *
8134  * PageLRU check without isolation or lru_lock could race so that
8135  * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
8136  * check without lock_page also may miss some movable non-lru pages at
8137  * race condition. So you can't expect this function should be exact.
8138  */
8139 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
8140 			 int migratetype, int flags)
8141 {
8142 	unsigned long found;
8143 	unsigned long iter = 0;
8144 	unsigned long pfn = page_to_pfn(page);
8145 	const char *reason = "unmovable page";
8146 
8147 	/*
8148 	 * TODO we could make this much more efficient by not checking every
8149 	 * page in the range if we know all of them are in MOVABLE_ZONE and
8150 	 * that the movable zone guarantees that pages are migratable but
8151 	 * the later is not the case right now unfortunatelly. E.g. movablecore
8152 	 * can still lead to having bootmem allocations in zone_movable.
8153 	 */
8154 
8155 	if (is_migrate_cma_page(page)) {
8156 		/*
8157 		 * CMA allocations (alloc_contig_range) really need to mark
8158 		 * isolate CMA pageblocks even when they are not movable in fact
8159 		 * so consider them movable here.
8160 		 */
8161 		if (is_migrate_cma(migratetype))
8162 			return false;
8163 
8164 		reason = "CMA page";
8165 		goto unmovable;
8166 	}
8167 
8168 	for (found = 0; iter < pageblock_nr_pages; iter++) {
8169 		unsigned long check = pfn + iter;
8170 
8171 		if (!pfn_valid_within(check))
8172 			continue;
8173 
8174 		page = pfn_to_page(check);
8175 
8176 		if (PageReserved(page))
8177 			goto unmovable;
8178 
8179 		/*
8180 		 * If the zone is movable and we have ruled out all reserved
8181 		 * pages then it should be reasonably safe to assume the rest
8182 		 * is movable.
8183 		 */
8184 		if (zone_idx(zone) == ZONE_MOVABLE)
8185 			continue;
8186 
8187 		/*
8188 		 * Hugepages are not in LRU lists, but they're movable.
8189 		 * We need not scan over tail pages because we don't
8190 		 * handle each tail page individually in migration.
8191 		 */
8192 		if (PageHuge(page)) {
8193 			struct page *head = compound_head(page);
8194 			unsigned int skip_pages;
8195 
8196 			if (!hugepage_migration_supported(page_hstate(head)))
8197 				goto unmovable;
8198 
8199 			skip_pages = (1 << compound_order(head)) - (page - head);
8200 			iter += skip_pages - 1;
8201 			continue;
8202 		}
8203 
8204 		/*
8205 		 * We can't use page_count without pin a page
8206 		 * because another CPU can free compound page.
8207 		 * This check already skips compound tails of THP
8208 		 * because their page->_refcount is zero at all time.
8209 		 */
8210 		if (!page_ref_count(page)) {
8211 			if (PageBuddy(page))
8212 				iter += (1 << page_order(page)) - 1;
8213 			continue;
8214 		}
8215 
8216 		/*
8217 		 * The HWPoisoned page may be not in buddy system, and
8218 		 * page_count() is not 0.
8219 		 */
8220 		if ((flags & SKIP_HWPOISON) && PageHWPoison(page))
8221 			continue;
8222 
8223 		if (__PageMovable(page))
8224 			continue;
8225 
8226 		if (!PageLRU(page))
8227 			found++;
8228 		/*
8229 		 * If there are RECLAIMABLE pages, we need to check
8230 		 * it.  But now, memory offline itself doesn't call
8231 		 * shrink_node_slabs() and it still to be fixed.
8232 		 */
8233 		/*
8234 		 * If the page is not RAM, page_count()should be 0.
8235 		 * we don't need more check. This is an _used_ not-movable page.
8236 		 *
8237 		 * The problematic thing here is PG_reserved pages. PG_reserved
8238 		 * is set to both of a memory hole page and a _used_ kernel
8239 		 * page at boot.
8240 		 */
8241 		if (found > count)
8242 			goto unmovable;
8243 	}
8244 	return false;
8245 unmovable:
8246 	WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
8247 	if (flags & REPORT_FAILURE)
8248 		dump_page(pfn_to_page(pfn + iter), reason);
8249 	return true;
8250 }
8251 
8252 #ifdef CONFIG_CONTIG_ALLOC
8253 static unsigned long pfn_max_align_down(unsigned long pfn)
8254 {
8255 	return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8256 			     pageblock_nr_pages) - 1);
8257 }
8258 
8259 static unsigned long pfn_max_align_up(unsigned long pfn)
8260 {
8261 	return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8262 				pageblock_nr_pages));
8263 }
8264 
8265 /* [start, end) must belong to a single zone. */
8266 static int __alloc_contig_migrate_range(struct compact_control *cc,
8267 					unsigned long start, unsigned long end)
8268 {
8269 	/* This function is based on compact_zone() from compaction.c. */
8270 	unsigned long nr_reclaimed;
8271 	unsigned long pfn = start;
8272 	unsigned int tries = 0;
8273 	int ret = 0;
8274 
8275 	migrate_prep();
8276 
8277 	while (pfn < end || !list_empty(&cc->migratepages)) {
8278 		if (fatal_signal_pending(current)) {
8279 			ret = -EINTR;
8280 			break;
8281 		}
8282 
8283 		if (list_empty(&cc->migratepages)) {
8284 			cc->nr_migratepages = 0;
8285 			pfn = isolate_migratepages_range(cc, pfn, end);
8286 			if (!pfn) {
8287 				ret = -EINTR;
8288 				break;
8289 			}
8290 			tries = 0;
8291 		} else if (++tries == 5) {
8292 			ret = ret < 0 ? ret : -EBUSY;
8293 			break;
8294 		}
8295 
8296 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8297 							&cc->migratepages);
8298 		cc->nr_migratepages -= nr_reclaimed;
8299 
8300 		ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
8301 				    NULL, 0, cc->mode, MR_CONTIG_RANGE);
8302 	}
8303 	if (ret < 0) {
8304 		putback_movable_pages(&cc->migratepages);
8305 		return ret;
8306 	}
8307 	return 0;
8308 }
8309 
8310 /**
8311  * alloc_contig_range() -- tries to allocate given range of pages
8312  * @start:	start PFN to allocate
8313  * @end:	one-past-the-last PFN to allocate
8314  * @migratetype:	migratetype of the underlaying pageblocks (either
8315  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
8316  *			in range must have the same migratetype and it must
8317  *			be either of the two.
8318  * @gfp_mask:	GFP mask to use during compaction
8319  *
8320  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
8321  * aligned.  The PFN range must belong to a single zone.
8322  *
8323  * The first thing this routine does is attempt to MIGRATE_ISOLATE all
8324  * pageblocks in the range.  Once isolated, the pageblocks should not
8325  * be modified by others.
8326  *
8327  * Return: zero on success or negative error code.  On success all
8328  * pages which PFN is in [start, end) are allocated for the caller and
8329  * need to be freed with free_contig_range().
8330  */
8331 int alloc_contig_range(unsigned long start, unsigned long end,
8332 		       unsigned migratetype, gfp_t gfp_mask)
8333 {
8334 	unsigned long outer_start, outer_end;
8335 	unsigned int order;
8336 	int ret = 0;
8337 
8338 	struct compact_control cc = {
8339 		.nr_migratepages = 0,
8340 		.order = -1,
8341 		.zone = page_zone(pfn_to_page(start)),
8342 		.mode = MIGRATE_SYNC,
8343 		.ignore_skip_hint = true,
8344 		.no_set_skip_hint = true,
8345 		.gfp_mask = current_gfp_context(gfp_mask),
8346 	};
8347 	INIT_LIST_HEAD(&cc.migratepages);
8348 
8349 	/*
8350 	 * What we do here is we mark all pageblocks in range as
8351 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
8352 	 * have different sizes, and due to the way page allocator
8353 	 * work, we align the range to biggest of the two pages so
8354 	 * that page allocator won't try to merge buddies from
8355 	 * different pageblocks and change MIGRATE_ISOLATE to some
8356 	 * other migration type.
8357 	 *
8358 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
8359 	 * migrate the pages from an unaligned range (ie. pages that
8360 	 * we are interested in).  This will put all the pages in
8361 	 * range back to page allocator as MIGRATE_ISOLATE.
8362 	 *
8363 	 * When this is done, we take the pages in range from page
8364 	 * allocator removing them from the buddy system.  This way
8365 	 * page allocator will never consider using them.
8366 	 *
8367 	 * This lets us mark the pageblocks back as
8368 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
8369 	 * aligned range but not in the unaligned, original range are
8370 	 * put back to page allocator so that buddy can use them.
8371 	 */
8372 
8373 	ret = start_isolate_page_range(pfn_max_align_down(start),
8374 				       pfn_max_align_up(end), migratetype, 0);
8375 	if (ret < 0)
8376 		return ret;
8377 
8378 	/*
8379 	 * In case of -EBUSY, we'd like to know which page causes problem.
8380 	 * So, just fall through. test_pages_isolated() has a tracepoint
8381 	 * which will report the busy page.
8382 	 *
8383 	 * It is possible that busy pages could become available before
8384 	 * the call to test_pages_isolated, and the range will actually be
8385 	 * allocated.  So, if we fall through be sure to clear ret so that
8386 	 * -EBUSY is not accidentally used or returned to caller.
8387 	 */
8388 	ret = __alloc_contig_migrate_range(&cc, start, end);
8389 	if (ret && ret != -EBUSY)
8390 		goto done;
8391 	ret =0;
8392 
8393 	/*
8394 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
8395 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
8396 	 * more, all pages in [start, end) are free in page allocator.
8397 	 * What we are going to do is to allocate all pages from
8398 	 * [start, end) (that is remove them from page allocator).
8399 	 *
8400 	 * The only problem is that pages at the beginning and at the
8401 	 * end of interesting range may be not aligned with pages that
8402 	 * page allocator holds, ie. they can be part of higher order
8403 	 * pages.  Because of this, we reserve the bigger range and
8404 	 * once this is done free the pages we are not interested in.
8405 	 *
8406 	 * We don't have to hold zone->lock here because the pages are
8407 	 * isolated thus they won't get removed from buddy.
8408 	 */
8409 
8410 	lru_add_drain_all();
8411 
8412 	order = 0;
8413 	outer_start = start;
8414 	while (!PageBuddy(pfn_to_page(outer_start))) {
8415 		if (++order >= MAX_ORDER) {
8416 			outer_start = start;
8417 			break;
8418 		}
8419 		outer_start &= ~0UL << order;
8420 	}
8421 
8422 	if (outer_start != start) {
8423 		order = page_order(pfn_to_page(outer_start));
8424 
8425 		/*
8426 		 * outer_start page could be small order buddy page and
8427 		 * it doesn't include start page. Adjust outer_start
8428 		 * in this case to report failed page properly
8429 		 * on tracepoint in test_pages_isolated()
8430 		 */
8431 		if (outer_start + (1UL << order) <= start)
8432 			outer_start = start;
8433 	}
8434 
8435 	/* Make sure the range is really isolated. */
8436 	if (test_pages_isolated(outer_start, end, false)) {
8437 		pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8438 			__func__, outer_start, end);
8439 		ret = -EBUSY;
8440 		goto done;
8441 	}
8442 
8443 	/* Grab isolated pages from freelists. */
8444 	outer_end = isolate_freepages_range(&cc, outer_start, end);
8445 	if (!outer_end) {
8446 		ret = -EBUSY;
8447 		goto done;
8448 	}
8449 
8450 	/* Free head and tail (if any) */
8451 	if (start != outer_start)
8452 		free_contig_range(outer_start, start - outer_start);
8453 	if (end != outer_end)
8454 		free_contig_range(end, outer_end - end);
8455 
8456 done:
8457 	undo_isolate_page_range(pfn_max_align_down(start),
8458 				pfn_max_align_up(end), migratetype);
8459 	return ret;
8460 }
8461 #endif /* CONFIG_CONTIG_ALLOC */
8462 
8463 void free_contig_range(unsigned long pfn, unsigned int nr_pages)
8464 {
8465 	unsigned int count = 0;
8466 
8467 	for (; nr_pages--; pfn++) {
8468 		struct page *page = pfn_to_page(pfn);
8469 
8470 		count += page_count(page) != 1;
8471 		__free_page(page);
8472 	}
8473 	WARN(count != 0, "%d pages are still in use!\n", count);
8474 }
8475 
8476 #ifdef CONFIG_MEMORY_HOTPLUG
8477 /*
8478  * The zone indicated has a new number of managed_pages; batch sizes and percpu
8479  * page high values need to be recalulated.
8480  */
8481 void __meminit zone_pcp_update(struct zone *zone)
8482 {
8483 	unsigned cpu;
8484 	mutex_lock(&pcp_batch_high_lock);
8485 	for_each_possible_cpu(cpu)
8486 		pageset_set_high_and_batch(zone,
8487 				per_cpu_ptr(zone->pageset, cpu));
8488 	mutex_unlock(&pcp_batch_high_lock);
8489 }
8490 #endif
8491 
8492 void zone_pcp_reset(struct zone *zone)
8493 {
8494 	unsigned long flags;
8495 	int cpu;
8496 	struct per_cpu_pageset *pset;
8497 
8498 	/* avoid races with drain_pages()  */
8499 	local_irq_save(flags);
8500 	if (zone->pageset != &boot_pageset) {
8501 		for_each_online_cpu(cpu) {
8502 			pset = per_cpu_ptr(zone->pageset, cpu);
8503 			drain_zonestat(zone, pset);
8504 		}
8505 		free_percpu(zone->pageset);
8506 		zone->pageset = &boot_pageset;
8507 	}
8508 	local_irq_restore(flags);
8509 }
8510 
8511 #ifdef CONFIG_MEMORY_HOTREMOVE
8512 /*
8513  * All pages in the range must be in a single zone and isolated
8514  * before calling this.
8515  */
8516 unsigned long
8517 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8518 {
8519 	struct page *page;
8520 	struct zone *zone;
8521 	unsigned int order, i;
8522 	unsigned long pfn;
8523 	unsigned long flags;
8524 	unsigned long offlined_pages = 0;
8525 
8526 	/* find the first valid pfn */
8527 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
8528 		if (pfn_valid(pfn))
8529 			break;
8530 	if (pfn == end_pfn)
8531 		return offlined_pages;
8532 
8533 	offline_mem_sections(pfn, end_pfn);
8534 	zone = page_zone(pfn_to_page(pfn));
8535 	spin_lock_irqsave(&zone->lock, flags);
8536 	pfn = start_pfn;
8537 	while (pfn < end_pfn) {
8538 		if (!pfn_valid(pfn)) {
8539 			pfn++;
8540 			continue;
8541 		}
8542 		page = pfn_to_page(pfn);
8543 		/*
8544 		 * The HWPoisoned page may be not in buddy system, and
8545 		 * page_count() is not 0.
8546 		 */
8547 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8548 			pfn++;
8549 			SetPageReserved(page);
8550 			offlined_pages++;
8551 			continue;
8552 		}
8553 
8554 		BUG_ON(page_count(page));
8555 		BUG_ON(!PageBuddy(page));
8556 		order = page_order(page);
8557 		offlined_pages += 1 << order;
8558 #ifdef CONFIG_DEBUG_VM
8559 		pr_info("remove from free list %lx %d %lx\n",
8560 			pfn, 1 << order, end_pfn);
8561 #endif
8562 		del_page_from_free_area(page, &zone->free_area[order]);
8563 		for (i = 0; i < (1 << order); i++)
8564 			SetPageReserved((page+i));
8565 		pfn += (1 << order);
8566 	}
8567 	spin_unlock_irqrestore(&zone->lock, flags);
8568 
8569 	return offlined_pages;
8570 }
8571 #endif
8572 
8573 bool is_free_buddy_page(struct page *page)
8574 {
8575 	struct zone *zone = page_zone(page);
8576 	unsigned long pfn = page_to_pfn(page);
8577 	unsigned long flags;
8578 	unsigned int order;
8579 
8580 	spin_lock_irqsave(&zone->lock, flags);
8581 	for (order = 0; order < MAX_ORDER; order++) {
8582 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8583 
8584 		if (PageBuddy(page_head) && page_order(page_head) >= order)
8585 			break;
8586 	}
8587 	spin_unlock_irqrestore(&zone->lock, flags);
8588 
8589 	return order < MAX_ORDER;
8590 }
8591 
8592 #ifdef CONFIG_MEMORY_FAILURE
8593 /*
8594  * Set PG_hwpoison flag if a given page is confirmed to be a free page.  This
8595  * test is performed under the zone lock to prevent a race against page
8596  * allocation.
8597  */
8598 bool set_hwpoison_free_buddy_page(struct page *page)
8599 {
8600 	struct zone *zone = page_zone(page);
8601 	unsigned long pfn = page_to_pfn(page);
8602 	unsigned long flags;
8603 	unsigned int order;
8604 	bool hwpoisoned = false;
8605 
8606 	spin_lock_irqsave(&zone->lock, flags);
8607 	for (order = 0; order < MAX_ORDER; order++) {
8608 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8609 
8610 		if (PageBuddy(page_head) && page_order(page_head) >= order) {
8611 			if (!TestSetPageHWPoison(page))
8612 				hwpoisoned = true;
8613 			break;
8614 		}
8615 	}
8616 	spin_unlock_irqrestore(&zone->lock, flags);
8617 
8618 	return hwpoisoned;
8619 }
8620 #endif
8621