xref: /openbmc/linux/mm/page_alloc.c (revision ff4a7481c3898ffc3cc271d6aca431d190c37247)
1 /*
2  *  linux/mm/page_alloc.c
3  *
4  *  Manages the free list, the system allocates free pages here.
5  *  Note that kmalloc() lives in slab.c
6  *
7  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
8  *  Swap reorganised 29.12.95, Stephen Tweedie
9  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15  */
16 
17 #include <linux/stddef.h>
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/memblock.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/kasan.h>
27 #include <linux/module.h>
28 #include <linux/suspend.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/slab.h>
32 #include <linux/ratelimit.h>
33 #include <linux/oom.h>
34 #include <linux/topology.h>
35 #include <linux/sysctl.h>
36 #include <linux/cpu.h>
37 #include <linux/cpuset.h>
38 #include <linux/memory_hotplug.h>
39 #include <linux/nodemask.h>
40 #include <linux/vmalloc.h>
41 #include <linux/vmstat.h>
42 #include <linux/mempolicy.h>
43 #include <linux/memremap.h>
44 #include <linux/stop_machine.h>
45 #include <linux/sort.h>
46 #include <linux/pfn.h>
47 #include <linux/backing-dev.h>
48 #include <linux/fault-inject.h>
49 #include <linux/page-isolation.h>
50 #include <linux/page_ext.h>
51 #include <linux/debugobjects.h>
52 #include <linux/kmemleak.h>
53 #include <linux/compaction.h>
54 #include <trace/events/kmem.h>
55 #include <trace/events/oom.h>
56 #include <linux/prefetch.h>
57 #include <linux/mm_inline.h>
58 #include <linux/migrate.h>
59 #include <linux/hugetlb.h>
60 #include <linux/sched/rt.h>
61 #include <linux/sched/mm.h>
62 #include <linux/page_owner.h>
63 #include <linux/kthread.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/lockdep.h>
67 #include <linux/nmi.h>
68 #include <linux/psi.h>
69 
70 #include <asm/sections.h>
71 #include <asm/tlbflush.h>
72 #include <asm/div64.h>
73 #include "internal.h"
74 
75 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
76 static DEFINE_MUTEX(pcp_batch_high_lock);
77 #define MIN_PERCPU_PAGELIST_FRACTION	(8)
78 
79 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
80 DEFINE_PER_CPU(int, numa_node);
81 EXPORT_PER_CPU_SYMBOL(numa_node);
82 #endif
83 
84 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
85 
86 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
87 /*
88  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
89  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
90  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
91  * defined in <linux/topology.h>.
92  */
93 DEFINE_PER_CPU(int, _numa_mem_);		/* Kernel "local memory" node */
94 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
95 int _node_numa_mem_[MAX_NUMNODES];
96 #endif
97 
98 /* work_structs for global per-cpu drains */
99 DEFINE_MUTEX(pcpu_drain_mutex);
100 DEFINE_PER_CPU(struct work_struct, pcpu_drain);
101 
102 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
103 volatile unsigned long latent_entropy __latent_entropy;
104 EXPORT_SYMBOL(latent_entropy);
105 #endif
106 
107 /*
108  * Array of node states.
109  */
110 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
111 	[N_POSSIBLE] = NODE_MASK_ALL,
112 	[N_ONLINE] = { { [0] = 1UL } },
113 #ifndef CONFIG_NUMA
114 	[N_NORMAL_MEMORY] = { { [0] = 1UL } },
115 #ifdef CONFIG_HIGHMEM
116 	[N_HIGH_MEMORY] = { { [0] = 1UL } },
117 #endif
118 	[N_MEMORY] = { { [0] = 1UL } },
119 	[N_CPU] = { { [0] = 1UL } },
120 #endif	/* NUMA */
121 };
122 EXPORT_SYMBOL(node_states);
123 
124 /* Protect totalram_pages and zone->managed_pages */
125 static DEFINE_SPINLOCK(managed_page_count_lock);
126 
127 unsigned long totalram_pages __read_mostly;
128 unsigned long totalreserve_pages __read_mostly;
129 unsigned long totalcma_pages __read_mostly;
130 
131 int percpu_pagelist_fraction;
132 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
133 
134 /*
135  * A cached value of the page's pageblock's migratetype, used when the page is
136  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
137  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
138  * Also the migratetype set in the page does not necessarily match the pcplist
139  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
140  * other index - this ensures that it will be put on the correct CMA freelist.
141  */
142 static inline int get_pcppage_migratetype(struct page *page)
143 {
144 	return page->index;
145 }
146 
147 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
148 {
149 	page->index = migratetype;
150 }
151 
152 #ifdef CONFIG_PM_SLEEP
153 /*
154  * The following functions are used by the suspend/hibernate code to temporarily
155  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
156  * while devices are suspended.  To avoid races with the suspend/hibernate code,
157  * they should always be called with system_transition_mutex held
158  * (gfp_allowed_mask also should only be modified with system_transition_mutex
159  * held, unless the suspend/hibernate code is guaranteed not to run in parallel
160  * with that modification).
161  */
162 
163 static gfp_t saved_gfp_mask;
164 
165 void pm_restore_gfp_mask(void)
166 {
167 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
168 	if (saved_gfp_mask) {
169 		gfp_allowed_mask = saved_gfp_mask;
170 		saved_gfp_mask = 0;
171 	}
172 }
173 
174 void pm_restrict_gfp_mask(void)
175 {
176 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
177 	WARN_ON(saved_gfp_mask);
178 	saved_gfp_mask = gfp_allowed_mask;
179 	gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
180 }
181 
182 bool pm_suspended_storage(void)
183 {
184 	if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
185 		return false;
186 	return true;
187 }
188 #endif /* CONFIG_PM_SLEEP */
189 
190 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
191 unsigned int pageblock_order __read_mostly;
192 #endif
193 
194 static void __free_pages_ok(struct page *page, unsigned int order);
195 
196 /*
197  * results with 256, 32 in the lowmem_reserve sysctl:
198  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
199  *	1G machine -> (16M dma, 784M normal, 224M high)
200  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
201  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
202  *	HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
203  *
204  * TBD: should special case ZONE_DMA32 machines here - in those we normally
205  * don't need any ZONE_NORMAL reservation
206  */
207 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
208 #ifdef CONFIG_ZONE_DMA
209 	[ZONE_DMA] = 256,
210 #endif
211 #ifdef CONFIG_ZONE_DMA32
212 	[ZONE_DMA32] = 256,
213 #endif
214 	[ZONE_NORMAL] = 32,
215 #ifdef CONFIG_HIGHMEM
216 	[ZONE_HIGHMEM] = 0,
217 #endif
218 	[ZONE_MOVABLE] = 0,
219 };
220 
221 EXPORT_SYMBOL(totalram_pages);
222 
223 static char * const zone_names[MAX_NR_ZONES] = {
224 #ifdef CONFIG_ZONE_DMA
225 	 "DMA",
226 #endif
227 #ifdef CONFIG_ZONE_DMA32
228 	 "DMA32",
229 #endif
230 	 "Normal",
231 #ifdef CONFIG_HIGHMEM
232 	 "HighMem",
233 #endif
234 	 "Movable",
235 #ifdef CONFIG_ZONE_DEVICE
236 	 "Device",
237 #endif
238 };
239 
240 char * const migratetype_names[MIGRATE_TYPES] = {
241 	"Unmovable",
242 	"Movable",
243 	"Reclaimable",
244 	"HighAtomic",
245 #ifdef CONFIG_CMA
246 	"CMA",
247 #endif
248 #ifdef CONFIG_MEMORY_ISOLATION
249 	"Isolate",
250 #endif
251 };
252 
253 compound_page_dtor * const compound_page_dtors[] = {
254 	NULL,
255 	free_compound_page,
256 #ifdef CONFIG_HUGETLB_PAGE
257 	free_huge_page,
258 #endif
259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
260 	free_transhuge_page,
261 #endif
262 };
263 
264 int min_free_kbytes = 1024;
265 int user_min_free_kbytes = -1;
266 int watermark_scale_factor = 10;
267 
268 static unsigned long nr_kernel_pages __meminitdata;
269 static unsigned long nr_all_pages __meminitdata;
270 static unsigned long dma_reserve __meminitdata;
271 
272 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
273 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __meminitdata;
274 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __meminitdata;
275 static unsigned long required_kernelcore __initdata;
276 static unsigned long required_kernelcore_percent __initdata;
277 static unsigned long required_movablecore __initdata;
278 static unsigned long required_movablecore_percent __initdata;
279 static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata;
280 static bool mirrored_kernelcore __meminitdata;
281 
282 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
283 int movable_zone;
284 EXPORT_SYMBOL(movable_zone);
285 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
286 
287 #if MAX_NUMNODES > 1
288 int nr_node_ids __read_mostly = MAX_NUMNODES;
289 int nr_online_nodes __read_mostly = 1;
290 EXPORT_SYMBOL(nr_node_ids);
291 EXPORT_SYMBOL(nr_online_nodes);
292 #endif
293 
294 int page_group_by_mobility_disabled __read_mostly;
295 
296 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
297 /* Returns true if the struct page for the pfn is uninitialised */
298 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
299 {
300 	int nid = early_pfn_to_nid(pfn);
301 
302 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
303 		return true;
304 
305 	return false;
306 }
307 
308 /*
309  * Returns true when the remaining initialisation should be deferred until
310  * later in the boot cycle when it can be parallelised.
311  */
312 static bool __meminit
313 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
314 {
315 	static unsigned long prev_end_pfn, nr_initialised;
316 
317 	/*
318 	 * prev_end_pfn static that contains the end of previous zone
319 	 * No need to protect because called very early in boot before smp_init.
320 	 */
321 	if (prev_end_pfn != end_pfn) {
322 		prev_end_pfn = end_pfn;
323 		nr_initialised = 0;
324 	}
325 
326 	/* Always populate low zones for address-constrained allocations */
327 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
328 		return false;
329 	nr_initialised++;
330 	if ((nr_initialised > NODE_DATA(nid)->static_init_pgcnt) &&
331 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
332 		NODE_DATA(nid)->first_deferred_pfn = pfn;
333 		return true;
334 	}
335 	return false;
336 }
337 #else
338 static inline bool early_page_uninitialised(unsigned long pfn)
339 {
340 	return false;
341 }
342 
343 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
344 {
345 	return false;
346 }
347 #endif
348 
349 /* Return a pointer to the bitmap storing bits affecting a block of pages */
350 static inline unsigned long *get_pageblock_bitmap(struct page *page,
351 							unsigned long pfn)
352 {
353 #ifdef CONFIG_SPARSEMEM
354 	return __pfn_to_section(pfn)->pageblock_flags;
355 #else
356 	return page_zone(page)->pageblock_flags;
357 #endif /* CONFIG_SPARSEMEM */
358 }
359 
360 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
361 {
362 #ifdef CONFIG_SPARSEMEM
363 	pfn &= (PAGES_PER_SECTION-1);
364 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
365 #else
366 	pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
367 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
368 #endif /* CONFIG_SPARSEMEM */
369 }
370 
371 /**
372  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
373  * @page: The page within the block of interest
374  * @pfn: The target page frame number
375  * @end_bitidx: The last bit of interest to retrieve
376  * @mask: mask of bits that the caller is interested in
377  *
378  * Return: pageblock_bits flags
379  */
380 static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
381 					unsigned long pfn,
382 					unsigned long end_bitidx,
383 					unsigned long mask)
384 {
385 	unsigned long *bitmap;
386 	unsigned long bitidx, word_bitidx;
387 	unsigned long word;
388 
389 	bitmap = get_pageblock_bitmap(page, pfn);
390 	bitidx = pfn_to_bitidx(page, pfn);
391 	word_bitidx = bitidx / BITS_PER_LONG;
392 	bitidx &= (BITS_PER_LONG-1);
393 
394 	word = bitmap[word_bitidx];
395 	bitidx += end_bitidx;
396 	return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
397 }
398 
399 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
400 					unsigned long end_bitidx,
401 					unsigned long mask)
402 {
403 	return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
404 }
405 
406 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
407 {
408 	return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
409 }
410 
411 /**
412  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
413  * @page: The page within the block of interest
414  * @flags: The flags to set
415  * @pfn: The target page frame number
416  * @end_bitidx: The last bit of interest
417  * @mask: mask of bits that the caller is interested in
418  */
419 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
420 					unsigned long pfn,
421 					unsigned long end_bitidx,
422 					unsigned long mask)
423 {
424 	unsigned long *bitmap;
425 	unsigned long bitidx, word_bitidx;
426 	unsigned long old_word, word;
427 
428 	BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
429 
430 	bitmap = get_pageblock_bitmap(page, pfn);
431 	bitidx = pfn_to_bitidx(page, pfn);
432 	word_bitidx = bitidx / BITS_PER_LONG;
433 	bitidx &= (BITS_PER_LONG-1);
434 
435 	VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
436 
437 	bitidx += end_bitidx;
438 	mask <<= (BITS_PER_LONG - bitidx - 1);
439 	flags <<= (BITS_PER_LONG - bitidx - 1);
440 
441 	word = READ_ONCE(bitmap[word_bitidx]);
442 	for (;;) {
443 		old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
444 		if (word == old_word)
445 			break;
446 		word = old_word;
447 	}
448 }
449 
450 void set_pageblock_migratetype(struct page *page, int migratetype)
451 {
452 	if (unlikely(page_group_by_mobility_disabled &&
453 		     migratetype < MIGRATE_PCPTYPES))
454 		migratetype = MIGRATE_UNMOVABLE;
455 
456 	set_pageblock_flags_group(page, (unsigned long)migratetype,
457 					PB_migrate, PB_migrate_end);
458 }
459 
460 #ifdef CONFIG_DEBUG_VM
461 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
462 {
463 	int ret = 0;
464 	unsigned seq;
465 	unsigned long pfn = page_to_pfn(page);
466 	unsigned long sp, start_pfn;
467 
468 	do {
469 		seq = zone_span_seqbegin(zone);
470 		start_pfn = zone->zone_start_pfn;
471 		sp = zone->spanned_pages;
472 		if (!zone_spans_pfn(zone, pfn))
473 			ret = 1;
474 	} while (zone_span_seqretry(zone, seq));
475 
476 	if (ret)
477 		pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
478 			pfn, zone_to_nid(zone), zone->name,
479 			start_pfn, start_pfn + sp);
480 
481 	return ret;
482 }
483 
484 static int page_is_consistent(struct zone *zone, struct page *page)
485 {
486 	if (!pfn_valid_within(page_to_pfn(page)))
487 		return 0;
488 	if (zone != page_zone(page))
489 		return 0;
490 
491 	return 1;
492 }
493 /*
494  * Temporary debugging check for pages not lying within a given zone.
495  */
496 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
497 {
498 	if (page_outside_zone_boundaries(zone, page))
499 		return 1;
500 	if (!page_is_consistent(zone, page))
501 		return 1;
502 
503 	return 0;
504 }
505 #else
506 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
507 {
508 	return 0;
509 }
510 #endif
511 
512 static void bad_page(struct page *page, const char *reason,
513 		unsigned long bad_flags)
514 {
515 	static unsigned long resume;
516 	static unsigned long nr_shown;
517 	static unsigned long nr_unshown;
518 
519 	/*
520 	 * Allow a burst of 60 reports, then keep quiet for that minute;
521 	 * or allow a steady drip of one report per second.
522 	 */
523 	if (nr_shown == 60) {
524 		if (time_before(jiffies, resume)) {
525 			nr_unshown++;
526 			goto out;
527 		}
528 		if (nr_unshown) {
529 			pr_alert(
530 			      "BUG: Bad page state: %lu messages suppressed\n",
531 				nr_unshown);
532 			nr_unshown = 0;
533 		}
534 		nr_shown = 0;
535 	}
536 	if (nr_shown++ == 0)
537 		resume = jiffies + 60 * HZ;
538 
539 	pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
540 		current->comm, page_to_pfn(page));
541 	__dump_page(page, reason);
542 	bad_flags &= page->flags;
543 	if (bad_flags)
544 		pr_alert("bad because of flags: %#lx(%pGp)\n",
545 						bad_flags, &bad_flags);
546 	dump_page_owner(page);
547 
548 	print_modules();
549 	dump_stack();
550 out:
551 	/* Leave bad fields for debug, except PageBuddy could make trouble */
552 	page_mapcount_reset(page); /* remove PageBuddy */
553 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
554 }
555 
556 /*
557  * Higher-order pages are called "compound pages".  They are structured thusly:
558  *
559  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
560  *
561  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
562  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
563  *
564  * The first tail page's ->compound_dtor holds the offset in array of compound
565  * page destructors. See compound_page_dtors.
566  *
567  * The first tail page's ->compound_order holds the order of allocation.
568  * This usage means that zero-order pages may not be compound.
569  */
570 
571 void free_compound_page(struct page *page)
572 {
573 	__free_pages_ok(page, compound_order(page));
574 }
575 
576 void prep_compound_page(struct page *page, unsigned int order)
577 {
578 	int i;
579 	int nr_pages = 1 << order;
580 
581 	set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
582 	set_compound_order(page, order);
583 	__SetPageHead(page);
584 	for (i = 1; i < nr_pages; i++) {
585 		struct page *p = page + i;
586 		set_page_count(p, 0);
587 		p->mapping = TAIL_MAPPING;
588 		set_compound_head(p, page);
589 	}
590 	atomic_set(compound_mapcount_ptr(page), -1);
591 }
592 
593 #ifdef CONFIG_DEBUG_PAGEALLOC
594 unsigned int _debug_guardpage_minorder;
595 bool _debug_pagealloc_enabled __read_mostly
596 			= IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
597 EXPORT_SYMBOL(_debug_pagealloc_enabled);
598 bool _debug_guardpage_enabled __read_mostly;
599 
600 static int __init early_debug_pagealloc(char *buf)
601 {
602 	if (!buf)
603 		return -EINVAL;
604 	return kstrtobool(buf, &_debug_pagealloc_enabled);
605 }
606 early_param("debug_pagealloc", early_debug_pagealloc);
607 
608 static bool need_debug_guardpage(void)
609 {
610 	/* If we don't use debug_pagealloc, we don't need guard page */
611 	if (!debug_pagealloc_enabled())
612 		return false;
613 
614 	if (!debug_guardpage_minorder())
615 		return false;
616 
617 	return true;
618 }
619 
620 static void init_debug_guardpage(void)
621 {
622 	if (!debug_pagealloc_enabled())
623 		return;
624 
625 	if (!debug_guardpage_minorder())
626 		return;
627 
628 	_debug_guardpage_enabled = true;
629 }
630 
631 struct page_ext_operations debug_guardpage_ops = {
632 	.need = need_debug_guardpage,
633 	.init = init_debug_guardpage,
634 };
635 
636 static int __init debug_guardpage_minorder_setup(char *buf)
637 {
638 	unsigned long res;
639 
640 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
641 		pr_err("Bad debug_guardpage_minorder value\n");
642 		return 0;
643 	}
644 	_debug_guardpage_minorder = res;
645 	pr_info("Setting debug_guardpage_minorder to %lu\n", res);
646 	return 0;
647 }
648 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
649 
650 static inline bool set_page_guard(struct zone *zone, struct page *page,
651 				unsigned int order, int migratetype)
652 {
653 	struct page_ext *page_ext;
654 
655 	if (!debug_guardpage_enabled())
656 		return false;
657 
658 	if (order >= debug_guardpage_minorder())
659 		return false;
660 
661 	page_ext = lookup_page_ext(page);
662 	if (unlikely(!page_ext))
663 		return false;
664 
665 	__set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
666 
667 	INIT_LIST_HEAD(&page->lru);
668 	set_page_private(page, order);
669 	/* Guard pages are not available for any usage */
670 	__mod_zone_freepage_state(zone, -(1 << order), migratetype);
671 
672 	return true;
673 }
674 
675 static inline void clear_page_guard(struct zone *zone, struct page *page,
676 				unsigned int order, int migratetype)
677 {
678 	struct page_ext *page_ext;
679 
680 	if (!debug_guardpage_enabled())
681 		return;
682 
683 	page_ext = lookup_page_ext(page);
684 	if (unlikely(!page_ext))
685 		return;
686 
687 	__clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
688 
689 	set_page_private(page, 0);
690 	if (!is_migrate_isolate(migratetype))
691 		__mod_zone_freepage_state(zone, (1 << order), migratetype);
692 }
693 #else
694 struct page_ext_operations debug_guardpage_ops;
695 static inline bool set_page_guard(struct zone *zone, struct page *page,
696 			unsigned int order, int migratetype) { return false; }
697 static inline void clear_page_guard(struct zone *zone, struct page *page,
698 				unsigned int order, int migratetype) {}
699 #endif
700 
701 static inline void set_page_order(struct page *page, unsigned int order)
702 {
703 	set_page_private(page, order);
704 	__SetPageBuddy(page);
705 }
706 
707 static inline void rmv_page_order(struct page *page)
708 {
709 	__ClearPageBuddy(page);
710 	set_page_private(page, 0);
711 }
712 
713 /*
714  * This function checks whether a page is free && is the buddy
715  * we can coalesce a page and its buddy if
716  * (a) the buddy is not in a hole (check before calling!) &&
717  * (b) the buddy is in the buddy system &&
718  * (c) a page and its buddy have the same order &&
719  * (d) a page and its buddy are in the same zone.
720  *
721  * For recording whether a page is in the buddy system, we set PageBuddy.
722  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
723  *
724  * For recording page's order, we use page_private(page).
725  */
726 static inline int page_is_buddy(struct page *page, struct page *buddy,
727 							unsigned int order)
728 {
729 	if (page_is_guard(buddy) && page_order(buddy) == order) {
730 		if (page_zone_id(page) != page_zone_id(buddy))
731 			return 0;
732 
733 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
734 
735 		return 1;
736 	}
737 
738 	if (PageBuddy(buddy) && page_order(buddy) == order) {
739 		/*
740 		 * zone check is done late to avoid uselessly
741 		 * calculating zone/node ids for pages that could
742 		 * never merge.
743 		 */
744 		if (page_zone_id(page) != page_zone_id(buddy))
745 			return 0;
746 
747 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
748 
749 		return 1;
750 	}
751 	return 0;
752 }
753 
754 /*
755  * Freeing function for a buddy system allocator.
756  *
757  * The concept of a buddy system is to maintain direct-mapped table
758  * (containing bit values) for memory blocks of various "orders".
759  * The bottom level table contains the map for the smallest allocatable
760  * units of memory (here, pages), and each level above it describes
761  * pairs of units from the levels below, hence, "buddies".
762  * At a high level, all that happens here is marking the table entry
763  * at the bottom level available, and propagating the changes upward
764  * as necessary, plus some accounting needed to play nicely with other
765  * parts of the VM system.
766  * At each level, we keep a list of pages, which are heads of continuous
767  * free pages of length of (1 << order) and marked with PageBuddy.
768  * Page's order is recorded in page_private(page) field.
769  * So when we are allocating or freeing one, we can derive the state of the
770  * other.  That is, if we allocate a small block, and both were
771  * free, the remainder of the region must be split into blocks.
772  * If a block is freed, and its buddy is also free, then this
773  * triggers coalescing into a block of larger size.
774  *
775  * -- nyc
776  */
777 
778 static inline void __free_one_page(struct page *page,
779 		unsigned long pfn,
780 		struct zone *zone, unsigned int order,
781 		int migratetype)
782 {
783 	unsigned long combined_pfn;
784 	unsigned long uninitialized_var(buddy_pfn);
785 	struct page *buddy;
786 	unsigned int max_order;
787 
788 	max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
789 
790 	VM_BUG_ON(!zone_is_initialized(zone));
791 	VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
792 
793 	VM_BUG_ON(migratetype == -1);
794 	if (likely(!is_migrate_isolate(migratetype)))
795 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
796 
797 	VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
798 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
799 
800 continue_merging:
801 	while (order < max_order - 1) {
802 		buddy_pfn = __find_buddy_pfn(pfn, order);
803 		buddy = page + (buddy_pfn - pfn);
804 
805 		if (!pfn_valid_within(buddy_pfn))
806 			goto done_merging;
807 		if (!page_is_buddy(page, buddy, order))
808 			goto done_merging;
809 		/*
810 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
811 		 * merge with it and move up one order.
812 		 */
813 		if (page_is_guard(buddy)) {
814 			clear_page_guard(zone, buddy, order, migratetype);
815 		} else {
816 			list_del(&buddy->lru);
817 			zone->free_area[order].nr_free--;
818 			rmv_page_order(buddy);
819 		}
820 		combined_pfn = buddy_pfn & pfn;
821 		page = page + (combined_pfn - pfn);
822 		pfn = combined_pfn;
823 		order++;
824 	}
825 	if (max_order < MAX_ORDER) {
826 		/* If we are here, it means order is >= pageblock_order.
827 		 * We want to prevent merge between freepages on isolate
828 		 * pageblock and normal pageblock. Without this, pageblock
829 		 * isolation could cause incorrect freepage or CMA accounting.
830 		 *
831 		 * We don't want to hit this code for the more frequent
832 		 * low-order merging.
833 		 */
834 		if (unlikely(has_isolate_pageblock(zone))) {
835 			int buddy_mt;
836 
837 			buddy_pfn = __find_buddy_pfn(pfn, order);
838 			buddy = page + (buddy_pfn - pfn);
839 			buddy_mt = get_pageblock_migratetype(buddy);
840 
841 			if (migratetype != buddy_mt
842 					&& (is_migrate_isolate(migratetype) ||
843 						is_migrate_isolate(buddy_mt)))
844 				goto done_merging;
845 		}
846 		max_order++;
847 		goto continue_merging;
848 	}
849 
850 done_merging:
851 	set_page_order(page, order);
852 
853 	/*
854 	 * If this is not the largest possible page, check if the buddy
855 	 * of the next-highest order is free. If it is, it's possible
856 	 * that pages are being freed that will coalesce soon. In case,
857 	 * that is happening, add the free page to the tail of the list
858 	 * so it's less likely to be used soon and more likely to be merged
859 	 * as a higher order page
860 	 */
861 	if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
862 		struct page *higher_page, *higher_buddy;
863 		combined_pfn = buddy_pfn & pfn;
864 		higher_page = page + (combined_pfn - pfn);
865 		buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
866 		higher_buddy = higher_page + (buddy_pfn - combined_pfn);
867 		if (pfn_valid_within(buddy_pfn) &&
868 		    page_is_buddy(higher_page, higher_buddy, order + 1)) {
869 			list_add_tail(&page->lru,
870 				&zone->free_area[order].free_list[migratetype]);
871 			goto out;
872 		}
873 	}
874 
875 	list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
876 out:
877 	zone->free_area[order].nr_free++;
878 }
879 
880 /*
881  * A bad page could be due to a number of fields. Instead of multiple branches,
882  * try and check multiple fields with one check. The caller must do a detailed
883  * check if necessary.
884  */
885 static inline bool page_expected_state(struct page *page,
886 					unsigned long check_flags)
887 {
888 	if (unlikely(atomic_read(&page->_mapcount) != -1))
889 		return false;
890 
891 	if (unlikely((unsigned long)page->mapping |
892 			page_ref_count(page) |
893 #ifdef CONFIG_MEMCG
894 			(unsigned long)page->mem_cgroup |
895 #endif
896 			(page->flags & check_flags)))
897 		return false;
898 
899 	return true;
900 }
901 
902 static void free_pages_check_bad(struct page *page)
903 {
904 	const char *bad_reason;
905 	unsigned long bad_flags;
906 
907 	bad_reason = NULL;
908 	bad_flags = 0;
909 
910 	if (unlikely(atomic_read(&page->_mapcount) != -1))
911 		bad_reason = "nonzero mapcount";
912 	if (unlikely(page->mapping != NULL))
913 		bad_reason = "non-NULL mapping";
914 	if (unlikely(page_ref_count(page) != 0))
915 		bad_reason = "nonzero _refcount";
916 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
917 		bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
918 		bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
919 	}
920 #ifdef CONFIG_MEMCG
921 	if (unlikely(page->mem_cgroup))
922 		bad_reason = "page still charged to cgroup";
923 #endif
924 	bad_page(page, bad_reason, bad_flags);
925 }
926 
927 static inline int free_pages_check(struct page *page)
928 {
929 	if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
930 		return 0;
931 
932 	/* Something has gone sideways, find it */
933 	free_pages_check_bad(page);
934 	return 1;
935 }
936 
937 static int free_tail_pages_check(struct page *head_page, struct page *page)
938 {
939 	int ret = 1;
940 
941 	/*
942 	 * We rely page->lru.next never has bit 0 set, unless the page
943 	 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
944 	 */
945 	BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
946 
947 	if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
948 		ret = 0;
949 		goto out;
950 	}
951 	switch (page - head_page) {
952 	case 1:
953 		/* the first tail page: ->mapping may be compound_mapcount() */
954 		if (unlikely(compound_mapcount(page))) {
955 			bad_page(page, "nonzero compound_mapcount", 0);
956 			goto out;
957 		}
958 		break;
959 	case 2:
960 		/*
961 		 * the second tail page: ->mapping is
962 		 * deferred_list.next -- ignore value.
963 		 */
964 		break;
965 	default:
966 		if (page->mapping != TAIL_MAPPING) {
967 			bad_page(page, "corrupted mapping in tail page", 0);
968 			goto out;
969 		}
970 		break;
971 	}
972 	if (unlikely(!PageTail(page))) {
973 		bad_page(page, "PageTail not set", 0);
974 		goto out;
975 	}
976 	if (unlikely(compound_head(page) != head_page)) {
977 		bad_page(page, "compound_head not consistent", 0);
978 		goto out;
979 	}
980 	ret = 0;
981 out:
982 	page->mapping = NULL;
983 	clear_compound_head(page);
984 	return ret;
985 }
986 
987 static __always_inline bool free_pages_prepare(struct page *page,
988 					unsigned int order, bool check_free)
989 {
990 	int bad = 0;
991 
992 	VM_BUG_ON_PAGE(PageTail(page), page);
993 
994 	trace_mm_page_free(page, order);
995 
996 	/*
997 	 * Check tail pages before head page information is cleared to
998 	 * avoid checking PageCompound for order-0 pages.
999 	 */
1000 	if (unlikely(order)) {
1001 		bool compound = PageCompound(page);
1002 		int i;
1003 
1004 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1005 
1006 		if (compound)
1007 			ClearPageDoubleMap(page);
1008 		for (i = 1; i < (1 << order); i++) {
1009 			if (compound)
1010 				bad += free_tail_pages_check(page, page + i);
1011 			if (unlikely(free_pages_check(page + i))) {
1012 				bad++;
1013 				continue;
1014 			}
1015 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1016 		}
1017 	}
1018 	if (PageMappingFlags(page))
1019 		page->mapping = NULL;
1020 	if (memcg_kmem_enabled() && PageKmemcg(page))
1021 		memcg_kmem_uncharge(page, order);
1022 	if (check_free)
1023 		bad += free_pages_check(page);
1024 	if (bad)
1025 		return false;
1026 
1027 	page_cpupid_reset_last(page);
1028 	page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1029 	reset_page_owner(page, order);
1030 
1031 	if (!PageHighMem(page)) {
1032 		debug_check_no_locks_freed(page_address(page),
1033 					   PAGE_SIZE << order);
1034 		debug_check_no_obj_freed(page_address(page),
1035 					   PAGE_SIZE << order);
1036 	}
1037 	arch_free_page(page, order);
1038 	kernel_poison_pages(page, 1 << order, 0);
1039 	kernel_map_pages(page, 1 << order, 0);
1040 	kasan_free_pages(page, order);
1041 
1042 	return true;
1043 }
1044 
1045 #ifdef CONFIG_DEBUG_VM
1046 static inline bool free_pcp_prepare(struct page *page)
1047 {
1048 	return free_pages_prepare(page, 0, true);
1049 }
1050 
1051 static inline bool bulkfree_pcp_prepare(struct page *page)
1052 {
1053 	return false;
1054 }
1055 #else
1056 static bool free_pcp_prepare(struct page *page)
1057 {
1058 	return free_pages_prepare(page, 0, false);
1059 }
1060 
1061 static bool bulkfree_pcp_prepare(struct page *page)
1062 {
1063 	return free_pages_check(page);
1064 }
1065 #endif /* CONFIG_DEBUG_VM */
1066 
1067 static inline void prefetch_buddy(struct page *page)
1068 {
1069 	unsigned long pfn = page_to_pfn(page);
1070 	unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1071 	struct page *buddy = page + (buddy_pfn - pfn);
1072 
1073 	prefetch(buddy);
1074 }
1075 
1076 /*
1077  * Frees a number of pages from the PCP lists
1078  * Assumes all pages on list are in same zone, and of same order.
1079  * count is the number of pages to free.
1080  *
1081  * If the zone was previously in an "all pages pinned" state then look to
1082  * see if this freeing clears that state.
1083  *
1084  * And clear the zone's pages_scanned counter, to hold off the "all pages are
1085  * pinned" detection logic.
1086  */
1087 static void free_pcppages_bulk(struct zone *zone, int count,
1088 					struct per_cpu_pages *pcp)
1089 {
1090 	int migratetype = 0;
1091 	int batch_free = 0;
1092 	int prefetch_nr = 0;
1093 	bool isolated_pageblocks;
1094 	struct page *page, *tmp;
1095 	LIST_HEAD(head);
1096 
1097 	while (count) {
1098 		struct list_head *list;
1099 
1100 		/*
1101 		 * Remove pages from lists in a round-robin fashion. A
1102 		 * batch_free count is maintained that is incremented when an
1103 		 * empty list is encountered.  This is so more pages are freed
1104 		 * off fuller lists instead of spinning excessively around empty
1105 		 * lists
1106 		 */
1107 		do {
1108 			batch_free++;
1109 			if (++migratetype == MIGRATE_PCPTYPES)
1110 				migratetype = 0;
1111 			list = &pcp->lists[migratetype];
1112 		} while (list_empty(list));
1113 
1114 		/* This is the only non-empty list. Free them all. */
1115 		if (batch_free == MIGRATE_PCPTYPES)
1116 			batch_free = count;
1117 
1118 		do {
1119 			page = list_last_entry(list, struct page, lru);
1120 			/* must delete to avoid corrupting pcp list */
1121 			list_del(&page->lru);
1122 			pcp->count--;
1123 
1124 			if (bulkfree_pcp_prepare(page))
1125 				continue;
1126 
1127 			list_add_tail(&page->lru, &head);
1128 
1129 			/*
1130 			 * We are going to put the page back to the global
1131 			 * pool, prefetch its buddy to speed up later access
1132 			 * under zone->lock. It is believed the overhead of
1133 			 * an additional test and calculating buddy_pfn here
1134 			 * can be offset by reduced memory latency later. To
1135 			 * avoid excessive prefetching due to large count, only
1136 			 * prefetch buddy for the first pcp->batch nr of pages.
1137 			 */
1138 			if (prefetch_nr++ < pcp->batch)
1139 				prefetch_buddy(page);
1140 		} while (--count && --batch_free && !list_empty(list));
1141 	}
1142 
1143 	spin_lock(&zone->lock);
1144 	isolated_pageblocks = has_isolate_pageblock(zone);
1145 
1146 	/*
1147 	 * Use safe version since after __free_one_page(),
1148 	 * page->lru.next will not point to original list.
1149 	 */
1150 	list_for_each_entry_safe(page, tmp, &head, lru) {
1151 		int mt = get_pcppage_migratetype(page);
1152 		/* MIGRATE_ISOLATE page should not go to pcplists */
1153 		VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1154 		/* Pageblock could have been isolated meanwhile */
1155 		if (unlikely(isolated_pageblocks))
1156 			mt = get_pageblock_migratetype(page);
1157 
1158 		__free_one_page(page, page_to_pfn(page), zone, 0, mt);
1159 		trace_mm_page_pcpu_drain(page, 0, mt);
1160 	}
1161 	spin_unlock(&zone->lock);
1162 }
1163 
1164 static void free_one_page(struct zone *zone,
1165 				struct page *page, unsigned long pfn,
1166 				unsigned int order,
1167 				int migratetype)
1168 {
1169 	spin_lock(&zone->lock);
1170 	if (unlikely(has_isolate_pageblock(zone) ||
1171 		is_migrate_isolate(migratetype))) {
1172 		migratetype = get_pfnblock_migratetype(page, pfn);
1173 	}
1174 	__free_one_page(page, pfn, zone, order, migratetype);
1175 	spin_unlock(&zone->lock);
1176 }
1177 
1178 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1179 				unsigned long zone, int nid)
1180 {
1181 	mm_zero_struct_page(page);
1182 	set_page_links(page, zone, nid, pfn);
1183 	init_page_count(page);
1184 	page_mapcount_reset(page);
1185 	page_cpupid_reset_last(page);
1186 	page_kasan_tag_reset(page);
1187 
1188 	INIT_LIST_HEAD(&page->lru);
1189 #ifdef WANT_PAGE_VIRTUAL
1190 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1191 	if (!is_highmem_idx(zone))
1192 		set_page_address(page, __va(pfn << PAGE_SHIFT));
1193 #endif
1194 }
1195 
1196 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1197 static void __meminit init_reserved_page(unsigned long pfn)
1198 {
1199 	pg_data_t *pgdat;
1200 	int nid, zid;
1201 
1202 	if (!early_page_uninitialised(pfn))
1203 		return;
1204 
1205 	nid = early_pfn_to_nid(pfn);
1206 	pgdat = NODE_DATA(nid);
1207 
1208 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1209 		struct zone *zone = &pgdat->node_zones[zid];
1210 
1211 		if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1212 			break;
1213 	}
1214 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1215 }
1216 #else
1217 static inline void init_reserved_page(unsigned long pfn)
1218 {
1219 }
1220 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1221 
1222 /*
1223  * Initialised pages do not have PageReserved set. This function is
1224  * called for each range allocated by the bootmem allocator and
1225  * marks the pages PageReserved. The remaining valid pages are later
1226  * sent to the buddy page allocator.
1227  */
1228 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1229 {
1230 	unsigned long start_pfn = PFN_DOWN(start);
1231 	unsigned long end_pfn = PFN_UP(end);
1232 
1233 	for (; start_pfn < end_pfn; start_pfn++) {
1234 		if (pfn_valid(start_pfn)) {
1235 			struct page *page = pfn_to_page(start_pfn);
1236 
1237 			init_reserved_page(start_pfn);
1238 
1239 			/* Avoid false-positive PageTail() */
1240 			INIT_LIST_HEAD(&page->lru);
1241 
1242 			/*
1243 			 * no need for atomic set_bit because the struct
1244 			 * page is not visible yet so nobody should
1245 			 * access it yet.
1246 			 */
1247 			__SetPageReserved(page);
1248 		}
1249 	}
1250 }
1251 
1252 static void __free_pages_ok(struct page *page, unsigned int order)
1253 {
1254 	unsigned long flags;
1255 	int migratetype;
1256 	unsigned long pfn = page_to_pfn(page);
1257 
1258 	if (!free_pages_prepare(page, order, true))
1259 		return;
1260 
1261 	migratetype = get_pfnblock_migratetype(page, pfn);
1262 	local_irq_save(flags);
1263 	__count_vm_events(PGFREE, 1 << order);
1264 	free_one_page(page_zone(page), page, pfn, order, migratetype);
1265 	local_irq_restore(flags);
1266 }
1267 
1268 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1269 {
1270 	unsigned int nr_pages = 1 << order;
1271 	struct page *p = page;
1272 	unsigned int loop;
1273 
1274 	prefetchw(p);
1275 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1276 		prefetchw(p + 1);
1277 		__ClearPageReserved(p);
1278 		set_page_count(p, 0);
1279 	}
1280 	__ClearPageReserved(p);
1281 	set_page_count(p, 0);
1282 
1283 	page_zone(page)->managed_pages += nr_pages;
1284 	set_page_refcounted(page);
1285 	__free_pages(page, order);
1286 }
1287 
1288 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1289 	defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1290 
1291 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1292 
1293 int __meminit early_pfn_to_nid(unsigned long pfn)
1294 {
1295 	static DEFINE_SPINLOCK(early_pfn_lock);
1296 	int nid;
1297 
1298 	spin_lock(&early_pfn_lock);
1299 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1300 	if (nid < 0)
1301 		nid = first_online_node;
1302 	spin_unlock(&early_pfn_lock);
1303 
1304 	return nid;
1305 }
1306 #endif
1307 
1308 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1309 static inline bool __meminit __maybe_unused
1310 meminit_pfn_in_nid(unsigned long pfn, int node,
1311 		   struct mminit_pfnnid_cache *state)
1312 {
1313 	int nid;
1314 
1315 	nid = __early_pfn_to_nid(pfn, state);
1316 	if (nid >= 0 && nid != node)
1317 		return false;
1318 	return true;
1319 }
1320 
1321 /* Only safe to use early in boot when initialisation is single-threaded */
1322 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1323 {
1324 	return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1325 }
1326 
1327 #else
1328 
1329 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1330 {
1331 	return true;
1332 }
1333 static inline bool __meminit  __maybe_unused
1334 meminit_pfn_in_nid(unsigned long pfn, int node,
1335 		   struct mminit_pfnnid_cache *state)
1336 {
1337 	return true;
1338 }
1339 #endif
1340 
1341 
1342 void __init memblock_free_pages(struct page *page, unsigned long pfn,
1343 							unsigned int order)
1344 {
1345 	if (early_page_uninitialised(pfn))
1346 		return;
1347 	return __free_pages_boot_core(page, order);
1348 }
1349 
1350 /*
1351  * Check that the whole (or subset of) a pageblock given by the interval of
1352  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1353  * with the migration of free compaction scanner. The scanners then need to
1354  * use only pfn_valid_within() check for arches that allow holes within
1355  * pageblocks.
1356  *
1357  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1358  *
1359  * It's possible on some configurations to have a setup like node0 node1 node0
1360  * i.e. it's possible that all pages within a zones range of pages do not
1361  * belong to a single zone. We assume that a border between node0 and node1
1362  * can occur within a single pageblock, but not a node0 node1 node0
1363  * interleaving within a single pageblock. It is therefore sufficient to check
1364  * the first and last page of a pageblock and avoid checking each individual
1365  * page in a pageblock.
1366  */
1367 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1368 				     unsigned long end_pfn, struct zone *zone)
1369 {
1370 	struct page *start_page;
1371 	struct page *end_page;
1372 
1373 	/* end_pfn is one past the range we are checking */
1374 	end_pfn--;
1375 
1376 	if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1377 		return NULL;
1378 
1379 	start_page = pfn_to_online_page(start_pfn);
1380 	if (!start_page)
1381 		return NULL;
1382 
1383 	if (page_zone(start_page) != zone)
1384 		return NULL;
1385 
1386 	end_page = pfn_to_page(end_pfn);
1387 
1388 	/* This gives a shorter code than deriving page_zone(end_page) */
1389 	if (page_zone_id(start_page) != page_zone_id(end_page))
1390 		return NULL;
1391 
1392 	return start_page;
1393 }
1394 
1395 void set_zone_contiguous(struct zone *zone)
1396 {
1397 	unsigned long block_start_pfn = zone->zone_start_pfn;
1398 	unsigned long block_end_pfn;
1399 
1400 	block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1401 	for (; block_start_pfn < zone_end_pfn(zone);
1402 			block_start_pfn = block_end_pfn,
1403 			 block_end_pfn += pageblock_nr_pages) {
1404 
1405 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1406 
1407 		if (!__pageblock_pfn_to_page(block_start_pfn,
1408 					     block_end_pfn, zone))
1409 			return;
1410 	}
1411 
1412 	/* We confirm that there is no hole */
1413 	zone->contiguous = true;
1414 }
1415 
1416 void clear_zone_contiguous(struct zone *zone)
1417 {
1418 	zone->contiguous = false;
1419 }
1420 
1421 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1422 static void __init deferred_free_range(unsigned long pfn,
1423 				       unsigned long nr_pages)
1424 {
1425 	struct page *page;
1426 	unsigned long i;
1427 
1428 	if (!nr_pages)
1429 		return;
1430 
1431 	page = pfn_to_page(pfn);
1432 
1433 	/* Free a large naturally-aligned chunk if possible */
1434 	if (nr_pages == pageblock_nr_pages &&
1435 	    (pfn & (pageblock_nr_pages - 1)) == 0) {
1436 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1437 		__free_pages_boot_core(page, pageblock_order);
1438 		return;
1439 	}
1440 
1441 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1442 		if ((pfn & (pageblock_nr_pages - 1)) == 0)
1443 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1444 		__free_pages_boot_core(page, 0);
1445 	}
1446 }
1447 
1448 /* Completion tracking for deferred_init_memmap() threads */
1449 static atomic_t pgdat_init_n_undone __initdata;
1450 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1451 
1452 static inline void __init pgdat_init_report_one_done(void)
1453 {
1454 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1455 		complete(&pgdat_init_all_done_comp);
1456 }
1457 
1458 /*
1459  * Returns true if page needs to be initialized or freed to buddy allocator.
1460  *
1461  * First we check if pfn is valid on architectures where it is possible to have
1462  * holes within pageblock_nr_pages. On systems where it is not possible, this
1463  * function is optimized out.
1464  *
1465  * Then, we check if a current large page is valid by only checking the validity
1466  * of the head pfn.
1467  *
1468  * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave
1469  * within a node: a pfn is between start and end of a node, but does not belong
1470  * to this memory node.
1471  */
1472 static inline bool __init
1473 deferred_pfn_valid(int nid, unsigned long pfn,
1474 		   struct mminit_pfnnid_cache *nid_init_state)
1475 {
1476 	if (!pfn_valid_within(pfn))
1477 		return false;
1478 	if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1479 		return false;
1480 	if (!meminit_pfn_in_nid(pfn, nid, nid_init_state))
1481 		return false;
1482 	return true;
1483 }
1484 
1485 /*
1486  * Free pages to buddy allocator. Try to free aligned pages in
1487  * pageblock_nr_pages sizes.
1488  */
1489 static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
1490 				       unsigned long end_pfn)
1491 {
1492 	struct mminit_pfnnid_cache nid_init_state = { };
1493 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1494 	unsigned long nr_free = 0;
1495 
1496 	for (; pfn < end_pfn; pfn++) {
1497 		if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1498 			deferred_free_range(pfn - nr_free, nr_free);
1499 			nr_free = 0;
1500 		} else if (!(pfn & nr_pgmask)) {
1501 			deferred_free_range(pfn - nr_free, nr_free);
1502 			nr_free = 1;
1503 			touch_nmi_watchdog();
1504 		} else {
1505 			nr_free++;
1506 		}
1507 	}
1508 	/* Free the last block of pages to allocator */
1509 	deferred_free_range(pfn - nr_free, nr_free);
1510 }
1511 
1512 /*
1513  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
1514  * by performing it only once every pageblock_nr_pages.
1515  * Return number of pages initialized.
1516  */
1517 static unsigned long  __init deferred_init_pages(int nid, int zid,
1518 						 unsigned long pfn,
1519 						 unsigned long end_pfn)
1520 {
1521 	struct mminit_pfnnid_cache nid_init_state = { };
1522 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1523 	unsigned long nr_pages = 0;
1524 	struct page *page = NULL;
1525 
1526 	for (; pfn < end_pfn; pfn++) {
1527 		if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1528 			page = NULL;
1529 			continue;
1530 		} else if (!page || !(pfn & nr_pgmask)) {
1531 			page = pfn_to_page(pfn);
1532 			touch_nmi_watchdog();
1533 		} else {
1534 			page++;
1535 		}
1536 		__init_single_page(page, pfn, zid, nid);
1537 		nr_pages++;
1538 	}
1539 	return (nr_pages);
1540 }
1541 
1542 /* Initialise remaining memory on a node */
1543 static int __init deferred_init_memmap(void *data)
1544 {
1545 	pg_data_t *pgdat = data;
1546 	int nid = pgdat->node_id;
1547 	unsigned long start = jiffies;
1548 	unsigned long nr_pages = 0;
1549 	unsigned long spfn, epfn, first_init_pfn, flags;
1550 	phys_addr_t spa, epa;
1551 	int zid;
1552 	struct zone *zone;
1553 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1554 	u64 i;
1555 
1556 	/* Bind memory initialisation thread to a local node if possible */
1557 	if (!cpumask_empty(cpumask))
1558 		set_cpus_allowed_ptr(current, cpumask);
1559 
1560 	pgdat_resize_lock(pgdat, &flags);
1561 	first_init_pfn = pgdat->first_deferred_pfn;
1562 	if (first_init_pfn == ULONG_MAX) {
1563 		pgdat_resize_unlock(pgdat, &flags);
1564 		pgdat_init_report_one_done();
1565 		return 0;
1566 	}
1567 
1568 	/* Sanity check boundaries */
1569 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1570 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1571 	pgdat->first_deferred_pfn = ULONG_MAX;
1572 
1573 	/* Only the highest zone is deferred so find it */
1574 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1575 		zone = pgdat->node_zones + zid;
1576 		if (first_init_pfn < zone_end_pfn(zone))
1577 			break;
1578 	}
1579 	first_init_pfn = max(zone->zone_start_pfn, first_init_pfn);
1580 
1581 	/*
1582 	 * Initialize and free pages. We do it in two loops: first we initialize
1583 	 * struct page, than free to buddy allocator, because while we are
1584 	 * freeing pages we can access pages that are ahead (computing buddy
1585 	 * page in __free_one_page()).
1586 	 */
1587 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1588 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1589 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1590 		nr_pages += deferred_init_pages(nid, zid, spfn, epfn);
1591 	}
1592 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1593 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1594 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1595 		deferred_free_pages(nid, zid, spfn, epfn);
1596 	}
1597 	pgdat_resize_unlock(pgdat, &flags);
1598 
1599 	/* Sanity check that the next zone really is unpopulated */
1600 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1601 
1602 	pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1603 					jiffies_to_msecs(jiffies - start));
1604 
1605 	pgdat_init_report_one_done();
1606 	return 0;
1607 }
1608 
1609 /*
1610  * During boot we initialize deferred pages on-demand, as needed, but once
1611  * page_alloc_init_late() has finished, the deferred pages are all initialized,
1612  * and we can permanently disable that path.
1613  */
1614 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
1615 
1616 /*
1617  * If this zone has deferred pages, try to grow it by initializing enough
1618  * deferred pages to satisfy the allocation specified by order, rounded up to
1619  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
1620  * of SECTION_SIZE bytes by initializing struct pages in increments of
1621  * PAGES_PER_SECTION * sizeof(struct page) bytes.
1622  *
1623  * Return true when zone was grown, otherwise return false. We return true even
1624  * when we grow less than requested, to let the caller decide if there are
1625  * enough pages to satisfy the allocation.
1626  *
1627  * Note: We use noinline because this function is needed only during boot, and
1628  * it is called from a __ref function _deferred_grow_zone. This way we are
1629  * making sure that it is not inlined into permanent text section.
1630  */
1631 static noinline bool __init
1632 deferred_grow_zone(struct zone *zone, unsigned int order)
1633 {
1634 	int zid = zone_idx(zone);
1635 	int nid = zone_to_nid(zone);
1636 	pg_data_t *pgdat = NODE_DATA(nid);
1637 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1638 	unsigned long nr_pages = 0;
1639 	unsigned long first_init_pfn, spfn, epfn, t, flags;
1640 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1641 	phys_addr_t spa, epa;
1642 	u64 i;
1643 
1644 	/* Only the last zone may have deferred pages */
1645 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1646 		return false;
1647 
1648 	pgdat_resize_lock(pgdat, &flags);
1649 
1650 	/*
1651 	 * If deferred pages have been initialized while we were waiting for
1652 	 * the lock, return true, as the zone was grown.  The caller will retry
1653 	 * this zone.  We won't return to this function since the caller also
1654 	 * has this static branch.
1655 	 */
1656 	if (!static_branch_unlikely(&deferred_pages)) {
1657 		pgdat_resize_unlock(pgdat, &flags);
1658 		return true;
1659 	}
1660 
1661 	/*
1662 	 * If someone grew this zone while we were waiting for spinlock, return
1663 	 * true, as there might be enough pages already.
1664 	 */
1665 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1666 		pgdat_resize_unlock(pgdat, &flags);
1667 		return true;
1668 	}
1669 
1670 	first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn);
1671 
1672 	if (first_init_pfn >= pgdat_end_pfn(pgdat)) {
1673 		pgdat_resize_unlock(pgdat, &flags);
1674 		return false;
1675 	}
1676 
1677 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1678 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1679 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1680 
1681 		while (spfn < epfn && nr_pages < nr_pages_needed) {
1682 			t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION);
1683 			first_deferred_pfn = min(t, epfn);
1684 			nr_pages += deferred_init_pages(nid, zid, spfn,
1685 							first_deferred_pfn);
1686 			spfn = first_deferred_pfn;
1687 		}
1688 
1689 		if (nr_pages >= nr_pages_needed)
1690 			break;
1691 	}
1692 
1693 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1694 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1695 		epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa));
1696 		deferred_free_pages(nid, zid, spfn, epfn);
1697 
1698 		if (first_deferred_pfn == epfn)
1699 			break;
1700 	}
1701 	pgdat->first_deferred_pfn = first_deferred_pfn;
1702 	pgdat_resize_unlock(pgdat, &flags);
1703 
1704 	return nr_pages > 0;
1705 }
1706 
1707 /*
1708  * deferred_grow_zone() is __init, but it is called from
1709  * get_page_from_freelist() during early boot until deferred_pages permanently
1710  * disables this call. This is why we have refdata wrapper to avoid warning,
1711  * and to ensure that the function body gets unloaded.
1712  */
1713 static bool __ref
1714 _deferred_grow_zone(struct zone *zone, unsigned int order)
1715 {
1716 	return deferred_grow_zone(zone, order);
1717 }
1718 
1719 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1720 
1721 void __init page_alloc_init_late(void)
1722 {
1723 	struct zone *zone;
1724 
1725 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1726 	int nid;
1727 
1728 	/* There will be num_node_state(N_MEMORY) threads */
1729 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1730 	for_each_node_state(nid, N_MEMORY) {
1731 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1732 	}
1733 
1734 	/* Block until all are initialised */
1735 	wait_for_completion(&pgdat_init_all_done_comp);
1736 
1737 	/*
1738 	 * We initialized the rest of the deferred pages.  Permanently disable
1739 	 * on-demand struct page initialization.
1740 	 */
1741 	static_branch_disable(&deferred_pages);
1742 
1743 	/* Reinit limits that are based on free pages after the kernel is up */
1744 	files_maxfiles_init();
1745 #endif
1746 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
1747 	/* Discard memblock private memory */
1748 	memblock_discard();
1749 #endif
1750 
1751 	for_each_populated_zone(zone)
1752 		set_zone_contiguous(zone);
1753 }
1754 
1755 #ifdef CONFIG_CMA
1756 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1757 void __init init_cma_reserved_pageblock(struct page *page)
1758 {
1759 	unsigned i = pageblock_nr_pages;
1760 	struct page *p = page;
1761 
1762 	do {
1763 		__ClearPageReserved(p);
1764 		set_page_count(p, 0);
1765 	} while (++p, --i);
1766 
1767 	set_pageblock_migratetype(page, MIGRATE_CMA);
1768 
1769 	if (pageblock_order >= MAX_ORDER) {
1770 		i = pageblock_nr_pages;
1771 		p = page;
1772 		do {
1773 			set_page_refcounted(p);
1774 			__free_pages(p, MAX_ORDER - 1);
1775 			p += MAX_ORDER_NR_PAGES;
1776 		} while (i -= MAX_ORDER_NR_PAGES);
1777 	} else {
1778 		set_page_refcounted(page);
1779 		__free_pages(page, pageblock_order);
1780 	}
1781 
1782 	adjust_managed_page_count(page, pageblock_nr_pages);
1783 }
1784 #endif
1785 
1786 /*
1787  * The order of subdivision here is critical for the IO subsystem.
1788  * Please do not alter this order without good reasons and regression
1789  * testing. Specifically, as large blocks of memory are subdivided,
1790  * the order in which smaller blocks are delivered depends on the order
1791  * they're subdivided in this function. This is the primary factor
1792  * influencing the order in which pages are delivered to the IO
1793  * subsystem according to empirical testing, and this is also justified
1794  * by considering the behavior of a buddy system containing a single
1795  * large block of memory acted on by a series of small allocations.
1796  * This behavior is a critical factor in sglist merging's success.
1797  *
1798  * -- nyc
1799  */
1800 static inline void expand(struct zone *zone, struct page *page,
1801 	int low, int high, struct free_area *area,
1802 	int migratetype)
1803 {
1804 	unsigned long size = 1 << high;
1805 
1806 	while (high > low) {
1807 		area--;
1808 		high--;
1809 		size >>= 1;
1810 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1811 
1812 		/*
1813 		 * Mark as guard pages (or page), that will allow to
1814 		 * merge back to allocator when buddy will be freed.
1815 		 * Corresponding page table entries will not be touched,
1816 		 * pages will stay not present in virtual address space
1817 		 */
1818 		if (set_page_guard(zone, &page[size], high, migratetype))
1819 			continue;
1820 
1821 		list_add(&page[size].lru, &area->free_list[migratetype]);
1822 		area->nr_free++;
1823 		set_page_order(&page[size], high);
1824 	}
1825 }
1826 
1827 static void check_new_page_bad(struct page *page)
1828 {
1829 	const char *bad_reason = NULL;
1830 	unsigned long bad_flags = 0;
1831 
1832 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1833 		bad_reason = "nonzero mapcount";
1834 	if (unlikely(page->mapping != NULL))
1835 		bad_reason = "non-NULL mapping";
1836 	if (unlikely(page_ref_count(page) != 0))
1837 		bad_reason = "nonzero _count";
1838 	if (unlikely(page->flags & __PG_HWPOISON)) {
1839 		bad_reason = "HWPoisoned (hardware-corrupted)";
1840 		bad_flags = __PG_HWPOISON;
1841 		/* Don't complain about hwpoisoned pages */
1842 		page_mapcount_reset(page); /* remove PageBuddy */
1843 		return;
1844 	}
1845 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1846 		bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1847 		bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1848 	}
1849 #ifdef CONFIG_MEMCG
1850 	if (unlikely(page->mem_cgroup))
1851 		bad_reason = "page still charged to cgroup";
1852 #endif
1853 	bad_page(page, bad_reason, bad_flags);
1854 }
1855 
1856 /*
1857  * This page is about to be returned from the page allocator
1858  */
1859 static inline int check_new_page(struct page *page)
1860 {
1861 	if (likely(page_expected_state(page,
1862 				PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1863 		return 0;
1864 
1865 	check_new_page_bad(page);
1866 	return 1;
1867 }
1868 
1869 static inline bool free_pages_prezeroed(void)
1870 {
1871 	return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1872 		page_poisoning_enabled();
1873 }
1874 
1875 #ifdef CONFIG_DEBUG_VM
1876 static bool check_pcp_refill(struct page *page)
1877 {
1878 	return false;
1879 }
1880 
1881 static bool check_new_pcp(struct page *page)
1882 {
1883 	return check_new_page(page);
1884 }
1885 #else
1886 static bool check_pcp_refill(struct page *page)
1887 {
1888 	return check_new_page(page);
1889 }
1890 static bool check_new_pcp(struct page *page)
1891 {
1892 	return false;
1893 }
1894 #endif /* CONFIG_DEBUG_VM */
1895 
1896 static bool check_new_pages(struct page *page, unsigned int order)
1897 {
1898 	int i;
1899 	for (i = 0; i < (1 << order); i++) {
1900 		struct page *p = page + i;
1901 
1902 		if (unlikely(check_new_page(p)))
1903 			return true;
1904 	}
1905 
1906 	return false;
1907 }
1908 
1909 inline void post_alloc_hook(struct page *page, unsigned int order,
1910 				gfp_t gfp_flags)
1911 {
1912 	set_page_private(page, 0);
1913 	set_page_refcounted(page);
1914 
1915 	arch_alloc_page(page, order);
1916 	kernel_map_pages(page, 1 << order, 1);
1917 	kernel_poison_pages(page, 1 << order, 1);
1918 	kasan_alloc_pages(page, order);
1919 	set_page_owner(page, order, gfp_flags);
1920 }
1921 
1922 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1923 							unsigned int alloc_flags)
1924 {
1925 	int i;
1926 
1927 	post_alloc_hook(page, order, gfp_flags);
1928 
1929 	if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
1930 		for (i = 0; i < (1 << order); i++)
1931 			clear_highpage(page + i);
1932 
1933 	if (order && (gfp_flags & __GFP_COMP))
1934 		prep_compound_page(page, order);
1935 
1936 	/*
1937 	 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1938 	 * allocate the page. The expectation is that the caller is taking
1939 	 * steps that will free more memory. The caller should avoid the page
1940 	 * being used for !PFMEMALLOC purposes.
1941 	 */
1942 	if (alloc_flags & ALLOC_NO_WATERMARKS)
1943 		set_page_pfmemalloc(page);
1944 	else
1945 		clear_page_pfmemalloc(page);
1946 }
1947 
1948 /*
1949  * Go through the free lists for the given migratetype and remove
1950  * the smallest available page from the freelists
1951  */
1952 static __always_inline
1953 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1954 						int migratetype)
1955 {
1956 	unsigned int current_order;
1957 	struct free_area *area;
1958 	struct page *page;
1959 
1960 	/* Find a page of the appropriate size in the preferred list */
1961 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1962 		area = &(zone->free_area[current_order]);
1963 		page = list_first_entry_or_null(&area->free_list[migratetype],
1964 							struct page, lru);
1965 		if (!page)
1966 			continue;
1967 		list_del(&page->lru);
1968 		rmv_page_order(page);
1969 		area->nr_free--;
1970 		expand(zone, page, order, current_order, area, migratetype);
1971 		set_pcppage_migratetype(page, migratetype);
1972 		return page;
1973 	}
1974 
1975 	return NULL;
1976 }
1977 
1978 
1979 /*
1980  * This array describes the order lists are fallen back to when
1981  * the free lists for the desirable migrate type are depleted
1982  */
1983 static int fallbacks[MIGRATE_TYPES][4] = {
1984 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
1985 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
1986 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1987 #ifdef CONFIG_CMA
1988 	[MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
1989 #endif
1990 #ifdef CONFIG_MEMORY_ISOLATION
1991 	[MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
1992 #endif
1993 };
1994 
1995 #ifdef CONFIG_CMA
1996 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1997 					unsigned int order)
1998 {
1999 	return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2000 }
2001 #else
2002 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2003 					unsigned int order) { return NULL; }
2004 #endif
2005 
2006 /*
2007  * Move the free pages in a range to the free lists of the requested type.
2008  * Note that start_page and end_pages are not aligned on a pageblock
2009  * boundary. If alignment is required, use move_freepages_block()
2010  */
2011 static int move_freepages(struct zone *zone,
2012 			  struct page *start_page, struct page *end_page,
2013 			  int migratetype, int *num_movable)
2014 {
2015 	struct page *page;
2016 	unsigned int order;
2017 	int pages_moved = 0;
2018 
2019 #ifndef CONFIG_HOLES_IN_ZONE
2020 	/*
2021 	 * page_zone is not safe to call in this context when
2022 	 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
2023 	 * anyway as we check zone boundaries in move_freepages_block().
2024 	 * Remove at a later date when no bug reports exist related to
2025 	 * grouping pages by mobility
2026 	 */
2027 	VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) &&
2028 	          pfn_valid(page_to_pfn(end_page)) &&
2029 	          page_zone(start_page) != page_zone(end_page));
2030 #endif
2031 	for (page = start_page; page <= end_page;) {
2032 		if (!pfn_valid_within(page_to_pfn(page))) {
2033 			page++;
2034 			continue;
2035 		}
2036 
2037 		/* Make sure we are not inadvertently changing nodes */
2038 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2039 
2040 		if (!PageBuddy(page)) {
2041 			/*
2042 			 * We assume that pages that could be isolated for
2043 			 * migration are movable. But we don't actually try
2044 			 * isolating, as that would be expensive.
2045 			 */
2046 			if (num_movable &&
2047 					(PageLRU(page) || __PageMovable(page)))
2048 				(*num_movable)++;
2049 
2050 			page++;
2051 			continue;
2052 		}
2053 
2054 		order = page_order(page);
2055 		list_move(&page->lru,
2056 			  &zone->free_area[order].free_list[migratetype]);
2057 		page += 1 << order;
2058 		pages_moved += 1 << order;
2059 	}
2060 
2061 	return pages_moved;
2062 }
2063 
2064 int move_freepages_block(struct zone *zone, struct page *page,
2065 				int migratetype, int *num_movable)
2066 {
2067 	unsigned long start_pfn, end_pfn;
2068 	struct page *start_page, *end_page;
2069 
2070 	if (num_movable)
2071 		*num_movable = 0;
2072 
2073 	start_pfn = page_to_pfn(page);
2074 	start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2075 	start_page = pfn_to_page(start_pfn);
2076 	end_page = start_page + pageblock_nr_pages - 1;
2077 	end_pfn = start_pfn + pageblock_nr_pages - 1;
2078 
2079 	/* Do not cross zone boundaries */
2080 	if (!zone_spans_pfn(zone, start_pfn))
2081 		start_page = page;
2082 	if (!zone_spans_pfn(zone, end_pfn))
2083 		return 0;
2084 
2085 	return move_freepages(zone, start_page, end_page, migratetype,
2086 								num_movable);
2087 }
2088 
2089 static void change_pageblock_range(struct page *pageblock_page,
2090 					int start_order, int migratetype)
2091 {
2092 	int nr_pageblocks = 1 << (start_order - pageblock_order);
2093 
2094 	while (nr_pageblocks--) {
2095 		set_pageblock_migratetype(pageblock_page, migratetype);
2096 		pageblock_page += pageblock_nr_pages;
2097 	}
2098 }
2099 
2100 /*
2101  * When we are falling back to another migratetype during allocation, try to
2102  * steal extra free pages from the same pageblocks to satisfy further
2103  * allocations, instead of polluting multiple pageblocks.
2104  *
2105  * If we are stealing a relatively large buddy page, it is likely there will
2106  * be more free pages in the pageblock, so try to steal them all. For
2107  * reclaimable and unmovable allocations, we steal regardless of page size,
2108  * as fragmentation caused by those allocations polluting movable pageblocks
2109  * is worse than movable allocations stealing from unmovable and reclaimable
2110  * pageblocks.
2111  */
2112 static bool can_steal_fallback(unsigned int order, int start_mt)
2113 {
2114 	/*
2115 	 * Leaving this order check is intended, although there is
2116 	 * relaxed order check in next check. The reason is that
2117 	 * we can actually steal whole pageblock if this condition met,
2118 	 * but, below check doesn't guarantee it and that is just heuristic
2119 	 * so could be changed anytime.
2120 	 */
2121 	if (order >= pageblock_order)
2122 		return true;
2123 
2124 	if (order >= pageblock_order / 2 ||
2125 		start_mt == MIGRATE_RECLAIMABLE ||
2126 		start_mt == MIGRATE_UNMOVABLE ||
2127 		page_group_by_mobility_disabled)
2128 		return true;
2129 
2130 	return false;
2131 }
2132 
2133 /*
2134  * This function implements actual steal behaviour. If order is large enough,
2135  * we can steal whole pageblock. If not, we first move freepages in this
2136  * pageblock to our migratetype and determine how many already-allocated pages
2137  * are there in the pageblock with a compatible migratetype. If at least half
2138  * of pages are free or compatible, we can change migratetype of the pageblock
2139  * itself, so pages freed in the future will be put on the correct free list.
2140  */
2141 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2142 					int start_type, bool whole_block)
2143 {
2144 	unsigned int current_order = page_order(page);
2145 	struct free_area *area;
2146 	int free_pages, movable_pages, alike_pages;
2147 	int old_block_type;
2148 
2149 	old_block_type = get_pageblock_migratetype(page);
2150 
2151 	/*
2152 	 * This can happen due to races and we want to prevent broken
2153 	 * highatomic accounting.
2154 	 */
2155 	if (is_migrate_highatomic(old_block_type))
2156 		goto single_page;
2157 
2158 	/* Take ownership for orders >= pageblock_order */
2159 	if (current_order >= pageblock_order) {
2160 		change_pageblock_range(page, current_order, start_type);
2161 		goto single_page;
2162 	}
2163 
2164 	/* We are not allowed to try stealing from the whole block */
2165 	if (!whole_block)
2166 		goto single_page;
2167 
2168 	free_pages = move_freepages_block(zone, page, start_type,
2169 						&movable_pages);
2170 	/*
2171 	 * Determine how many pages are compatible with our allocation.
2172 	 * For movable allocation, it's the number of movable pages which
2173 	 * we just obtained. For other types it's a bit more tricky.
2174 	 */
2175 	if (start_type == MIGRATE_MOVABLE) {
2176 		alike_pages = movable_pages;
2177 	} else {
2178 		/*
2179 		 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2180 		 * to MOVABLE pageblock, consider all non-movable pages as
2181 		 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2182 		 * vice versa, be conservative since we can't distinguish the
2183 		 * exact migratetype of non-movable pages.
2184 		 */
2185 		if (old_block_type == MIGRATE_MOVABLE)
2186 			alike_pages = pageblock_nr_pages
2187 						- (free_pages + movable_pages);
2188 		else
2189 			alike_pages = 0;
2190 	}
2191 
2192 	/* moving whole block can fail due to zone boundary conditions */
2193 	if (!free_pages)
2194 		goto single_page;
2195 
2196 	/*
2197 	 * If a sufficient number of pages in the block are either free or of
2198 	 * comparable migratability as our allocation, claim the whole block.
2199 	 */
2200 	if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2201 			page_group_by_mobility_disabled)
2202 		set_pageblock_migratetype(page, start_type);
2203 
2204 	return;
2205 
2206 single_page:
2207 	area = &zone->free_area[current_order];
2208 	list_move(&page->lru, &area->free_list[start_type]);
2209 }
2210 
2211 /*
2212  * Check whether there is a suitable fallback freepage with requested order.
2213  * If only_stealable is true, this function returns fallback_mt only if
2214  * we can steal other freepages all together. This would help to reduce
2215  * fragmentation due to mixed migratetype pages in one pageblock.
2216  */
2217 int find_suitable_fallback(struct free_area *area, unsigned int order,
2218 			int migratetype, bool only_stealable, bool *can_steal)
2219 {
2220 	int i;
2221 	int fallback_mt;
2222 
2223 	if (area->nr_free == 0)
2224 		return -1;
2225 
2226 	*can_steal = false;
2227 	for (i = 0;; i++) {
2228 		fallback_mt = fallbacks[migratetype][i];
2229 		if (fallback_mt == MIGRATE_TYPES)
2230 			break;
2231 
2232 		if (list_empty(&area->free_list[fallback_mt]))
2233 			continue;
2234 
2235 		if (can_steal_fallback(order, migratetype))
2236 			*can_steal = true;
2237 
2238 		if (!only_stealable)
2239 			return fallback_mt;
2240 
2241 		if (*can_steal)
2242 			return fallback_mt;
2243 	}
2244 
2245 	return -1;
2246 }
2247 
2248 /*
2249  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2250  * there are no empty page blocks that contain a page with a suitable order
2251  */
2252 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2253 				unsigned int alloc_order)
2254 {
2255 	int mt;
2256 	unsigned long max_managed, flags;
2257 
2258 	/*
2259 	 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2260 	 * Check is race-prone but harmless.
2261 	 */
2262 	max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2263 	if (zone->nr_reserved_highatomic >= max_managed)
2264 		return;
2265 
2266 	spin_lock_irqsave(&zone->lock, flags);
2267 
2268 	/* Recheck the nr_reserved_highatomic limit under the lock */
2269 	if (zone->nr_reserved_highatomic >= max_managed)
2270 		goto out_unlock;
2271 
2272 	/* Yoink! */
2273 	mt = get_pageblock_migratetype(page);
2274 	if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2275 	    && !is_migrate_cma(mt)) {
2276 		zone->nr_reserved_highatomic += pageblock_nr_pages;
2277 		set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2278 		move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2279 	}
2280 
2281 out_unlock:
2282 	spin_unlock_irqrestore(&zone->lock, flags);
2283 }
2284 
2285 /*
2286  * Used when an allocation is about to fail under memory pressure. This
2287  * potentially hurts the reliability of high-order allocations when under
2288  * intense memory pressure but failed atomic allocations should be easier
2289  * to recover from than an OOM.
2290  *
2291  * If @force is true, try to unreserve a pageblock even though highatomic
2292  * pageblock is exhausted.
2293  */
2294 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2295 						bool force)
2296 {
2297 	struct zonelist *zonelist = ac->zonelist;
2298 	unsigned long flags;
2299 	struct zoneref *z;
2300 	struct zone *zone;
2301 	struct page *page;
2302 	int order;
2303 	bool ret;
2304 
2305 	for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2306 								ac->nodemask) {
2307 		/*
2308 		 * Preserve at least one pageblock unless memory pressure
2309 		 * is really high.
2310 		 */
2311 		if (!force && zone->nr_reserved_highatomic <=
2312 					pageblock_nr_pages)
2313 			continue;
2314 
2315 		spin_lock_irqsave(&zone->lock, flags);
2316 		for (order = 0; order < MAX_ORDER; order++) {
2317 			struct free_area *area = &(zone->free_area[order]);
2318 
2319 			page = list_first_entry_or_null(
2320 					&area->free_list[MIGRATE_HIGHATOMIC],
2321 					struct page, lru);
2322 			if (!page)
2323 				continue;
2324 
2325 			/*
2326 			 * In page freeing path, migratetype change is racy so
2327 			 * we can counter several free pages in a pageblock
2328 			 * in this loop althoug we changed the pageblock type
2329 			 * from highatomic to ac->migratetype. So we should
2330 			 * adjust the count once.
2331 			 */
2332 			if (is_migrate_highatomic_page(page)) {
2333 				/*
2334 				 * It should never happen but changes to
2335 				 * locking could inadvertently allow a per-cpu
2336 				 * drain to add pages to MIGRATE_HIGHATOMIC
2337 				 * while unreserving so be safe and watch for
2338 				 * underflows.
2339 				 */
2340 				zone->nr_reserved_highatomic -= min(
2341 						pageblock_nr_pages,
2342 						zone->nr_reserved_highatomic);
2343 			}
2344 
2345 			/*
2346 			 * Convert to ac->migratetype and avoid the normal
2347 			 * pageblock stealing heuristics. Minimally, the caller
2348 			 * is doing the work and needs the pages. More
2349 			 * importantly, if the block was always converted to
2350 			 * MIGRATE_UNMOVABLE or another type then the number
2351 			 * of pageblocks that cannot be completely freed
2352 			 * may increase.
2353 			 */
2354 			set_pageblock_migratetype(page, ac->migratetype);
2355 			ret = move_freepages_block(zone, page, ac->migratetype,
2356 									NULL);
2357 			if (ret) {
2358 				spin_unlock_irqrestore(&zone->lock, flags);
2359 				return ret;
2360 			}
2361 		}
2362 		spin_unlock_irqrestore(&zone->lock, flags);
2363 	}
2364 
2365 	return false;
2366 }
2367 
2368 /*
2369  * Try finding a free buddy page on the fallback list and put it on the free
2370  * list of requested migratetype, possibly along with other pages from the same
2371  * block, depending on fragmentation avoidance heuristics. Returns true if
2372  * fallback was found so that __rmqueue_smallest() can grab it.
2373  *
2374  * The use of signed ints for order and current_order is a deliberate
2375  * deviation from the rest of this file, to make the for loop
2376  * condition simpler.
2377  */
2378 static __always_inline bool
2379 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
2380 {
2381 	struct free_area *area;
2382 	int current_order;
2383 	struct page *page;
2384 	int fallback_mt;
2385 	bool can_steal;
2386 
2387 	/*
2388 	 * Find the largest available free page in the other list. This roughly
2389 	 * approximates finding the pageblock with the most free pages, which
2390 	 * would be too costly to do exactly.
2391 	 */
2392 	for (current_order = MAX_ORDER - 1; current_order >= order;
2393 				--current_order) {
2394 		area = &(zone->free_area[current_order]);
2395 		fallback_mt = find_suitable_fallback(area, current_order,
2396 				start_migratetype, false, &can_steal);
2397 		if (fallback_mt == -1)
2398 			continue;
2399 
2400 		/*
2401 		 * We cannot steal all free pages from the pageblock and the
2402 		 * requested migratetype is movable. In that case it's better to
2403 		 * steal and split the smallest available page instead of the
2404 		 * largest available page, because even if the next movable
2405 		 * allocation falls back into a different pageblock than this
2406 		 * one, it won't cause permanent fragmentation.
2407 		 */
2408 		if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2409 					&& current_order > order)
2410 			goto find_smallest;
2411 
2412 		goto do_steal;
2413 	}
2414 
2415 	return false;
2416 
2417 find_smallest:
2418 	for (current_order = order; current_order < MAX_ORDER;
2419 							current_order++) {
2420 		area = &(zone->free_area[current_order]);
2421 		fallback_mt = find_suitable_fallback(area, current_order,
2422 				start_migratetype, false, &can_steal);
2423 		if (fallback_mt != -1)
2424 			break;
2425 	}
2426 
2427 	/*
2428 	 * This should not happen - we already found a suitable fallback
2429 	 * when looking for the largest page.
2430 	 */
2431 	VM_BUG_ON(current_order == MAX_ORDER);
2432 
2433 do_steal:
2434 	page = list_first_entry(&area->free_list[fallback_mt],
2435 							struct page, lru);
2436 
2437 	steal_suitable_fallback(zone, page, start_migratetype, can_steal);
2438 
2439 	trace_mm_page_alloc_extfrag(page, order, current_order,
2440 		start_migratetype, fallback_mt);
2441 
2442 	return true;
2443 
2444 }
2445 
2446 /*
2447  * Do the hard work of removing an element from the buddy allocator.
2448  * Call me with the zone->lock already held.
2449  */
2450 static __always_inline struct page *
2451 __rmqueue(struct zone *zone, unsigned int order, int migratetype)
2452 {
2453 	struct page *page;
2454 
2455 retry:
2456 	page = __rmqueue_smallest(zone, order, migratetype);
2457 	if (unlikely(!page)) {
2458 		if (migratetype == MIGRATE_MOVABLE)
2459 			page = __rmqueue_cma_fallback(zone, order);
2460 
2461 		if (!page && __rmqueue_fallback(zone, order, migratetype))
2462 			goto retry;
2463 	}
2464 
2465 	trace_mm_page_alloc_zone_locked(page, order, migratetype);
2466 	return page;
2467 }
2468 
2469 /*
2470  * Obtain a specified number of elements from the buddy allocator, all under
2471  * a single hold of the lock, for efficiency.  Add them to the supplied list.
2472  * Returns the number of new pages which were placed at *list.
2473  */
2474 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2475 			unsigned long count, struct list_head *list,
2476 			int migratetype)
2477 {
2478 	int i, alloced = 0;
2479 
2480 	spin_lock(&zone->lock);
2481 	for (i = 0; i < count; ++i) {
2482 		struct page *page = __rmqueue(zone, order, migratetype);
2483 		if (unlikely(page == NULL))
2484 			break;
2485 
2486 		if (unlikely(check_pcp_refill(page)))
2487 			continue;
2488 
2489 		/*
2490 		 * Split buddy pages returned by expand() are received here in
2491 		 * physical page order. The page is added to the tail of
2492 		 * caller's list. From the callers perspective, the linked list
2493 		 * is ordered by page number under some conditions. This is
2494 		 * useful for IO devices that can forward direction from the
2495 		 * head, thus also in the physical page order. This is useful
2496 		 * for IO devices that can merge IO requests if the physical
2497 		 * pages are ordered properly.
2498 		 */
2499 		list_add_tail(&page->lru, list);
2500 		alloced++;
2501 		if (is_migrate_cma(get_pcppage_migratetype(page)))
2502 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2503 					      -(1 << order));
2504 	}
2505 
2506 	/*
2507 	 * i pages were removed from the buddy list even if some leak due
2508 	 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2509 	 * on i. Do not confuse with 'alloced' which is the number of
2510 	 * pages added to the pcp list.
2511 	 */
2512 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2513 	spin_unlock(&zone->lock);
2514 	return alloced;
2515 }
2516 
2517 #ifdef CONFIG_NUMA
2518 /*
2519  * Called from the vmstat counter updater to drain pagesets of this
2520  * currently executing processor on remote nodes after they have
2521  * expired.
2522  *
2523  * Note that this function must be called with the thread pinned to
2524  * a single processor.
2525  */
2526 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2527 {
2528 	unsigned long flags;
2529 	int to_drain, batch;
2530 
2531 	local_irq_save(flags);
2532 	batch = READ_ONCE(pcp->batch);
2533 	to_drain = min(pcp->count, batch);
2534 	if (to_drain > 0)
2535 		free_pcppages_bulk(zone, to_drain, pcp);
2536 	local_irq_restore(flags);
2537 }
2538 #endif
2539 
2540 /*
2541  * Drain pcplists of the indicated processor and zone.
2542  *
2543  * The processor must either be the current processor and the
2544  * thread pinned to the current processor or a processor that
2545  * is not online.
2546  */
2547 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2548 {
2549 	unsigned long flags;
2550 	struct per_cpu_pageset *pset;
2551 	struct per_cpu_pages *pcp;
2552 
2553 	local_irq_save(flags);
2554 	pset = per_cpu_ptr(zone->pageset, cpu);
2555 
2556 	pcp = &pset->pcp;
2557 	if (pcp->count)
2558 		free_pcppages_bulk(zone, pcp->count, pcp);
2559 	local_irq_restore(flags);
2560 }
2561 
2562 /*
2563  * Drain pcplists of all zones on the indicated processor.
2564  *
2565  * The processor must either be the current processor and the
2566  * thread pinned to the current processor or a processor that
2567  * is not online.
2568  */
2569 static void drain_pages(unsigned int cpu)
2570 {
2571 	struct zone *zone;
2572 
2573 	for_each_populated_zone(zone) {
2574 		drain_pages_zone(cpu, zone);
2575 	}
2576 }
2577 
2578 /*
2579  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2580  *
2581  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2582  * the single zone's pages.
2583  */
2584 void drain_local_pages(struct zone *zone)
2585 {
2586 	int cpu = smp_processor_id();
2587 
2588 	if (zone)
2589 		drain_pages_zone(cpu, zone);
2590 	else
2591 		drain_pages(cpu);
2592 }
2593 
2594 static void drain_local_pages_wq(struct work_struct *work)
2595 {
2596 	/*
2597 	 * drain_all_pages doesn't use proper cpu hotplug protection so
2598 	 * we can race with cpu offline when the WQ can move this from
2599 	 * a cpu pinned worker to an unbound one. We can operate on a different
2600 	 * cpu which is allright but we also have to make sure to not move to
2601 	 * a different one.
2602 	 */
2603 	preempt_disable();
2604 	drain_local_pages(NULL);
2605 	preempt_enable();
2606 }
2607 
2608 /*
2609  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2610  *
2611  * When zone parameter is non-NULL, spill just the single zone's pages.
2612  *
2613  * Note that this can be extremely slow as the draining happens in a workqueue.
2614  */
2615 void drain_all_pages(struct zone *zone)
2616 {
2617 	int cpu;
2618 
2619 	/*
2620 	 * Allocate in the BSS so we wont require allocation in
2621 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2622 	 */
2623 	static cpumask_t cpus_with_pcps;
2624 
2625 	/*
2626 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
2627 	 * initialized.
2628 	 */
2629 	if (WARN_ON_ONCE(!mm_percpu_wq))
2630 		return;
2631 
2632 	/*
2633 	 * Do not drain if one is already in progress unless it's specific to
2634 	 * a zone. Such callers are primarily CMA and memory hotplug and need
2635 	 * the drain to be complete when the call returns.
2636 	 */
2637 	if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2638 		if (!zone)
2639 			return;
2640 		mutex_lock(&pcpu_drain_mutex);
2641 	}
2642 
2643 	/*
2644 	 * We don't care about racing with CPU hotplug event
2645 	 * as offline notification will cause the notified
2646 	 * cpu to drain that CPU pcps and on_each_cpu_mask
2647 	 * disables preemption as part of its processing
2648 	 */
2649 	for_each_online_cpu(cpu) {
2650 		struct per_cpu_pageset *pcp;
2651 		struct zone *z;
2652 		bool has_pcps = false;
2653 
2654 		if (zone) {
2655 			pcp = per_cpu_ptr(zone->pageset, cpu);
2656 			if (pcp->pcp.count)
2657 				has_pcps = true;
2658 		} else {
2659 			for_each_populated_zone(z) {
2660 				pcp = per_cpu_ptr(z->pageset, cpu);
2661 				if (pcp->pcp.count) {
2662 					has_pcps = true;
2663 					break;
2664 				}
2665 			}
2666 		}
2667 
2668 		if (has_pcps)
2669 			cpumask_set_cpu(cpu, &cpus_with_pcps);
2670 		else
2671 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
2672 	}
2673 
2674 	for_each_cpu(cpu, &cpus_with_pcps) {
2675 		struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu);
2676 		INIT_WORK(work, drain_local_pages_wq);
2677 		queue_work_on(cpu, mm_percpu_wq, work);
2678 	}
2679 	for_each_cpu(cpu, &cpus_with_pcps)
2680 		flush_work(per_cpu_ptr(&pcpu_drain, cpu));
2681 
2682 	mutex_unlock(&pcpu_drain_mutex);
2683 }
2684 
2685 #ifdef CONFIG_HIBERNATION
2686 
2687 /*
2688  * Touch the watchdog for every WD_PAGE_COUNT pages.
2689  */
2690 #define WD_PAGE_COUNT	(128*1024)
2691 
2692 void mark_free_pages(struct zone *zone)
2693 {
2694 	unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
2695 	unsigned long flags;
2696 	unsigned int order, t;
2697 	struct page *page;
2698 
2699 	if (zone_is_empty(zone))
2700 		return;
2701 
2702 	spin_lock_irqsave(&zone->lock, flags);
2703 
2704 	max_zone_pfn = zone_end_pfn(zone);
2705 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2706 		if (pfn_valid(pfn)) {
2707 			page = pfn_to_page(pfn);
2708 
2709 			if (!--page_count) {
2710 				touch_nmi_watchdog();
2711 				page_count = WD_PAGE_COUNT;
2712 			}
2713 
2714 			if (page_zone(page) != zone)
2715 				continue;
2716 
2717 			if (!swsusp_page_is_forbidden(page))
2718 				swsusp_unset_page_free(page);
2719 		}
2720 
2721 	for_each_migratetype_order(order, t) {
2722 		list_for_each_entry(page,
2723 				&zone->free_area[order].free_list[t], lru) {
2724 			unsigned long i;
2725 
2726 			pfn = page_to_pfn(page);
2727 			for (i = 0; i < (1UL << order); i++) {
2728 				if (!--page_count) {
2729 					touch_nmi_watchdog();
2730 					page_count = WD_PAGE_COUNT;
2731 				}
2732 				swsusp_set_page_free(pfn_to_page(pfn + i));
2733 			}
2734 		}
2735 	}
2736 	spin_unlock_irqrestore(&zone->lock, flags);
2737 }
2738 #endif /* CONFIG_PM */
2739 
2740 static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
2741 {
2742 	int migratetype;
2743 
2744 	if (!free_pcp_prepare(page))
2745 		return false;
2746 
2747 	migratetype = get_pfnblock_migratetype(page, pfn);
2748 	set_pcppage_migratetype(page, migratetype);
2749 	return true;
2750 }
2751 
2752 static void free_unref_page_commit(struct page *page, unsigned long pfn)
2753 {
2754 	struct zone *zone = page_zone(page);
2755 	struct per_cpu_pages *pcp;
2756 	int migratetype;
2757 
2758 	migratetype = get_pcppage_migratetype(page);
2759 	__count_vm_event(PGFREE);
2760 
2761 	/*
2762 	 * We only track unmovable, reclaimable and movable on pcp lists.
2763 	 * Free ISOLATE pages back to the allocator because they are being
2764 	 * offlined but treat HIGHATOMIC as movable pages so we can get those
2765 	 * areas back if necessary. Otherwise, we may have to free
2766 	 * excessively into the page allocator
2767 	 */
2768 	if (migratetype >= MIGRATE_PCPTYPES) {
2769 		if (unlikely(is_migrate_isolate(migratetype))) {
2770 			free_one_page(zone, page, pfn, 0, migratetype);
2771 			return;
2772 		}
2773 		migratetype = MIGRATE_MOVABLE;
2774 	}
2775 
2776 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
2777 	list_add(&page->lru, &pcp->lists[migratetype]);
2778 	pcp->count++;
2779 	if (pcp->count >= pcp->high) {
2780 		unsigned long batch = READ_ONCE(pcp->batch);
2781 		free_pcppages_bulk(zone, batch, pcp);
2782 	}
2783 }
2784 
2785 /*
2786  * Free a 0-order page
2787  */
2788 void free_unref_page(struct page *page)
2789 {
2790 	unsigned long flags;
2791 	unsigned long pfn = page_to_pfn(page);
2792 
2793 	if (!free_unref_page_prepare(page, pfn))
2794 		return;
2795 
2796 	local_irq_save(flags);
2797 	free_unref_page_commit(page, pfn);
2798 	local_irq_restore(flags);
2799 }
2800 
2801 /*
2802  * Free a list of 0-order pages
2803  */
2804 void free_unref_page_list(struct list_head *list)
2805 {
2806 	struct page *page, *next;
2807 	unsigned long flags, pfn;
2808 	int batch_count = 0;
2809 
2810 	/* Prepare pages for freeing */
2811 	list_for_each_entry_safe(page, next, list, lru) {
2812 		pfn = page_to_pfn(page);
2813 		if (!free_unref_page_prepare(page, pfn))
2814 			list_del(&page->lru);
2815 		set_page_private(page, pfn);
2816 	}
2817 
2818 	local_irq_save(flags);
2819 	list_for_each_entry_safe(page, next, list, lru) {
2820 		unsigned long pfn = page_private(page);
2821 
2822 		set_page_private(page, 0);
2823 		trace_mm_page_free_batched(page);
2824 		free_unref_page_commit(page, pfn);
2825 
2826 		/*
2827 		 * Guard against excessive IRQ disabled times when we get
2828 		 * a large list of pages to free.
2829 		 */
2830 		if (++batch_count == SWAP_CLUSTER_MAX) {
2831 			local_irq_restore(flags);
2832 			batch_count = 0;
2833 			local_irq_save(flags);
2834 		}
2835 	}
2836 	local_irq_restore(flags);
2837 }
2838 
2839 /*
2840  * split_page takes a non-compound higher-order page, and splits it into
2841  * n (1<<order) sub-pages: page[0..n]
2842  * Each sub-page must be freed individually.
2843  *
2844  * Note: this is probably too low level an operation for use in drivers.
2845  * Please consult with lkml before using this in your driver.
2846  */
2847 void split_page(struct page *page, unsigned int order)
2848 {
2849 	int i;
2850 
2851 	VM_BUG_ON_PAGE(PageCompound(page), page);
2852 	VM_BUG_ON_PAGE(!page_count(page), page);
2853 
2854 	for (i = 1; i < (1 << order); i++)
2855 		set_page_refcounted(page + i);
2856 	split_page_owner(page, order);
2857 }
2858 EXPORT_SYMBOL_GPL(split_page);
2859 
2860 int __isolate_free_page(struct page *page, unsigned int order)
2861 {
2862 	unsigned long watermark;
2863 	struct zone *zone;
2864 	int mt;
2865 
2866 	BUG_ON(!PageBuddy(page));
2867 
2868 	zone = page_zone(page);
2869 	mt = get_pageblock_migratetype(page);
2870 
2871 	if (!is_migrate_isolate(mt)) {
2872 		/*
2873 		 * Obey watermarks as if the page was being allocated. We can
2874 		 * emulate a high-order watermark check with a raised order-0
2875 		 * watermark, because we already know our high-order page
2876 		 * exists.
2877 		 */
2878 		watermark = min_wmark_pages(zone) + (1UL << order);
2879 		if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
2880 			return 0;
2881 
2882 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
2883 	}
2884 
2885 	/* Remove page from free list */
2886 	list_del(&page->lru);
2887 	zone->free_area[order].nr_free--;
2888 	rmv_page_order(page);
2889 
2890 	/*
2891 	 * Set the pageblock if the isolated page is at least half of a
2892 	 * pageblock
2893 	 */
2894 	if (order >= pageblock_order - 1) {
2895 		struct page *endpage = page + (1 << order) - 1;
2896 		for (; page < endpage; page += pageblock_nr_pages) {
2897 			int mt = get_pageblock_migratetype(page);
2898 			if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
2899 			    && !is_migrate_highatomic(mt))
2900 				set_pageblock_migratetype(page,
2901 							  MIGRATE_MOVABLE);
2902 		}
2903 	}
2904 
2905 
2906 	return 1UL << order;
2907 }
2908 
2909 /*
2910  * Update NUMA hit/miss statistics
2911  *
2912  * Must be called with interrupts disabled.
2913  */
2914 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
2915 {
2916 #ifdef CONFIG_NUMA
2917 	enum numa_stat_item local_stat = NUMA_LOCAL;
2918 
2919 	/* skip numa counters update if numa stats is disabled */
2920 	if (!static_branch_likely(&vm_numa_stat_key))
2921 		return;
2922 
2923 	if (zone_to_nid(z) != numa_node_id())
2924 		local_stat = NUMA_OTHER;
2925 
2926 	if (zone_to_nid(z) == zone_to_nid(preferred_zone))
2927 		__inc_numa_state(z, NUMA_HIT);
2928 	else {
2929 		__inc_numa_state(z, NUMA_MISS);
2930 		__inc_numa_state(preferred_zone, NUMA_FOREIGN);
2931 	}
2932 	__inc_numa_state(z, local_stat);
2933 #endif
2934 }
2935 
2936 /* Remove page from the per-cpu list, caller must protect the list */
2937 static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
2938 			struct per_cpu_pages *pcp,
2939 			struct list_head *list)
2940 {
2941 	struct page *page;
2942 
2943 	do {
2944 		if (list_empty(list)) {
2945 			pcp->count += rmqueue_bulk(zone, 0,
2946 					pcp->batch, list,
2947 					migratetype);
2948 			if (unlikely(list_empty(list)))
2949 				return NULL;
2950 		}
2951 
2952 		page = list_first_entry(list, struct page, lru);
2953 		list_del(&page->lru);
2954 		pcp->count--;
2955 	} while (check_new_pcp(page));
2956 
2957 	return page;
2958 }
2959 
2960 /* Lock and remove page from the per-cpu list */
2961 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2962 			struct zone *zone, unsigned int order,
2963 			gfp_t gfp_flags, int migratetype)
2964 {
2965 	struct per_cpu_pages *pcp;
2966 	struct list_head *list;
2967 	struct page *page;
2968 	unsigned long flags;
2969 
2970 	local_irq_save(flags);
2971 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
2972 	list = &pcp->lists[migratetype];
2973 	page = __rmqueue_pcplist(zone,  migratetype, pcp, list);
2974 	if (page) {
2975 		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2976 		zone_statistics(preferred_zone, zone);
2977 	}
2978 	local_irq_restore(flags);
2979 	return page;
2980 }
2981 
2982 /*
2983  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2984  */
2985 static inline
2986 struct page *rmqueue(struct zone *preferred_zone,
2987 			struct zone *zone, unsigned int order,
2988 			gfp_t gfp_flags, unsigned int alloc_flags,
2989 			int migratetype)
2990 {
2991 	unsigned long flags;
2992 	struct page *page;
2993 
2994 	if (likely(order == 0)) {
2995 		page = rmqueue_pcplist(preferred_zone, zone, order,
2996 				gfp_flags, migratetype);
2997 		goto out;
2998 	}
2999 
3000 	/*
3001 	 * We most definitely don't want callers attempting to
3002 	 * allocate greater than order-1 page units with __GFP_NOFAIL.
3003 	 */
3004 	WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3005 	spin_lock_irqsave(&zone->lock, flags);
3006 
3007 	do {
3008 		page = NULL;
3009 		if (alloc_flags & ALLOC_HARDER) {
3010 			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3011 			if (page)
3012 				trace_mm_page_alloc_zone_locked(page, order, migratetype);
3013 		}
3014 		if (!page)
3015 			page = __rmqueue(zone, order, migratetype);
3016 	} while (page && check_new_pages(page, order));
3017 	spin_unlock(&zone->lock);
3018 	if (!page)
3019 		goto failed;
3020 	__mod_zone_freepage_state(zone, -(1 << order),
3021 				  get_pcppage_migratetype(page));
3022 
3023 	__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3024 	zone_statistics(preferred_zone, zone);
3025 	local_irq_restore(flags);
3026 
3027 out:
3028 	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3029 	return page;
3030 
3031 failed:
3032 	local_irq_restore(flags);
3033 	return NULL;
3034 }
3035 
3036 #ifdef CONFIG_FAIL_PAGE_ALLOC
3037 
3038 static struct {
3039 	struct fault_attr attr;
3040 
3041 	bool ignore_gfp_highmem;
3042 	bool ignore_gfp_reclaim;
3043 	u32 min_order;
3044 } fail_page_alloc = {
3045 	.attr = FAULT_ATTR_INITIALIZER,
3046 	.ignore_gfp_reclaim = true,
3047 	.ignore_gfp_highmem = true,
3048 	.min_order = 1,
3049 };
3050 
3051 static int __init setup_fail_page_alloc(char *str)
3052 {
3053 	return setup_fault_attr(&fail_page_alloc.attr, str);
3054 }
3055 __setup("fail_page_alloc=", setup_fail_page_alloc);
3056 
3057 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3058 {
3059 	if (order < fail_page_alloc.min_order)
3060 		return false;
3061 	if (gfp_mask & __GFP_NOFAIL)
3062 		return false;
3063 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3064 		return false;
3065 	if (fail_page_alloc.ignore_gfp_reclaim &&
3066 			(gfp_mask & __GFP_DIRECT_RECLAIM))
3067 		return false;
3068 
3069 	return should_fail(&fail_page_alloc.attr, 1 << order);
3070 }
3071 
3072 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3073 
3074 static int __init fail_page_alloc_debugfs(void)
3075 {
3076 	umode_t mode = S_IFREG | 0600;
3077 	struct dentry *dir;
3078 
3079 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3080 					&fail_page_alloc.attr);
3081 	if (IS_ERR(dir))
3082 		return PTR_ERR(dir);
3083 
3084 	if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
3085 				&fail_page_alloc.ignore_gfp_reclaim))
3086 		goto fail;
3087 	if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3088 				&fail_page_alloc.ignore_gfp_highmem))
3089 		goto fail;
3090 	if (!debugfs_create_u32("min-order", mode, dir,
3091 				&fail_page_alloc.min_order))
3092 		goto fail;
3093 
3094 	return 0;
3095 fail:
3096 	debugfs_remove_recursive(dir);
3097 
3098 	return -ENOMEM;
3099 }
3100 
3101 late_initcall(fail_page_alloc_debugfs);
3102 
3103 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3104 
3105 #else /* CONFIG_FAIL_PAGE_ALLOC */
3106 
3107 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3108 {
3109 	return false;
3110 }
3111 
3112 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3113 
3114 /*
3115  * Return true if free base pages are above 'mark'. For high-order checks it
3116  * will return true of the order-0 watermark is reached and there is at least
3117  * one free page of a suitable size. Checking now avoids taking the zone lock
3118  * to check in the allocation paths if no pages are free.
3119  */
3120 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3121 			 int classzone_idx, unsigned int alloc_flags,
3122 			 long free_pages)
3123 {
3124 	long min = mark;
3125 	int o;
3126 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3127 
3128 	/* free_pages may go negative - that's OK */
3129 	free_pages -= (1 << order) - 1;
3130 
3131 	if (alloc_flags & ALLOC_HIGH)
3132 		min -= min / 2;
3133 
3134 	/*
3135 	 * If the caller does not have rights to ALLOC_HARDER then subtract
3136 	 * the high-atomic reserves. This will over-estimate the size of the
3137 	 * atomic reserve but it avoids a search.
3138 	 */
3139 	if (likely(!alloc_harder)) {
3140 		free_pages -= z->nr_reserved_highatomic;
3141 	} else {
3142 		/*
3143 		 * OOM victims can try even harder than normal ALLOC_HARDER
3144 		 * users on the grounds that it's definitely going to be in
3145 		 * the exit path shortly and free memory. Any allocation it
3146 		 * makes during the free path will be small and short-lived.
3147 		 */
3148 		if (alloc_flags & ALLOC_OOM)
3149 			min -= min / 2;
3150 		else
3151 			min -= min / 4;
3152 	}
3153 
3154 
3155 #ifdef CONFIG_CMA
3156 	/* If allocation can't use CMA areas don't use free CMA pages */
3157 	if (!(alloc_flags & ALLOC_CMA))
3158 		free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
3159 #endif
3160 
3161 	/*
3162 	 * Check watermarks for an order-0 allocation request. If these
3163 	 * are not met, then a high-order request also cannot go ahead
3164 	 * even if a suitable page happened to be free.
3165 	 */
3166 	if (free_pages <= min + z->lowmem_reserve[classzone_idx])
3167 		return false;
3168 
3169 	/* If this is an order-0 request then the watermark is fine */
3170 	if (!order)
3171 		return true;
3172 
3173 	/* For a high-order request, check at least one suitable page is free */
3174 	for (o = order; o < MAX_ORDER; o++) {
3175 		struct free_area *area = &z->free_area[o];
3176 		int mt;
3177 
3178 		if (!area->nr_free)
3179 			continue;
3180 
3181 		for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3182 			if (!list_empty(&area->free_list[mt]))
3183 				return true;
3184 		}
3185 
3186 #ifdef CONFIG_CMA
3187 		if ((alloc_flags & ALLOC_CMA) &&
3188 		    !list_empty(&area->free_list[MIGRATE_CMA])) {
3189 			return true;
3190 		}
3191 #endif
3192 		if (alloc_harder &&
3193 			!list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
3194 			return true;
3195 	}
3196 	return false;
3197 }
3198 
3199 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3200 		      int classzone_idx, unsigned int alloc_flags)
3201 {
3202 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3203 					zone_page_state(z, NR_FREE_PAGES));
3204 }
3205 
3206 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3207 		unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3208 {
3209 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3210 	long cma_pages = 0;
3211 
3212 #ifdef CONFIG_CMA
3213 	/* If allocation can't use CMA areas don't use free CMA pages */
3214 	if (!(alloc_flags & ALLOC_CMA))
3215 		cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3216 #endif
3217 
3218 	/*
3219 	 * Fast check for order-0 only. If this fails then the reserves
3220 	 * need to be calculated. There is a corner case where the check
3221 	 * passes but only the high-order atomic reserve are free. If
3222 	 * the caller is !atomic then it'll uselessly search the free
3223 	 * list. That corner case is then slower but it is harmless.
3224 	 */
3225 	if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3226 		return true;
3227 
3228 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3229 					free_pages);
3230 }
3231 
3232 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3233 			unsigned long mark, int classzone_idx)
3234 {
3235 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3236 
3237 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3238 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3239 
3240 	return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
3241 								free_pages);
3242 }
3243 
3244 #ifdef CONFIG_NUMA
3245 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3246 {
3247 	return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3248 				RECLAIM_DISTANCE;
3249 }
3250 #else	/* CONFIG_NUMA */
3251 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3252 {
3253 	return true;
3254 }
3255 #endif	/* CONFIG_NUMA */
3256 
3257 /*
3258  * get_page_from_freelist goes through the zonelist trying to allocate
3259  * a page.
3260  */
3261 static struct page *
3262 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3263 						const struct alloc_context *ac)
3264 {
3265 	struct zoneref *z = ac->preferred_zoneref;
3266 	struct zone *zone;
3267 	struct pglist_data *last_pgdat_dirty_limit = NULL;
3268 
3269 	/*
3270 	 * Scan zonelist, looking for a zone with enough free.
3271 	 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3272 	 */
3273 	for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3274 								ac->nodemask) {
3275 		struct page *page;
3276 		unsigned long mark;
3277 
3278 		if (cpusets_enabled() &&
3279 			(alloc_flags & ALLOC_CPUSET) &&
3280 			!__cpuset_zone_allowed(zone, gfp_mask))
3281 				continue;
3282 		/*
3283 		 * When allocating a page cache page for writing, we
3284 		 * want to get it from a node that is within its dirty
3285 		 * limit, such that no single node holds more than its
3286 		 * proportional share of globally allowed dirty pages.
3287 		 * The dirty limits take into account the node's
3288 		 * lowmem reserves and high watermark so that kswapd
3289 		 * should be able to balance it without having to
3290 		 * write pages from its LRU list.
3291 		 *
3292 		 * XXX: For now, allow allocations to potentially
3293 		 * exceed the per-node dirty limit in the slowpath
3294 		 * (spread_dirty_pages unset) before going into reclaim,
3295 		 * which is important when on a NUMA setup the allowed
3296 		 * nodes are together not big enough to reach the
3297 		 * global limit.  The proper fix for these situations
3298 		 * will require awareness of nodes in the
3299 		 * dirty-throttling and the flusher threads.
3300 		 */
3301 		if (ac->spread_dirty_pages) {
3302 			if (last_pgdat_dirty_limit == zone->zone_pgdat)
3303 				continue;
3304 
3305 			if (!node_dirty_ok(zone->zone_pgdat)) {
3306 				last_pgdat_dirty_limit = zone->zone_pgdat;
3307 				continue;
3308 			}
3309 		}
3310 
3311 		mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
3312 		if (!zone_watermark_fast(zone, order, mark,
3313 				       ac_classzone_idx(ac), alloc_flags)) {
3314 			int ret;
3315 
3316 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3317 			/*
3318 			 * Watermark failed for this zone, but see if we can
3319 			 * grow this zone if it contains deferred pages.
3320 			 */
3321 			if (static_branch_unlikely(&deferred_pages)) {
3322 				if (_deferred_grow_zone(zone, order))
3323 					goto try_this_zone;
3324 			}
3325 #endif
3326 			/* Checked here to keep the fast path fast */
3327 			BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3328 			if (alloc_flags & ALLOC_NO_WATERMARKS)
3329 				goto try_this_zone;
3330 
3331 			if (node_reclaim_mode == 0 ||
3332 			    !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3333 				continue;
3334 
3335 			ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3336 			switch (ret) {
3337 			case NODE_RECLAIM_NOSCAN:
3338 				/* did not scan */
3339 				continue;
3340 			case NODE_RECLAIM_FULL:
3341 				/* scanned but unreclaimable */
3342 				continue;
3343 			default:
3344 				/* did we reclaim enough */
3345 				if (zone_watermark_ok(zone, order, mark,
3346 						ac_classzone_idx(ac), alloc_flags))
3347 					goto try_this_zone;
3348 
3349 				continue;
3350 			}
3351 		}
3352 
3353 try_this_zone:
3354 		page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3355 				gfp_mask, alloc_flags, ac->migratetype);
3356 		if (page) {
3357 			prep_new_page(page, order, gfp_mask, alloc_flags);
3358 
3359 			/*
3360 			 * If this is a high-order atomic allocation then check
3361 			 * if the pageblock should be reserved for the future
3362 			 */
3363 			if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3364 				reserve_highatomic_pageblock(page, zone, order);
3365 
3366 			return page;
3367 		} else {
3368 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3369 			/* Try again if zone has deferred pages */
3370 			if (static_branch_unlikely(&deferred_pages)) {
3371 				if (_deferred_grow_zone(zone, order))
3372 					goto try_this_zone;
3373 			}
3374 #endif
3375 		}
3376 	}
3377 
3378 	return NULL;
3379 }
3380 
3381 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3382 {
3383 	unsigned int filter = SHOW_MEM_FILTER_NODES;
3384 	static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
3385 
3386 	if (!__ratelimit(&show_mem_rs))
3387 		return;
3388 
3389 	/*
3390 	 * This documents exceptions given to allocations in certain
3391 	 * contexts that are allowed to allocate outside current's set
3392 	 * of allowed nodes.
3393 	 */
3394 	if (!(gfp_mask & __GFP_NOMEMALLOC))
3395 		if (tsk_is_oom_victim(current) ||
3396 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
3397 			filter &= ~SHOW_MEM_FILTER_NODES;
3398 	if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3399 		filter &= ~SHOW_MEM_FILTER_NODES;
3400 
3401 	show_mem(filter, nodemask);
3402 }
3403 
3404 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3405 {
3406 	struct va_format vaf;
3407 	va_list args;
3408 	static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3409 				      DEFAULT_RATELIMIT_BURST);
3410 
3411 	if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3412 		return;
3413 
3414 	va_start(args, fmt);
3415 	vaf.fmt = fmt;
3416 	vaf.va = &args;
3417 	pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl\n",
3418 			current->comm, &vaf, gfp_mask, &gfp_mask,
3419 			nodemask_pr_args(nodemask));
3420 	va_end(args);
3421 
3422 	cpuset_print_current_mems_allowed();
3423 
3424 	dump_stack();
3425 	warn_alloc_show_mem(gfp_mask, nodemask);
3426 }
3427 
3428 static inline struct page *
3429 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3430 			      unsigned int alloc_flags,
3431 			      const struct alloc_context *ac)
3432 {
3433 	struct page *page;
3434 
3435 	page = get_page_from_freelist(gfp_mask, order,
3436 			alloc_flags|ALLOC_CPUSET, ac);
3437 	/*
3438 	 * fallback to ignore cpuset restriction if our nodes
3439 	 * are depleted
3440 	 */
3441 	if (!page)
3442 		page = get_page_from_freelist(gfp_mask, order,
3443 				alloc_flags, ac);
3444 
3445 	return page;
3446 }
3447 
3448 static inline struct page *
3449 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3450 	const struct alloc_context *ac, unsigned long *did_some_progress)
3451 {
3452 	struct oom_control oc = {
3453 		.zonelist = ac->zonelist,
3454 		.nodemask = ac->nodemask,
3455 		.memcg = NULL,
3456 		.gfp_mask = gfp_mask,
3457 		.order = order,
3458 	};
3459 	struct page *page;
3460 
3461 	*did_some_progress = 0;
3462 
3463 	/*
3464 	 * Acquire the oom lock.  If that fails, somebody else is
3465 	 * making progress for us.
3466 	 */
3467 	if (!mutex_trylock(&oom_lock)) {
3468 		*did_some_progress = 1;
3469 		schedule_timeout_uninterruptible(1);
3470 		return NULL;
3471 	}
3472 
3473 	/*
3474 	 * Go through the zonelist yet one more time, keep very high watermark
3475 	 * here, this is only to catch a parallel oom killing, we must fail if
3476 	 * we're still under heavy pressure. But make sure that this reclaim
3477 	 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3478 	 * allocation which will never fail due to oom_lock already held.
3479 	 */
3480 	page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3481 				      ~__GFP_DIRECT_RECLAIM, order,
3482 				      ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3483 	if (page)
3484 		goto out;
3485 
3486 	/* Coredumps can quickly deplete all memory reserves */
3487 	if (current->flags & PF_DUMPCORE)
3488 		goto out;
3489 	/* The OOM killer will not help higher order allocs */
3490 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3491 		goto out;
3492 	/*
3493 	 * We have already exhausted all our reclaim opportunities without any
3494 	 * success so it is time to admit defeat. We will skip the OOM killer
3495 	 * because it is very likely that the caller has a more reasonable
3496 	 * fallback than shooting a random task.
3497 	 */
3498 	if (gfp_mask & __GFP_RETRY_MAYFAIL)
3499 		goto out;
3500 	/* The OOM killer does not needlessly kill tasks for lowmem */
3501 	if (ac->high_zoneidx < ZONE_NORMAL)
3502 		goto out;
3503 	if (pm_suspended_storage())
3504 		goto out;
3505 	/*
3506 	 * XXX: GFP_NOFS allocations should rather fail than rely on
3507 	 * other request to make a forward progress.
3508 	 * We are in an unfortunate situation where out_of_memory cannot
3509 	 * do much for this context but let's try it to at least get
3510 	 * access to memory reserved if the current task is killed (see
3511 	 * out_of_memory). Once filesystems are ready to handle allocation
3512 	 * failures more gracefully we should just bail out here.
3513 	 */
3514 
3515 	/* The OOM killer may not free memory on a specific node */
3516 	if (gfp_mask & __GFP_THISNODE)
3517 		goto out;
3518 
3519 	/* Exhausted what can be done so it's blame time */
3520 	if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3521 		*did_some_progress = 1;
3522 
3523 		/*
3524 		 * Help non-failing allocations by giving them access to memory
3525 		 * reserves
3526 		 */
3527 		if (gfp_mask & __GFP_NOFAIL)
3528 			page = __alloc_pages_cpuset_fallback(gfp_mask, order,
3529 					ALLOC_NO_WATERMARKS, ac);
3530 	}
3531 out:
3532 	mutex_unlock(&oom_lock);
3533 	return page;
3534 }
3535 
3536 /*
3537  * Maximum number of compaction retries wit a progress before OOM
3538  * killer is consider as the only way to move forward.
3539  */
3540 #define MAX_COMPACT_RETRIES 16
3541 
3542 #ifdef CONFIG_COMPACTION
3543 /* Try memory compaction for high-order allocations before reclaim */
3544 static struct page *
3545 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3546 		unsigned int alloc_flags, const struct alloc_context *ac,
3547 		enum compact_priority prio, enum compact_result *compact_result)
3548 {
3549 	struct page *page;
3550 	unsigned long pflags;
3551 	unsigned int noreclaim_flag;
3552 
3553 	if (!order)
3554 		return NULL;
3555 
3556 	psi_memstall_enter(&pflags);
3557 	noreclaim_flag = memalloc_noreclaim_save();
3558 
3559 	*compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3560 									prio);
3561 
3562 	memalloc_noreclaim_restore(noreclaim_flag);
3563 	psi_memstall_leave(&pflags);
3564 
3565 	if (*compact_result <= COMPACT_INACTIVE)
3566 		return NULL;
3567 
3568 	/*
3569 	 * At least in one zone compaction wasn't deferred or skipped, so let's
3570 	 * count a compaction stall
3571 	 */
3572 	count_vm_event(COMPACTSTALL);
3573 
3574 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3575 
3576 	if (page) {
3577 		struct zone *zone = page_zone(page);
3578 
3579 		zone->compact_blockskip_flush = false;
3580 		compaction_defer_reset(zone, order, true);
3581 		count_vm_event(COMPACTSUCCESS);
3582 		return page;
3583 	}
3584 
3585 	/*
3586 	 * It's bad if compaction run occurs and fails. The most likely reason
3587 	 * is that pages exist, but not enough to satisfy watermarks.
3588 	 */
3589 	count_vm_event(COMPACTFAIL);
3590 
3591 	cond_resched();
3592 
3593 	return NULL;
3594 }
3595 
3596 static inline bool
3597 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3598 		     enum compact_result compact_result,
3599 		     enum compact_priority *compact_priority,
3600 		     int *compaction_retries)
3601 {
3602 	int max_retries = MAX_COMPACT_RETRIES;
3603 	int min_priority;
3604 	bool ret = false;
3605 	int retries = *compaction_retries;
3606 	enum compact_priority priority = *compact_priority;
3607 
3608 	if (!order)
3609 		return false;
3610 
3611 	if (compaction_made_progress(compact_result))
3612 		(*compaction_retries)++;
3613 
3614 	/*
3615 	 * compaction considers all the zone as desperately out of memory
3616 	 * so it doesn't really make much sense to retry except when the
3617 	 * failure could be caused by insufficient priority
3618 	 */
3619 	if (compaction_failed(compact_result))
3620 		goto check_priority;
3621 
3622 	/*
3623 	 * make sure the compaction wasn't deferred or didn't bail out early
3624 	 * due to locks contention before we declare that we should give up.
3625 	 * But do not retry if the given zonelist is not suitable for
3626 	 * compaction.
3627 	 */
3628 	if (compaction_withdrawn(compact_result)) {
3629 		ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3630 		goto out;
3631 	}
3632 
3633 	/*
3634 	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
3635 	 * costly ones because they are de facto nofail and invoke OOM
3636 	 * killer to move on while costly can fail and users are ready
3637 	 * to cope with that. 1/4 retries is rather arbitrary but we
3638 	 * would need much more detailed feedback from compaction to
3639 	 * make a better decision.
3640 	 */
3641 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3642 		max_retries /= 4;
3643 	if (*compaction_retries <= max_retries) {
3644 		ret = true;
3645 		goto out;
3646 	}
3647 
3648 	/*
3649 	 * Make sure there are attempts at the highest priority if we exhausted
3650 	 * all retries or failed at the lower priorities.
3651 	 */
3652 check_priority:
3653 	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3654 			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3655 
3656 	if (*compact_priority > min_priority) {
3657 		(*compact_priority)--;
3658 		*compaction_retries = 0;
3659 		ret = true;
3660 	}
3661 out:
3662 	trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3663 	return ret;
3664 }
3665 #else
3666 static inline struct page *
3667 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3668 		unsigned int alloc_flags, const struct alloc_context *ac,
3669 		enum compact_priority prio, enum compact_result *compact_result)
3670 {
3671 	*compact_result = COMPACT_SKIPPED;
3672 	return NULL;
3673 }
3674 
3675 static inline bool
3676 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3677 		     enum compact_result compact_result,
3678 		     enum compact_priority *compact_priority,
3679 		     int *compaction_retries)
3680 {
3681 	struct zone *zone;
3682 	struct zoneref *z;
3683 
3684 	if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3685 		return false;
3686 
3687 	/*
3688 	 * There are setups with compaction disabled which would prefer to loop
3689 	 * inside the allocator rather than hit the oom killer prematurely.
3690 	 * Let's give them a good hope and keep retrying while the order-0
3691 	 * watermarks are OK.
3692 	 */
3693 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3694 					ac->nodemask) {
3695 		if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3696 					ac_classzone_idx(ac), alloc_flags))
3697 			return true;
3698 	}
3699 	return false;
3700 }
3701 #endif /* CONFIG_COMPACTION */
3702 
3703 #ifdef CONFIG_LOCKDEP
3704 static struct lockdep_map __fs_reclaim_map =
3705 	STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
3706 
3707 static bool __need_fs_reclaim(gfp_t gfp_mask)
3708 {
3709 	gfp_mask = current_gfp_context(gfp_mask);
3710 
3711 	/* no reclaim without waiting on it */
3712 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
3713 		return false;
3714 
3715 	/* this guy won't enter reclaim */
3716 	if (current->flags & PF_MEMALLOC)
3717 		return false;
3718 
3719 	/* We're only interested __GFP_FS allocations for now */
3720 	if (!(gfp_mask & __GFP_FS))
3721 		return false;
3722 
3723 	if (gfp_mask & __GFP_NOLOCKDEP)
3724 		return false;
3725 
3726 	return true;
3727 }
3728 
3729 void __fs_reclaim_acquire(void)
3730 {
3731 	lock_map_acquire(&__fs_reclaim_map);
3732 }
3733 
3734 void __fs_reclaim_release(void)
3735 {
3736 	lock_map_release(&__fs_reclaim_map);
3737 }
3738 
3739 void fs_reclaim_acquire(gfp_t gfp_mask)
3740 {
3741 	if (__need_fs_reclaim(gfp_mask))
3742 		__fs_reclaim_acquire();
3743 }
3744 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
3745 
3746 void fs_reclaim_release(gfp_t gfp_mask)
3747 {
3748 	if (__need_fs_reclaim(gfp_mask))
3749 		__fs_reclaim_release();
3750 }
3751 EXPORT_SYMBOL_GPL(fs_reclaim_release);
3752 #endif
3753 
3754 /* Perform direct synchronous page reclaim */
3755 static int
3756 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3757 					const struct alloc_context *ac)
3758 {
3759 	struct reclaim_state reclaim_state;
3760 	int progress;
3761 	unsigned int noreclaim_flag;
3762 	unsigned long pflags;
3763 
3764 	cond_resched();
3765 
3766 	/* We now go into synchronous reclaim */
3767 	cpuset_memory_pressure_bump();
3768 	psi_memstall_enter(&pflags);
3769 	fs_reclaim_acquire(gfp_mask);
3770 	noreclaim_flag = memalloc_noreclaim_save();
3771 	reclaim_state.reclaimed_slab = 0;
3772 	current->reclaim_state = &reclaim_state;
3773 
3774 	progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3775 								ac->nodemask);
3776 
3777 	current->reclaim_state = NULL;
3778 	memalloc_noreclaim_restore(noreclaim_flag);
3779 	fs_reclaim_release(gfp_mask);
3780 	psi_memstall_leave(&pflags);
3781 
3782 	cond_resched();
3783 
3784 	return progress;
3785 }
3786 
3787 /* The really slow allocator path where we enter direct reclaim */
3788 static inline struct page *
3789 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3790 		unsigned int alloc_flags, const struct alloc_context *ac,
3791 		unsigned long *did_some_progress)
3792 {
3793 	struct page *page = NULL;
3794 	bool drained = false;
3795 
3796 	*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3797 	if (unlikely(!(*did_some_progress)))
3798 		return NULL;
3799 
3800 retry:
3801 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3802 
3803 	/*
3804 	 * If an allocation failed after direct reclaim, it could be because
3805 	 * pages are pinned on the per-cpu lists or in high alloc reserves.
3806 	 * Shrink them them and try again
3807 	 */
3808 	if (!page && !drained) {
3809 		unreserve_highatomic_pageblock(ac, false);
3810 		drain_all_pages(NULL);
3811 		drained = true;
3812 		goto retry;
3813 	}
3814 
3815 	return page;
3816 }
3817 
3818 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
3819 			     const struct alloc_context *ac)
3820 {
3821 	struct zoneref *z;
3822 	struct zone *zone;
3823 	pg_data_t *last_pgdat = NULL;
3824 	enum zone_type high_zoneidx = ac->high_zoneidx;
3825 
3826 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx,
3827 					ac->nodemask) {
3828 		if (last_pgdat != zone->zone_pgdat)
3829 			wakeup_kswapd(zone, gfp_mask, order, high_zoneidx);
3830 		last_pgdat = zone->zone_pgdat;
3831 	}
3832 }
3833 
3834 static inline unsigned int
3835 gfp_to_alloc_flags(gfp_t gfp_mask)
3836 {
3837 	unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3838 
3839 	/* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3840 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3841 
3842 	/*
3843 	 * The caller may dip into page reserves a bit more if the caller
3844 	 * cannot run direct reclaim, or if the caller has realtime scheduling
3845 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
3846 	 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3847 	 */
3848 	alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3849 
3850 	if (gfp_mask & __GFP_ATOMIC) {
3851 		/*
3852 		 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3853 		 * if it can't schedule.
3854 		 */
3855 		if (!(gfp_mask & __GFP_NOMEMALLOC))
3856 			alloc_flags |= ALLOC_HARDER;
3857 		/*
3858 		 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3859 		 * comment for __cpuset_node_allowed().
3860 		 */
3861 		alloc_flags &= ~ALLOC_CPUSET;
3862 	} else if (unlikely(rt_task(current)) && !in_interrupt())
3863 		alloc_flags |= ALLOC_HARDER;
3864 
3865 #ifdef CONFIG_CMA
3866 	if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3867 		alloc_flags |= ALLOC_CMA;
3868 #endif
3869 	return alloc_flags;
3870 }
3871 
3872 static bool oom_reserves_allowed(struct task_struct *tsk)
3873 {
3874 	if (!tsk_is_oom_victim(tsk))
3875 		return false;
3876 
3877 	/*
3878 	 * !MMU doesn't have oom reaper so give access to memory reserves
3879 	 * only to the thread with TIF_MEMDIE set
3880 	 */
3881 	if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
3882 		return false;
3883 
3884 	return true;
3885 }
3886 
3887 /*
3888  * Distinguish requests which really need access to full memory
3889  * reserves from oom victims which can live with a portion of it
3890  */
3891 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
3892 {
3893 	if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3894 		return 0;
3895 	if (gfp_mask & __GFP_MEMALLOC)
3896 		return ALLOC_NO_WATERMARKS;
3897 	if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3898 		return ALLOC_NO_WATERMARKS;
3899 	if (!in_interrupt()) {
3900 		if (current->flags & PF_MEMALLOC)
3901 			return ALLOC_NO_WATERMARKS;
3902 		else if (oom_reserves_allowed(current))
3903 			return ALLOC_OOM;
3904 	}
3905 
3906 	return 0;
3907 }
3908 
3909 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3910 {
3911 	return !!__gfp_pfmemalloc_flags(gfp_mask);
3912 }
3913 
3914 /*
3915  * Checks whether it makes sense to retry the reclaim to make a forward progress
3916  * for the given allocation request.
3917  *
3918  * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3919  * without success, or when we couldn't even meet the watermark if we
3920  * reclaimed all remaining pages on the LRU lists.
3921  *
3922  * Returns true if a retry is viable or false to enter the oom path.
3923  */
3924 static inline bool
3925 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3926 		     struct alloc_context *ac, int alloc_flags,
3927 		     bool did_some_progress, int *no_progress_loops)
3928 {
3929 	struct zone *zone;
3930 	struct zoneref *z;
3931 	bool ret = false;
3932 
3933 	/*
3934 	 * Costly allocations might have made a progress but this doesn't mean
3935 	 * their order will become available due to high fragmentation so
3936 	 * always increment the no progress counter for them
3937 	 */
3938 	if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3939 		*no_progress_loops = 0;
3940 	else
3941 		(*no_progress_loops)++;
3942 
3943 	/*
3944 	 * Make sure we converge to OOM if we cannot make any progress
3945 	 * several times in the row.
3946 	 */
3947 	if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
3948 		/* Before OOM, exhaust highatomic_reserve */
3949 		return unreserve_highatomic_pageblock(ac, true);
3950 	}
3951 
3952 	/*
3953 	 * Keep reclaiming pages while there is a chance this will lead
3954 	 * somewhere.  If none of the target zones can satisfy our allocation
3955 	 * request even if all reclaimable pages are considered then we are
3956 	 * screwed and have to go OOM.
3957 	 */
3958 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3959 					ac->nodemask) {
3960 		unsigned long available;
3961 		unsigned long reclaimable;
3962 		unsigned long min_wmark = min_wmark_pages(zone);
3963 		bool wmark;
3964 
3965 		available = reclaimable = zone_reclaimable_pages(zone);
3966 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3967 
3968 		/*
3969 		 * Would the allocation succeed if we reclaimed all
3970 		 * reclaimable pages?
3971 		 */
3972 		wmark = __zone_watermark_ok(zone, order, min_wmark,
3973 				ac_classzone_idx(ac), alloc_flags, available);
3974 		trace_reclaim_retry_zone(z, order, reclaimable,
3975 				available, min_wmark, *no_progress_loops, wmark);
3976 		if (wmark) {
3977 			/*
3978 			 * If we didn't make any progress and have a lot of
3979 			 * dirty + writeback pages then we should wait for
3980 			 * an IO to complete to slow down the reclaim and
3981 			 * prevent from pre mature OOM
3982 			 */
3983 			if (!did_some_progress) {
3984 				unsigned long write_pending;
3985 
3986 				write_pending = zone_page_state_snapshot(zone,
3987 							NR_ZONE_WRITE_PENDING);
3988 
3989 				if (2 * write_pending > reclaimable) {
3990 					congestion_wait(BLK_RW_ASYNC, HZ/10);
3991 					return true;
3992 				}
3993 			}
3994 
3995 			ret = true;
3996 			goto out;
3997 		}
3998 	}
3999 
4000 out:
4001 	/*
4002 	 * Memory allocation/reclaim might be called from a WQ context and the
4003 	 * current implementation of the WQ concurrency control doesn't
4004 	 * recognize that a particular WQ is congested if the worker thread is
4005 	 * looping without ever sleeping. Therefore we have to do a short sleep
4006 	 * here rather than calling cond_resched().
4007 	 */
4008 	if (current->flags & PF_WQ_WORKER)
4009 		schedule_timeout_uninterruptible(1);
4010 	else
4011 		cond_resched();
4012 	return ret;
4013 }
4014 
4015 static inline bool
4016 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4017 {
4018 	/*
4019 	 * It's possible that cpuset's mems_allowed and the nodemask from
4020 	 * mempolicy don't intersect. This should be normally dealt with by
4021 	 * policy_nodemask(), but it's possible to race with cpuset update in
4022 	 * such a way the check therein was true, and then it became false
4023 	 * before we got our cpuset_mems_cookie here.
4024 	 * This assumes that for all allocations, ac->nodemask can come only
4025 	 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4026 	 * when it does not intersect with the cpuset restrictions) or the
4027 	 * caller can deal with a violated nodemask.
4028 	 */
4029 	if (cpusets_enabled() && ac->nodemask &&
4030 			!cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4031 		ac->nodemask = NULL;
4032 		return true;
4033 	}
4034 
4035 	/*
4036 	 * When updating a task's mems_allowed or mempolicy nodemask, it is
4037 	 * possible to race with parallel threads in such a way that our
4038 	 * allocation can fail while the mask is being updated. If we are about
4039 	 * to fail, check if the cpuset changed during allocation and if so,
4040 	 * retry.
4041 	 */
4042 	if (read_mems_allowed_retry(cpuset_mems_cookie))
4043 		return true;
4044 
4045 	return false;
4046 }
4047 
4048 static inline struct page *
4049 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4050 						struct alloc_context *ac)
4051 {
4052 	bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4053 	const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4054 	struct page *page = NULL;
4055 	unsigned int alloc_flags;
4056 	unsigned long did_some_progress;
4057 	enum compact_priority compact_priority;
4058 	enum compact_result compact_result;
4059 	int compaction_retries;
4060 	int no_progress_loops;
4061 	unsigned int cpuset_mems_cookie;
4062 	int reserve_flags;
4063 
4064 	/*
4065 	 * We also sanity check to catch abuse of atomic reserves being used by
4066 	 * callers that are not in atomic context.
4067 	 */
4068 	if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4069 				(__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4070 		gfp_mask &= ~__GFP_ATOMIC;
4071 
4072 retry_cpuset:
4073 	compaction_retries = 0;
4074 	no_progress_loops = 0;
4075 	compact_priority = DEF_COMPACT_PRIORITY;
4076 	cpuset_mems_cookie = read_mems_allowed_begin();
4077 
4078 	/*
4079 	 * The fast path uses conservative alloc_flags to succeed only until
4080 	 * kswapd needs to be woken up, and to avoid the cost of setting up
4081 	 * alloc_flags precisely. So we do that now.
4082 	 */
4083 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
4084 
4085 	/*
4086 	 * We need to recalculate the starting point for the zonelist iterator
4087 	 * because we might have used different nodemask in the fast path, or
4088 	 * there was a cpuset modification and we are retrying - otherwise we
4089 	 * could end up iterating over non-eligible zones endlessly.
4090 	 */
4091 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4092 					ac->high_zoneidx, ac->nodemask);
4093 	if (!ac->preferred_zoneref->zone)
4094 		goto nopage;
4095 
4096 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4097 		wake_all_kswapds(order, gfp_mask, ac);
4098 
4099 	/*
4100 	 * The adjusted alloc_flags might result in immediate success, so try
4101 	 * that first
4102 	 */
4103 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4104 	if (page)
4105 		goto got_pg;
4106 
4107 	/*
4108 	 * For costly allocations, try direct compaction first, as it's likely
4109 	 * that we have enough base pages and don't need to reclaim. For non-
4110 	 * movable high-order allocations, do that as well, as compaction will
4111 	 * try prevent permanent fragmentation by migrating from blocks of the
4112 	 * same migratetype.
4113 	 * Don't try this for allocations that are allowed to ignore
4114 	 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4115 	 */
4116 	if (can_direct_reclaim &&
4117 			(costly_order ||
4118 			   (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4119 			&& !gfp_pfmemalloc_allowed(gfp_mask)) {
4120 		page = __alloc_pages_direct_compact(gfp_mask, order,
4121 						alloc_flags, ac,
4122 						INIT_COMPACT_PRIORITY,
4123 						&compact_result);
4124 		if (page)
4125 			goto got_pg;
4126 
4127 		/*
4128 		 * Checks for costly allocations with __GFP_NORETRY, which
4129 		 * includes THP page fault allocations
4130 		 */
4131 		if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4132 			/*
4133 			 * If compaction is deferred for high-order allocations,
4134 			 * it is because sync compaction recently failed. If
4135 			 * this is the case and the caller requested a THP
4136 			 * allocation, we do not want to heavily disrupt the
4137 			 * system, so we fail the allocation instead of entering
4138 			 * direct reclaim.
4139 			 */
4140 			if (compact_result == COMPACT_DEFERRED)
4141 				goto nopage;
4142 
4143 			/*
4144 			 * Looks like reclaim/compaction is worth trying, but
4145 			 * sync compaction could be very expensive, so keep
4146 			 * using async compaction.
4147 			 */
4148 			compact_priority = INIT_COMPACT_PRIORITY;
4149 		}
4150 	}
4151 
4152 retry:
4153 	/* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4154 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4155 		wake_all_kswapds(order, gfp_mask, ac);
4156 
4157 	reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4158 	if (reserve_flags)
4159 		alloc_flags = reserve_flags;
4160 
4161 	/*
4162 	 * Reset the nodemask and zonelist iterators if memory policies can be
4163 	 * ignored. These allocations are high priority and system rather than
4164 	 * user oriented.
4165 	 */
4166 	if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4167 		ac->nodemask = NULL;
4168 		ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4169 					ac->high_zoneidx, ac->nodemask);
4170 	}
4171 
4172 	/* Attempt with potentially adjusted zonelist and alloc_flags */
4173 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4174 	if (page)
4175 		goto got_pg;
4176 
4177 	/* Caller is not willing to reclaim, we can't balance anything */
4178 	if (!can_direct_reclaim)
4179 		goto nopage;
4180 
4181 	/* Avoid recursion of direct reclaim */
4182 	if (current->flags & PF_MEMALLOC)
4183 		goto nopage;
4184 
4185 	/* Try direct reclaim and then allocating */
4186 	page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4187 							&did_some_progress);
4188 	if (page)
4189 		goto got_pg;
4190 
4191 	/* Try direct compaction and then allocating */
4192 	page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4193 					compact_priority, &compact_result);
4194 	if (page)
4195 		goto got_pg;
4196 
4197 	/* Do not loop if specifically requested */
4198 	if (gfp_mask & __GFP_NORETRY)
4199 		goto nopage;
4200 
4201 	/*
4202 	 * Do not retry costly high order allocations unless they are
4203 	 * __GFP_RETRY_MAYFAIL
4204 	 */
4205 	if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4206 		goto nopage;
4207 
4208 	if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4209 				 did_some_progress > 0, &no_progress_loops))
4210 		goto retry;
4211 
4212 	/*
4213 	 * It doesn't make any sense to retry for the compaction if the order-0
4214 	 * reclaim is not able to make any progress because the current
4215 	 * implementation of the compaction depends on the sufficient amount
4216 	 * of free memory (see __compaction_suitable)
4217 	 */
4218 	if (did_some_progress > 0 &&
4219 			should_compact_retry(ac, order, alloc_flags,
4220 				compact_result, &compact_priority,
4221 				&compaction_retries))
4222 		goto retry;
4223 
4224 
4225 	/* Deal with possible cpuset update races before we start OOM killing */
4226 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4227 		goto retry_cpuset;
4228 
4229 	/* Reclaim has failed us, start killing things */
4230 	page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4231 	if (page)
4232 		goto got_pg;
4233 
4234 	/* Avoid allocations with no watermarks from looping endlessly */
4235 	if (tsk_is_oom_victim(current) &&
4236 	    (alloc_flags == ALLOC_OOM ||
4237 	     (gfp_mask & __GFP_NOMEMALLOC)))
4238 		goto nopage;
4239 
4240 	/* Retry as long as the OOM killer is making progress */
4241 	if (did_some_progress) {
4242 		no_progress_loops = 0;
4243 		goto retry;
4244 	}
4245 
4246 nopage:
4247 	/* Deal with possible cpuset update races before we fail */
4248 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4249 		goto retry_cpuset;
4250 
4251 	/*
4252 	 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4253 	 * we always retry
4254 	 */
4255 	if (gfp_mask & __GFP_NOFAIL) {
4256 		/*
4257 		 * All existing users of the __GFP_NOFAIL are blockable, so warn
4258 		 * of any new users that actually require GFP_NOWAIT
4259 		 */
4260 		if (WARN_ON_ONCE(!can_direct_reclaim))
4261 			goto fail;
4262 
4263 		/*
4264 		 * PF_MEMALLOC request from this context is rather bizarre
4265 		 * because we cannot reclaim anything and only can loop waiting
4266 		 * for somebody to do a work for us
4267 		 */
4268 		WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4269 
4270 		/*
4271 		 * non failing costly orders are a hard requirement which we
4272 		 * are not prepared for much so let's warn about these users
4273 		 * so that we can identify them and convert them to something
4274 		 * else.
4275 		 */
4276 		WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4277 
4278 		/*
4279 		 * Help non-failing allocations by giving them access to memory
4280 		 * reserves but do not use ALLOC_NO_WATERMARKS because this
4281 		 * could deplete whole memory reserves which would just make
4282 		 * the situation worse
4283 		 */
4284 		page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4285 		if (page)
4286 			goto got_pg;
4287 
4288 		cond_resched();
4289 		goto retry;
4290 	}
4291 fail:
4292 	warn_alloc(gfp_mask, ac->nodemask,
4293 			"page allocation failure: order:%u", order);
4294 got_pg:
4295 	return page;
4296 }
4297 
4298 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4299 		int preferred_nid, nodemask_t *nodemask,
4300 		struct alloc_context *ac, gfp_t *alloc_mask,
4301 		unsigned int *alloc_flags)
4302 {
4303 	ac->high_zoneidx = gfp_zone(gfp_mask);
4304 	ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4305 	ac->nodemask = nodemask;
4306 	ac->migratetype = gfpflags_to_migratetype(gfp_mask);
4307 
4308 	if (cpusets_enabled()) {
4309 		*alloc_mask |= __GFP_HARDWALL;
4310 		if (!ac->nodemask)
4311 			ac->nodemask = &cpuset_current_mems_allowed;
4312 		else
4313 			*alloc_flags |= ALLOC_CPUSET;
4314 	}
4315 
4316 	fs_reclaim_acquire(gfp_mask);
4317 	fs_reclaim_release(gfp_mask);
4318 
4319 	might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4320 
4321 	if (should_fail_alloc_page(gfp_mask, order))
4322 		return false;
4323 
4324 	if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4325 		*alloc_flags |= ALLOC_CMA;
4326 
4327 	return true;
4328 }
4329 
4330 /* Determine whether to spread dirty pages and what the first usable zone */
4331 static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac)
4332 {
4333 	/* Dirty zone balancing only done in the fast path */
4334 	ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4335 
4336 	/*
4337 	 * The preferred zone is used for statistics but crucially it is
4338 	 * also used as the starting point for the zonelist iterator. It
4339 	 * may get reset for allocations that ignore memory policies.
4340 	 */
4341 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4342 					ac->high_zoneidx, ac->nodemask);
4343 }
4344 
4345 /*
4346  * This is the 'heart' of the zoned buddy allocator.
4347  */
4348 struct page *
4349 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4350 							nodemask_t *nodemask)
4351 {
4352 	struct page *page;
4353 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
4354 	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4355 	struct alloc_context ac = { };
4356 
4357 	/*
4358 	 * There are several places where we assume that the order value is sane
4359 	 * so bail out early if the request is out of bound.
4360 	 */
4361 	if (unlikely(order >= MAX_ORDER)) {
4362 		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4363 		return NULL;
4364 	}
4365 
4366 	gfp_mask &= gfp_allowed_mask;
4367 	alloc_mask = gfp_mask;
4368 	if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4369 		return NULL;
4370 
4371 	finalise_ac(gfp_mask, &ac);
4372 
4373 	/* First allocation attempt */
4374 	page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4375 	if (likely(page))
4376 		goto out;
4377 
4378 	/*
4379 	 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4380 	 * resp. GFP_NOIO which has to be inherited for all allocation requests
4381 	 * from a particular context which has been marked by
4382 	 * memalloc_no{fs,io}_{save,restore}.
4383 	 */
4384 	alloc_mask = current_gfp_context(gfp_mask);
4385 	ac.spread_dirty_pages = false;
4386 
4387 	/*
4388 	 * Restore the original nodemask if it was potentially replaced with
4389 	 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4390 	 */
4391 	if (unlikely(ac.nodemask != nodemask))
4392 		ac.nodemask = nodemask;
4393 
4394 	page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4395 
4396 out:
4397 	if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4398 	    unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4399 		__free_pages(page, order);
4400 		page = NULL;
4401 	}
4402 
4403 	trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4404 
4405 	return page;
4406 }
4407 EXPORT_SYMBOL(__alloc_pages_nodemask);
4408 
4409 /*
4410  * Common helper functions. Never use with __GFP_HIGHMEM because the returned
4411  * address cannot represent highmem pages. Use alloc_pages and then kmap if
4412  * you need to access high mem.
4413  */
4414 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4415 {
4416 	struct page *page;
4417 
4418 	page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
4419 	if (!page)
4420 		return 0;
4421 	return (unsigned long) page_address(page);
4422 }
4423 EXPORT_SYMBOL(__get_free_pages);
4424 
4425 unsigned long get_zeroed_page(gfp_t gfp_mask)
4426 {
4427 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4428 }
4429 EXPORT_SYMBOL(get_zeroed_page);
4430 
4431 void __free_pages(struct page *page, unsigned int order)
4432 {
4433 	if (put_page_testzero(page)) {
4434 		if (order == 0)
4435 			free_unref_page(page);
4436 		else
4437 			__free_pages_ok(page, order);
4438 	}
4439 }
4440 
4441 EXPORT_SYMBOL(__free_pages);
4442 
4443 void free_pages(unsigned long addr, unsigned int order)
4444 {
4445 	if (addr != 0) {
4446 		VM_BUG_ON(!virt_addr_valid((void *)addr));
4447 		__free_pages(virt_to_page((void *)addr), order);
4448 	}
4449 }
4450 
4451 EXPORT_SYMBOL(free_pages);
4452 
4453 /*
4454  * Page Fragment:
4455  *  An arbitrary-length arbitrary-offset area of memory which resides
4456  *  within a 0 or higher order page.  Multiple fragments within that page
4457  *  are individually refcounted, in the page's reference counter.
4458  *
4459  * The page_frag functions below provide a simple allocation framework for
4460  * page fragments.  This is used by the network stack and network device
4461  * drivers to provide a backing region of memory for use as either an
4462  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4463  */
4464 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4465 					     gfp_t gfp_mask)
4466 {
4467 	struct page *page = NULL;
4468 	gfp_t gfp = gfp_mask;
4469 
4470 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4471 	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4472 		    __GFP_NOMEMALLOC;
4473 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4474 				PAGE_FRAG_CACHE_MAX_ORDER);
4475 	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4476 #endif
4477 	if (unlikely(!page))
4478 		page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4479 
4480 	nc->va = page ? page_address(page) : NULL;
4481 
4482 	return page;
4483 }
4484 
4485 void __page_frag_cache_drain(struct page *page, unsigned int count)
4486 {
4487 	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4488 
4489 	if (page_ref_sub_and_test(page, count)) {
4490 		unsigned int order = compound_order(page);
4491 
4492 		if (order == 0)
4493 			free_unref_page(page);
4494 		else
4495 			__free_pages_ok(page, order);
4496 	}
4497 }
4498 EXPORT_SYMBOL(__page_frag_cache_drain);
4499 
4500 void *page_frag_alloc(struct page_frag_cache *nc,
4501 		      unsigned int fragsz, gfp_t gfp_mask)
4502 {
4503 	unsigned int size = PAGE_SIZE;
4504 	struct page *page;
4505 	int offset;
4506 
4507 	if (unlikely(!nc->va)) {
4508 refill:
4509 		page = __page_frag_cache_refill(nc, gfp_mask);
4510 		if (!page)
4511 			return NULL;
4512 
4513 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4514 		/* if size can vary use size else just use PAGE_SIZE */
4515 		size = nc->size;
4516 #endif
4517 		/* Even if we own the page, we do not use atomic_set().
4518 		 * This would break get_page_unless_zero() users.
4519 		 */
4520 		page_ref_add(page, size - 1);
4521 
4522 		/* reset page count bias and offset to start of new frag */
4523 		nc->pfmemalloc = page_is_pfmemalloc(page);
4524 		nc->pagecnt_bias = size;
4525 		nc->offset = size;
4526 	}
4527 
4528 	offset = nc->offset - fragsz;
4529 	if (unlikely(offset < 0)) {
4530 		page = virt_to_page(nc->va);
4531 
4532 		if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
4533 			goto refill;
4534 
4535 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4536 		/* if size can vary use size else just use PAGE_SIZE */
4537 		size = nc->size;
4538 #endif
4539 		/* OK, page count is 0, we can safely set it */
4540 		set_page_count(page, size);
4541 
4542 		/* reset page count bias and offset to start of new frag */
4543 		nc->pagecnt_bias = size;
4544 		offset = size - fragsz;
4545 	}
4546 
4547 	nc->pagecnt_bias--;
4548 	nc->offset = offset;
4549 
4550 	return nc->va + offset;
4551 }
4552 EXPORT_SYMBOL(page_frag_alloc);
4553 
4554 /*
4555  * Frees a page fragment allocated out of either a compound or order 0 page.
4556  */
4557 void page_frag_free(void *addr)
4558 {
4559 	struct page *page = virt_to_head_page(addr);
4560 
4561 	if (unlikely(put_page_testzero(page)))
4562 		__free_pages_ok(page, compound_order(page));
4563 }
4564 EXPORT_SYMBOL(page_frag_free);
4565 
4566 static void *make_alloc_exact(unsigned long addr, unsigned int order,
4567 		size_t size)
4568 {
4569 	if (addr) {
4570 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
4571 		unsigned long used = addr + PAGE_ALIGN(size);
4572 
4573 		split_page(virt_to_page((void *)addr), order);
4574 		while (used < alloc_end) {
4575 			free_page(used);
4576 			used += PAGE_SIZE;
4577 		}
4578 	}
4579 	return (void *)addr;
4580 }
4581 
4582 /**
4583  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4584  * @size: the number of bytes to allocate
4585  * @gfp_mask: GFP flags for the allocation
4586  *
4587  * This function is similar to alloc_pages(), except that it allocates the
4588  * minimum number of pages to satisfy the request.  alloc_pages() can only
4589  * allocate memory in power-of-two pages.
4590  *
4591  * This function is also limited by MAX_ORDER.
4592  *
4593  * Memory allocated by this function must be released by free_pages_exact().
4594  */
4595 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4596 {
4597 	unsigned int order = get_order(size);
4598 	unsigned long addr;
4599 
4600 	addr = __get_free_pages(gfp_mask, order);
4601 	return make_alloc_exact(addr, order, size);
4602 }
4603 EXPORT_SYMBOL(alloc_pages_exact);
4604 
4605 /**
4606  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4607  *			   pages on a node.
4608  * @nid: the preferred node ID where memory should be allocated
4609  * @size: the number of bytes to allocate
4610  * @gfp_mask: GFP flags for the allocation
4611  *
4612  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4613  * back.
4614  */
4615 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4616 {
4617 	unsigned int order = get_order(size);
4618 	struct page *p = alloc_pages_node(nid, gfp_mask, order);
4619 	if (!p)
4620 		return NULL;
4621 	return make_alloc_exact((unsigned long)page_address(p), order, size);
4622 }
4623 
4624 /**
4625  * free_pages_exact - release memory allocated via alloc_pages_exact()
4626  * @virt: the value returned by alloc_pages_exact.
4627  * @size: size of allocation, same value as passed to alloc_pages_exact().
4628  *
4629  * Release the memory allocated by a previous call to alloc_pages_exact.
4630  */
4631 void free_pages_exact(void *virt, size_t size)
4632 {
4633 	unsigned long addr = (unsigned long)virt;
4634 	unsigned long end = addr + PAGE_ALIGN(size);
4635 
4636 	while (addr < end) {
4637 		free_page(addr);
4638 		addr += PAGE_SIZE;
4639 	}
4640 }
4641 EXPORT_SYMBOL(free_pages_exact);
4642 
4643 /**
4644  * nr_free_zone_pages - count number of pages beyond high watermark
4645  * @offset: The zone index of the highest zone
4646  *
4647  * nr_free_zone_pages() counts the number of counts pages which are beyond the
4648  * high watermark within all zones at or below a given zone index.  For each
4649  * zone, the number of pages is calculated as:
4650  *
4651  *     nr_free_zone_pages = managed_pages - high_pages
4652  */
4653 static unsigned long nr_free_zone_pages(int offset)
4654 {
4655 	struct zoneref *z;
4656 	struct zone *zone;
4657 
4658 	/* Just pick one node, since fallback list is circular */
4659 	unsigned long sum = 0;
4660 
4661 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
4662 
4663 	for_each_zone_zonelist(zone, z, zonelist, offset) {
4664 		unsigned long size = zone->managed_pages;
4665 		unsigned long high = high_wmark_pages(zone);
4666 		if (size > high)
4667 			sum += size - high;
4668 	}
4669 
4670 	return sum;
4671 }
4672 
4673 /**
4674  * nr_free_buffer_pages - count number of pages beyond high watermark
4675  *
4676  * nr_free_buffer_pages() counts the number of pages which are beyond the high
4677  * watermark within ZONE_DMA and ZONE_NORMAL.
4678  */
4679 unsigned long nr_free_buffer_pages(void)
4680 {
4681 	return nr_free_zone_pages(gfp_zone(GFP_USER));
4682 }
4683 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
4684 
4685 /**
4686  * nr_free_pagecache_pages - count number of pages beyond high watermark
4687  *
4688  * nr_free_pagecache_pages() counts the number of pages which are beyond the
4689  * high watermark within all zones.
4690  */
4691 unsigned long nr_free_pagecache_pages(void)
4692 {
4693 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
4694 }
4695 
4696 static inline void show_node(struct zone *zone)
4697 {
4698 	if (IS_ENABLED(CONFIG_NUMA))
4699 		printk("Node %d ", zone_to_nid(zone));
4700 }
4701 
4702 long si_mem_available(void)
4703 {
4704 	long available;
4705 	unsigned long pagecache;
4706 	unsigned long wmark_low = 0;
4707 	unsigned long pages[NR_LRU_LISTS];
4708 	unsigned long reclaimable;
4709 	struct zone *zone;
4710 	int lru;
4711 
4712 	for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4713 		pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4714 
4715 	for_each_zone(zone)
4716 		wmark_low += zone->watermark[WMARK_LOW];
4717 
4718 	/*
4719 	 * Estimate the amount of memory available for userspace allocations,
4720 	 * without causing swapping.
4721 	 */
4722 	available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
4723 
4724 	/*
4725 	 * Not all the page cache can be freed, otherwise the system will
4726 	 * start swapping. Assume at least half of the page cache, or the
4727 	 * low watermark worth of cache, needs to stay.
4728 	 */
4729 	pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4730 	pagecache -= min(pagecache / 2, wmark_low);
4731 	available += pagecache;
4732 
4733 	/*
4734 	 * Part of the reclaimable slab and other kernel memory consists of
4735 	 * items that are in use, and cannot be freed. Cap this estimate at the
4736 	 * low watermark.
4737 	 */
4738 	reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) +
4739 			global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
4740 	available += reclaimable - min(reclaimable / 2, wmark_low);
4741 
4742 	if (available < 0)
4743 		available = 0;
4744 	return available;
4745 }
4746 EXPORT_SYMBOL_GPL(si_mem_available);
4747 
4748 void si_meminfo(struct sysinfo *val)
4749 {
4750 	val->totalram = totalram_pages;
4751 	val->sharedram = global_node_page_state(NR_SHMEM);
4752 	val->freeram = global_zone_page_state(NR_FREE_PAGES);
4753 	val->bufferram = nr_blockdev_pages();
4754 	val->totalhigh = totalhigh_pages;
4755 	val->freehigh = nr_free_highpages();
4756 	val->mem_unit = PAGE_SIZE;
4757 }
4758 
4759 EXPORT_SYMBOL(si_meminfo);
4760 
4761 #ifdef CONFIG_NUMA
4762 void si_meminfo_node(struct sysinfo *val, int nid)
4763 {
4764 	int zone_type;		/* needs to be signed */
4765 	unsigned long managed_pages = 0;
4766 	unsigned long managed_highpages = 0;
4767 	unsigned long free_highpages = 0;
4768 	pg_data_t *pgdat = NODE_DATA(nid);
4769 
4770 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4771 		managed_pages += pgdat->node_zones[zone_type].managed_pages;
4772 	val->totalram = managed_pages;
4773 	val->sharedram = node_page_state(pgdat, NR_SHMEM);
4774 	val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
4775 #ifdef CONFIG_HIGHMEM
4776 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4777 		struct zone *zone = &pgdat->node_zones[zone_type];
4778 
4779 		if (is_highmem(zone)) {
4780 			managed_highpages += zone->managed_pages;
4781 			free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4782 		}
4783 	}
4784 	val->totalhigh = managed_highpages;
4785 	val->freehigh = free_highpages;
4786 #else
4787 	val->totalhigh = managed_highpages;
4788 	val->freehigh = free_highpages;
4789 #endif
4790 	val->mem_unit = PAGE_SIZE;
4791 }
4792 #endif
4793 
4794 /*
4795  * Determine whether the node should be displayed or not, depending on whether
4796  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4797  */
4798 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
4799 {
4800 	if (!(flags & SHOW_MEM_FILTER_NODES))
4801 		return false;
4802 
4803 	/*
4804 	 * no node mask - aka implicit memory numa policy. Do not bother with
4805 	 * the synchronization - read_mems_allowed_begin - because we do not
4806 	 * have to be precise here.
4807 	 */
4808 	if (!nodemask)
4809 		nodemask = &cpuset_current_mems_allowed;
4810 
4811 	return !node_isset(nid, *nodemask);
4812 }
4813 
4814 #define K(x) ((x) << (PAGE_SHIFT-10))
4815 
4816 static void show_migration_types(unsigned char type)
4817 {
4818 	static const char types[MIGRATE_TYPES] = {
4819 		[MIGRATE_UNMOVABLE]	= 'U',
4820 		[MIGRATE_MOVABLE]	= 'M',
4821 		[MIGRATE_RECLAIMABLE]	= 'E',
4822 		[MIGRATE_HIGHATOMIC]	= 'H',
4823 #ifdef CONFIG_CMA
4824 		[MIGRATE_CMA]		= 'C',
4825 #endif
4826 #ifdef CONFIG_MEMORY_ISOLATION
4827 		[MIGRATE_ISOLATE]	= 'I',
4828 #endif
4829 	};
4830 	char tmp[MIGRATE_TYPES + 1];
4831 	char *p = tmp;
4832 	int i;
4833 
4834 	for (i = 0; i < MIGRATE_TYPES; i++) {
4835 		if (type & (1 << i))
4836 			*p++ = types[i];
4837 	}
4838 
4839 	*p = '\0';
4840 	printk(KERN_CONT "(%s) ", tmp);
4841 }
4842 
4843 /*
4844  * Show free area list (used inside shift_scroll-lock stuff)
4845  * We also calculate the percentage fragmentation. We do this by counting the
4846  * memory on each free list with the exception of the first item on the list.
4847  *
4848  * Bits in @filter:
4849  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4850  *   cpuset.
4851  */
4852 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
4853 {
4854 	unsigned long free_pcp = 0;
4855 	int cpu;
4856 	struct zone *zone;
4857 	pg_data_t *pgdat;
4858 
4859 	for_each_populated_zone(zone) {
4860 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
4861 			continue;
4862 
4863 		for_each_online_cpu(cpu)
4864 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4865 	}
4866 
4867 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4868 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4869 		" unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4870 		" slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4871 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4872 		" free:%lu free_pcp:%lu free_cma:%lu\n",
4873 		global_node_page_state(NR_ACTIVE_ANON),
4874 		global_node_page_state(NR_INACTIVE_ANON),
4875 		global_node_page_state(NR_ISOLATED_ANON),
4876 		global_node_page_state(NR_ACTIVE_FILE),
4877 		global_node_page_state(NR_INACTIVE_FILE),
4878 		global_node_page_state(NR_ISOLATED_FILE),
4879 		global_node_page_state(NR_UNEVICTABLE),
4880 		global_node_page_state(NR_FILE_DIRTY),
4881 		global_node_page_state(NR_WRITEBACK),
4882 		global_node_page_state(NR_UNSTABLE_NFS),
4883 		global_node_page_state(NR_SLAB_RECLAIMABLE),
4884 		global_node_page_state(NR_SLAB_UNRECLAIMABLE),
4885 		global_node_page_state(NR_FILE_MAPPED),
4886 		global_node_page_state(NR_SHMEM),
4887 		global_zone_page_state(NR_PAGETABLE),
4888 		global_zone_page_state(NR_BOUNCE),
4889 		global_zone_page_state(NR_FREE_PAGES),
4890 		free_pcp,
4891 		global_zone_page_state(NR_FREE_CMA_PAGES));
4892 
4893 	for_each_online_pgdat(pgdat) {
4894 		if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
4895 			continue;
4896 
4897 		printk("Node %d"
4898 			" active_anon:%lukB"
4899 			" inactive_anon:%lukB"
4900 			" active_file:%lukB"
4901 			" inactive_file:%lukB"
4902 			" unevictable:%lukB"
4903 			" isolated(anon):%lukB"
4904 			" isolated(file):%lukB"
4905 			" mapped:%lukB"
4906 			" dirty:%lukB"
4907 			" writeback:%lukB"
4908 			" shmem:%lukB"
4909 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4910 			" shmem_thp: %lukB"
4911 			" shmem_pmdmapped: %lukB"
4912 			" anon_thp: %lukB"
4913 #endif
4914 			" writeback_tmp:%lukB"
4915 			" unstable:%lukB"
4916 			" all_unreclaimable? %s"
4917 			"\n",
4918 			pgdat->node_id,
4919 			K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4920 			K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4921 			K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4922 			K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4923 			K(node_page_state(pgdat, NR_UNEVICTABLE)),
4924 			K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4925 			K(node_page_state(pgdat, NR_ISOLATED_FILE)),
4926 			K(node_page_state(pgdat, NR_FILE_MAPPED)),
4927 			K(node_page_state(pgdat, NR_FILE_DIRTY)),
4928 			K(node_page_state(pgdat, NR_WRITEBACK)),
4929 			K(node_page_state(pgdat, NR_SHMEM)),
4930 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4931 			K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4932 			K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4933 					* HPAGE_PMD_NR),
4934 			K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4935 #endif
4936 			K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4937 			K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
4938 			pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
4939 				"yes" : "no");
4940 	}
4941 
4942 	for_each_populated_zone(zone) {
4943 		int i;
4944 
4945 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
4946 			continue;
4947 
4948 		free_pcp = 0;
4949 		for_each_online_cpu(cpu)
4950 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4951 
4952 		show_node(zone);
4953 		printk(KERN_CONT
4954 			"%s"
4955 			" free:%lukB"
4956 			" min:%lukB"
4957 			" low:%lukB"
4958 			" high:%lukB"
4959 			" active_anon:%lukB"
4960 			" inactive_anon:%lukB"
4961 			" active_file:%lukB"
4962 			" inactive_file:%lukB"
4963 			" unevictable:%lukB"
4964 			" writepending:%lukB"
4965 			" present:%lukB"
4966 			" managed:%lukB"
4967 			" mlocked:%lukB"
4968 			" kernel_stack:%lukB"
4969 			" pagetables:%lukB"
4970 			" bounce:%lukB"
4971 			" free_pcp:%lukB"
4972 			" local_pcp:%ukB"
4973 			" free_cma:%lukB"
4974 			"\n",
4975 			zone->name,
4976 			K(zone_page_state(zone, NR_FREE_PAGES)),
4977 			K(min_wmark_pages(zone)),
4978 			K(low_wmark_pages(zone)),
4979 			K(high_wmark_pages(zone)),
4980 			K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4981 			K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4982 			K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4983 			K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4984 			K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
4985 			K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
4986 			K(zone->present_pages),
4987 			K(zone->managed_pages),
4988 			K(zone_page_state(zone, NR_MLOCK)),
4989 			zone_page_state(zone, NR_KERNEL_STACK_KB),
4990 			K(zone_page_state(zone, NR_PAGETABLE)),
4991 			K(zone_page_state(zone, NR_BOUNCE)),
4992 			K(free_pcp),
4993 			K(this_cpu_read(zone->pageset->pcp.count)),
4994 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
4995 		printk("lowmem_reserve[]:");
4996 		for (i = 0; i < MAX_NR_ZONES; i++)
4997 			printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
4998 		printk(KERN_CONT "\n");
4999 	}
5000 
5001 	for_each_populated_zone(zone) {
5002 		unsigned int order;
5003 		unsigned long nr[MAX_ORDER], flags, total = 0;
5004 		unsigned char types[MAX_ORDER];
5005 
5006 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5007 			continue;
5008 		show_node(zone);
5009 		printk(KERN_CONT "%s: ", zone->name);
5010 
5011 		spin_lock_irqsave(&zone->lock, flags);
5012 		for (order = 0; order < MAX_ORDER; order++) {
5013 			struct free_area *area = &zone->free_area[order];
5014 			int type;
5015 
5016 			nr[order] = area->nr_free;
5017 			total += nr[order] << order;
5018 
5019 			types[order] = 0;
5020 			for (type = 0; type < MIGRATE_TYPES; type++) {
5021 				if (!list_empty(&area->free_list[type]))
5022 					types[order] |= 1 << type;
5023 			}
5024 		}
5025 		spin_unlock_irqrestore(&zone->lock, flags);
5026 		for (order = 0; order < MAX_ORDER; order++) {
5027 			printk(KERN_CONT "%lu*%lukB ",
5028 			       nr[order], K(1UL) << order);
5029 			if (nr[order])
5030 				show_migration_types(types[order]);
5031 		}
5032 		printk(KERN_CONT "= %lukB\n", K(total));
5033 	}
5034 
5035 	hugetlb_show_meminfo();
5036 
5037 	printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5038 
5039 	show_swap_cache_info();
5040 }
5041 
5042 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5043 {
5044 	zoneref->zone = zone;
5045 	zoneref->zone_idx = zone_idx(zone);
5046 }
5047 
5048 /*
5049  * Builds allocation fallback zone lists.
5050  *
5051  * Add all populated zones of a node to the zonelist.
5052  */
5053 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5054 {
5055 	struct zone *zone;
5056 	enum zone_type zone_type = MAX_NR_ZONES;
5057 	int nr_zones = 0;
5058 
5059 	do {
5060 		zone_type--;
5061 		zone = pgdat->node_zones + zone_type;
5062 		if (managed_zone(zone)) {
5063 			zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5064 			check_highest_zone(zone_type);
5065 		}
5066 	} while (zone_type);
5067 
5068 	return nr_zones;
5069 }
5070 
5071 #ifdef CONFIG_NUMA
5072 
5073 static int __parse_numa_zonelist_order(char *s)
5074 {
5075 	/*
5076 	 * We used to support different zonlists modes but they turned
5077 	 * out to be just not useful. Let's keep the warning in place
5078 	 * if somebody still use the cmd line parameter so that we do
5079 	 * not fail it silently
5080 	 */
5081 	if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5082 		pr_warn("Ignoring unsupported numa_zonelist_order value:  %s\n", s);
5083 		return -EINVAL;
5084 	}
5085 	return 0;
5086 }
5087 
5088 static __init int setup_numa_zonelist_order(char *s)
5089 {
5090 	if (!s)
5091 		return 0;
5092 
5093 	return __parse_numa_zonelist_order(s);
5094 }
5095 early_param("numa_zonelist_order", setup_numa_zonelist_order);
5096 
5097 char numa_zonelist_order[] = "Node";
5098 
5099 /*
5100  * sysctl handler for numa_zonelist_order
5101  */
5102 int numa_zonelist_order_handler(struct ctl_table *table, int write,
5103 		void __user *buffer, size_t *length,
5104 		loff_t *ppos)
5105 {
5106 	char *str;
5107 	int ret;
5108 
5109 	if (!write)
5110 		return proc_dostring(table, write, buffer, length, ppos);
5111 	str = memdup_user_nul(buffer, 16);
5112 	if (IS_ERR(str))
5113 		return PTR_ERR(str);
5114 
5115 	ret = __parse_numa_zonelist_order(str);
5116 	kfree(str);
5117 	return ret;
5118 }
5119 
5120 
5121 #define MAX_NODE_LOAD (nr_online_nodes)
5122 static int node_load[MAX_NUMNODES];
5123 
5124 /**
5125  * find_next_best_node - find the next node that should appear in a given node's fallback list
5126  * @node: node whose fallback list we're appending
5127  * @used_node_mask: nodemask_t of already used nodes
5128  *
5129  * We use a number of factors to determine which is the next node that should
5130  * appear on a given node's fallback list.  The node should not have appeared
5131  * already in @node's fallback list, and it should be the next closest node
5132  * according to the distance array (which contains arbitrary distance values
5133  * from each node to each node in the system), and should also prefer nodes
5134  * with no CPUs, since presumably they'll have very little allocation pressure
5135  * on them otherwise.
5136  * It returns -1 if no node is found.
5137  */
5138 static int find_next_best_node(int node, nodemask_t *used_node_mask)
5139 {
5140 	int n, val;
5141 	int min_val = INT_MAX;
5142 	int best_node = NUMA_NO_NODE;
5143 	const struct cpumask *tmp = cpumask_of_node(0);
5144 
5145 	/* Use the local node if we haven't already */
5146 	if (!node_isset(node, *used_node_mask)) {
5147 		node_set(node, *used_node_mask);
5148 		return node;
5149 	}
5150 
5151 	for_each_node_state(n, N_MEMORY) {
5152 
5153 		/* Don't want a node to appear more than once */
5154 		if (node_isset(n, *used_node_mask))
5155 			continue;
5156 
5157 		/* Use the distance array to find the distance */
5158 		val = node_distance(node, n);
5159 
5160 		/* Penalize nodes under us ("prefer the next node") */
5161 		val += (n < node);
5162 
5163 		/* Give preference to headless and unused nodes */
5164 		tmp = cpumask_of_node(n);
5165 		if (!cpumask_empty(tmp))
5166 			val += PENALTY_FOR_NODE_WITH_CPUS;
5167 
5168 		/* Slight preference for less loaded node */
5169 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5170 		val += node_load[n];
5171 
5172 		if (val < min_val) {
5173 			min_val = val;
5174 			best_node = n;
5175 		}
5176 	}
5177 
5178 	if (best_node >= 0)
5179 		node_set(best_node, *used_node_mask);
5180 
5181 	return best_node;
5182 }
5183 
5184 
5185 /*
5186  * Build zonelists ordered by node and zones within node.
5187  * This results in maximum locality--normal zone overflows into local
5188  * DMA zone, if any--but risks exhausting DMA zone.
5189  */
5190 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5191 		unsigned nr_nodes)
5192 {
5193 	struct zoneref *zonerefs;
5194 	int i;
5195 
5196 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5197 
5198 	for (i = 0; i < nr_nodes; i++) {
5199 		int nr_zones;
5200 
5201 		pg_data_t *node = NODE_DATA(node_order[i]);
5202 
5203 		nr_zones = build_zonerefs_node(node, zonerefs);
5204 		zonerefs += nr_zones;
5205 	}
5206 	zonerefs->zone = NULL;
5207 	zonerefs->zone_idx = 0;
5208 }
5209 
5210 /*
5211  * Build gfp_thisnode zonelists
5212  */
5213 static void build_thisnode_zonelists(pg_data_t *pgdat)
5214 {
5215 	struct zoneref *zonerefs;
5216 	int nr_zones;
5217 
5218 	zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5219 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5220 	zonerefs += nr_zones;
5221 	zonerefs->zone = NULL;
5222 	zonerefs->zone_idx = 0;
5223 }
5224 
5225 /*
5226  * Build zonelists ordered by zone and nodes within zones.
5227  * This results in conserving DMA zone[s] until all Normal memory is
5228  * exhausted, but results in overflowing to remote node while memory
5229  * may still exist in local DMA zone.
5230  */
5231 
5232 static void build_zonelists(pg_data_t *pgdat)
5233 {
5234 	static int node_order[MAX_NUMNODES];
5235 	int node, load, nr_nodes = 0;
5236 	nodemask_t used_mask;
5237 	int local_node, prev_node;
5238 
5239 	/* NUMA-aware ordering of nodes */
5240 	local_node = pgdat->node_id;
5241 	load = nr_online_nodes;
5242 	prev_node = local_node;
5243 	nodes_clear(used_mask);
5244 
5245 	memset(node_order, 0, sizeof(node_order));
5246 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5247 		/*
5248 		 * We don't want to pressure a particular node.
5249 		 * So adding penalty to the first node in same
5250 		 * distance group to make it round-robin.
5251 		 */
5252 		if (node_distance(local_node, node) !=
5253 		    node_distance(local_node, prev_node))
5254 			node_load[node] = load;
5255 
5256 		node_order[nr_nodes++] = node;
5257 		prev_node = node;
5258 		load--;
5259 	}
5260 
5261 	build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5262 	build_thisnode_zonelists(pgdat);
5263 }
5264 
5265 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5266 /*
5267  * Return node id of node used for "local" allocations.
5268  * I.e., first node id of first zone in arg node's generic zonelist.
5269  * Used for initializing percpu 'numa_mem', which is used primarily
5270  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5271  */
5272 int local_memory_node(int node)
5273 {
5274 	struct zoneref *z;
5275 
5276 	z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5277 				   gfp_zone(GFP_KERNEL),
5278 				   NULL);
5279 	return zone_to_nid(z->zone);
5280 }
5281 #endif
5282 
5283 static void setup_min_unmapped_ratio(void);
5284 static void setup_min_slab_ratio(void);
5285 #else	/* CONFIG_NUMA */
5286 
5287 static void build_zonelists(pg_data_t *pgdat)
5288 {
5289 	int node, local_node;
5290 	struct zoneref *zonerefs;
5291 	int nr_zones;
5292 
5293 	local_node = pgdat->node_id;
5294 
5295 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5296 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5297 	zonerefs += nr_zones;
5298 
5299 	/*
5300 	 * Now we build the zonelist so that it contains the zones
5301 	 * of all the other nodes.
5302 	 * We don't want to pressure a particular node, so when
5303 	 * building the zones for node N, we make sure that the
5304 	 * zones coming right after the local ones are those from
5305 	 * node N+1 (modulo N)
5306 	 */
5307 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5308 		if (!node_online(node))
5309 			continue;
5310 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5311 		zonerefs += nr_zones;
5312 	}
5313 	for (node = 0; node < local_node; node++) {
5314 		if (!node_online(node))
5315 			continue;
5316 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5317 		zonerefs += nr_zones;
5318 	}
5319 
5320 	zonerefs->zone = NULL;
5321 	zonerefs->zone_idx = 0;
5322 }
5323 
5324 #endif	/* CONFIG_NUMA */
5325 
5326 /*
5327  * Boot pageset table. One per cpu which is going to be used for all
5328  * zones and all nodes. The parameters will be set in such a way
5329  * that an item put on a list will immediately be handed over to
5330  * the buddy list. This is safe since pageset manipulation is done
5331  * with interrupts disabled.
5332  *
5333  * The boot_pagesets must be kept even after bootup is complete for
5334  * unused processors and/or zones. They do play a role for bootstrapping
5335  * hotplugged processors.
5336  *
5337  * zoneinfo_show() and maybe other functions do
5338  * not check if the processor is online before following the pageset pointer.
5339  * Other parts of the kernel may not check if the zone is available.
5340  */
5341 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5342 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5343 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5344 
5345 static void __build_all_zonelists(void *data)
5346 {
5347 	int nid;
5348 	int __maybe_unused cpu;
5349 	pg_data_t *self = data;
5350 	static DEFINE_SPINLOCK(lock);
5351 
5352 	spin_lock(&lock);
5353 
5354 #ifdef CONFIG_NUMA
5355 	memset(node_load, 0, sizeof(node_load));
5356 #endif
5357 
5358 	/*
5359 	 * This node is hotadded and no memory is yet present.   So just
5360 	 * building zonelists is fine - no need to touch other nodes.
5361 	 */
5362 	if (self && !node_online(self->node_id)) {
5363 		build_zonelists(self);
5364 	} else {
5365 		for_each_online_node(nid) {
5366 			pg_data_t *pgdat = NODE_DATA(nid);
5367 
5368 			build_zonelists(pgdat);
5369 		}
5370 
5371 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5372 		/*
5373 		 * We now know the "local memory node" for each node--
5374 		 * i.e., the node of the first zone in the generic zonelist.
5375 		 * Set up numa_mem percpu variable for on-line cpus.  During
5376 		 * boot, only the boot cpu should be on-line;  we'll init the
5377 		 * secondary cpus' numa_mem as they come on-line.  During
5378 		 * node/memory hotplug, we'll fixup all on-line cpus.
5379 		 */
5380 		for_each_online_cpu(cpu)
5381 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5382 #endif
5383 	}
5384 
5385 	spin_unlock(&lock);
5386 }
5387 
5388 static noinline void __init
5389 build_all_zonelists_init(void)
5390 {
5391 	int cpu;
5392 
5393 	__build_all_zonelists(NULL);
5394 
5395 	/*
5396 	 * Initialize the boot_pagesets that are going to be used
5397 	 * for bootstrapping processors. The real pagesets for
5398 	 * each zone will be allocated later when the per cpu
5399 	 * allocator is available.
5400 	 *
5401 	 * boot_pagesets are used also for bootstrapping offline
5402 	 * cpus if the system is already booted because the pagesets
5403 	 * are needed to initialize allocators on a specific cpu too.
5404 	 * F.e. the percpu allocator needs the page allocator which
5405 	 * needs the percpu allocator in order to allocate its pagesets
5406 	 * (a chicken-egg dilemma).
5407 	 */
5408 	for_each_possible_cpu(cpu)
5409 		setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5410 
5411 	mminit_verify_zonelist();
5412 	cpuset_init_current_mems_allowed();
5413 }
5414 
5415 /*
5416  * unless system_state == SYSTEM_BOOTING.
5417  *
5418  * __ref due to call of __init annotated helper build_all_zonelists_init
5419  * [protected by SYSTEM_BOOTING].
5420  */
5421 void __ref build_all_zonelists(pg_data_t *pgdat)
5422 {
5423 	if (system_state == SYSTEM_BOOTING) {
5424 		build_all_zonelists_init();
5425 	} else {
5426 		__build_all_zonelists(pgdat);
5427 		/* cpuset refresh routine should be here */
5428 	}
5429 	vm_total_pages = nr_free_pagecache_pages();
5430 	/*
5431 	 * Disable grouping by mobility if the number of pages in the
5432 	 * system is too low to allow the mechanism to work. It would be
5433 	 * more accurate, but expensive to check per-zone. This check is
5434 	 * made on memory-hotadd so a system can start with mobility
5435 	 * disabled and enable it later
5436 	 */
5437 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5438 		page_group_by_mobility_disabled = 1;
5439 	else
5440 		page_group_by_mobility_disabled = 0;
5441 
5442 	pr_info("Built %i zonelists, mobility grouping %s.  Total pages: %ld\n",
5443 		nr_online_nodes,
5444 		page_group_by_mobility_disabled ? "off" : "on",
5445 		vm_total_pages);
5446 #ifdef CONFIG_NUMA
5447 	pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5448 #endif
5449 }
5450 
5451 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
5452 static bool __meminit
5453 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
5454 {
5455 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5456 	static struct memblock_region *r;
5457 
5458 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5459 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
5460 			for_each_memblock(memory, r) {
5461 				if (*pfn < memblock_region_memory_end_pfn(r))
5462 					break;
5463 			}
5464 		}
5465 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
5466 		    memblock_is_mirror(r)) {
5467 			*pfn = memblock_region_memory_end_pfn(r);
5468 			return true;
5469 		}
5470 	}
5471 #endif
5472 	return false;
5473 }
5474 
5475 /*
5476  * Initially all pages are reserved - free ones are freed
5477  * up by memblock_free_all() once the early boot process is
5478  * done. Non-atomic initialization, single-pass.
5479  */
5480 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5481 		unsigned long start_pfn, enum memmap_context context,
5482 		struct vmem_altmap *altmap)
5483 {
5484 	unsigned long pfn, end_pfn = start_pfn + size;
5485 	struct page *page;
5486 
5487 	if (highest_memmap_pfn < end_pfn - 1)
5488 		highest_memmap_pfn = end_pfn - 1;
5489 
5490 #ifdef CONFIG_ZONE_DEVICE
5491 	/*
5492 	 * Honor reservation requested by the driver for this ZONE_DEVICE
5493 	 * memory. We limit the total number of pages to initialize to just
5494 	 * those that might contain the memory mapping. We will defer the
5495 	 * ZONE_DEVICE page initialization until after we have released
5496 	 * the hotplug lock.
5497 	 */
5498 	if (zone == ZONE_DEVICE) {
5499 		if (!altmap)
5500 			return;
5501 
5502 		if (start_pfn == altmap->base_pfn)
5503 			start_pfn += altmap->reserve;
5504 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5505 	}
5506 #endif
5507 
5508 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5509 		/*
5510 		 * There can be holes in boot-time mem_map[]s handed to this
5511 		 * function.  They do not exist on hotplugged memory.
5512 		 */
5513 		if (context == MEMMAP_EARLY) {
5514 			if (!early_pfn_valid(pfn))
5515 				continue;
5516 			if (!early_pfn_in_nid(pfn, nid))
5517 				continue;
5518 			if (overlap_memmap_init(zone, &pfn))
5519 				continue;
5520 			if (defer_init(nid, pfn, end_pfn))
5521 				break;
5522 		}
5523 
5524 		page = pfn_to_page(pfn);
5525 		__init_single_page(page, pfn, zone, nid);
5526 		if (context == MEMMAP_HOTPLUG)
5527 			__SetPageReserved(page);
5528 
5529 		/*
5530 		 * Mark the block movable so that blocks are reserved for
5531 		 * movable at startup. This will force kernel allocations
5532 		 * to reserve their blocks rather than leaking throughout
5533 		 * the address space during boot when many long-lived
5534 		 * kernel allocations are made.
5535 		 *
5536 		 * bitmap is created for zone's valid pfn range. but memmap
5537 		 * can be created for invalid pages (for alignment)
5538 		 * check here not to call set_pageblock_migratetype() against
5539 		 * pfn out of zone.
5540 		 */
5541 		if (!(pfn & (pageblock_nr_pages - 1))) {
5542 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5543 			cond_resched();
5544 		}
5545 	}
5546 #ifdef CONFIG_SPARSEMEM
5547 	/*
5548 	 * If the zone does not span the rest of the section then
5549 	 * we should at least initialize those pages. Otherwise we
5550 	 * could blow up on a poisoned page in some paths which depend
5551 	 * on full sections being initialized (e.g. memory hotplug).
5552 	 */
5553 	while (end_pfn % PAGES_PER_SECTION) {
5554 		__init_single_page(pfn_to_page(end_pfn), end_pfn, zone, nid);
5555 		end_pfn++;
5556 	}
5557 #endif
5558 }
5559 
5560 #ifdef CONFIG_ZONE_DEVICE
5561 void __ref memmap_init_zone_device(struct zone *zone,
5562 				   unsigned long start_pfn,
5563 				   unsigned long size,
5564 				   struct dev_pagemap *pgmap)
5565 {
5566 	unsigned long pfn, end_pfn = start_pfn + size;
5567 	struct pglist_data *pgdat = zone->zone_pgdat;
5568 	unsigned long zone_idx = zone_idx(zone);
5569 	unsigned long start = jiffies;
5570 	int nid = pgdat->node_id;
5571 
5572 	if (WARN_ON_ONCE(!pgmap || !is_dev_zone(zone)))
5573 		return;
5574 
5575 	/*
5576 	 * The call to memmap_init_zone should have already taken care
5577 	 * of the pages reserved for the memmap, so we can just jump to
5578 	 * the end of that region and start processing the device pages.
5579 	 */
5580 	if (pgmap->altmap_valid) {
5581 		struct vmem_altmap *altmap = &pgmap->altmap;
5582 
5583 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5584 		size = end_pfn - start_pfn;
5585 	}
5586 
5587 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5588 		struct page *page = pfn_to_page(pfn);
5589 
5590 		__init_single_page(page, pfn, zone_idx, nid);
5591 
5592 		/*
5593 		 * Mark page reserved as it will need to wait for onlining
5594 		 * phase for it to be fully associated with a zone.
5595 		 *
5596 		 * We can use the non-atomic __set_bit operation for setting
5597 		 * the flag as we are still initializing the pages.
5598 		 */
5599 		__SetPageReserved(page);
5600 
5601 		/*
5602 		 * ZONE_DEVICE pages union ->lru with a ->pgmap back
5603 		 * pointer and hmm_data.  It is a bug if a ZONE_DEVICE
5604 		 * page is ever freed or placed on a driver-private list.
5605 		 */
5606 		page->pgmap = pgmap;
5607 		page->hmm_data = 0;
5608 
5609 		/*
5610 		 * Mark the block movable so that blocks are reserved for
5611 		 * movable at startup. This will force kernel allocations
5612 		 * to reserve their blocks rather than leaking throughout
5613 		 * the address space during boot when many long-lived
5614 		 * kernel allocations are made.
5615 		 *
5616 		 * bitmap is created for zone's valid pfn range. but memmap
5617 		 * can be created for invalid pages (for alignment)
5618 		 * check here not to call set_pageblock_migratetype() against
5619 		 * pfn out of zone.
5620 		 *
5621 		 * Please note that MEMMAP_HOTPLUG path doesn't clear memmap
5622 		 * because this is done early in sparse_add_one_section
5623 		 */
5624 		if (!(pfn & (pageblock_nr_pages - 1))) {
5625 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5626 			cond_resched();
5627 		}
5628 	}
5629 
5630 	pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev),
5631 		size, jiffies_to_msecs(jiffies - start));
5632 }
5633 
5634 #endif
5635 static void __meminit zone_init_free_lists(struct zone *zone)
5636 {
5637 	unsigned int order, t;
5638 	for_each_migratetype_order(order, t) {
5639 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5640 		zone->free_area[order].nr_free = 0;
5641 	}
5642 }
5643 
5644 void __meminit __weak memmap_init(unsigned long size, int nid,
5645 				  unsigned long zone, unsigned long start_pfn)
5646 {
5647 	memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL);
5648 }
5649 
5650 static int zone_batchsize(struct zone *zone)
5651 {
5652 #ifdef CONFIG_MMU
5653 	int batch;
5654 
5655 	/*
5656 	 * The per-cpu-pages pools are set to around 1000th of the
5657 	 * size of the zone.
5658 	 */
5659 	batch = zone->managed_pages / 1024;
5660 	/* But no more than a meg. */
5661 	if (batch * PAGE_SIZE > 1024 * 1024)
5662 		batch = (1024 * 1024) / PAGE_SIZE;
5663 	batch /= 4;		/* We effectively *= 4 below */
5664 	if (batch < 1)
5665 		batch = 1;
5666 
5667 	/*
5668 	 * Clamp the batch to a 2^n - 1 value. Having a power
5669 	 * of 2 value was found to be more likely to have
5670 	 * suboptimal cache aliasing properties in some cases.
5671 	 *
5672 	 * For example if 2 tasks are alternately allocating
5673 	 * batches of pages, one task can end up with a lot
5674 	 * of pages of one half of the possible page colors
5675 	 * and the other with pages of the other colors.
5676 	 */
5677 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
5678 
5679 	return batch;
5680 
5681 #else
5682 	/* The deferral and batching of frees should be suppressed under NOMMU
5683 	 * conditions.
5684 	 *
5685 	 * The problem is that NOMMU needs to be able to allocate large chunks
5686 	 * of contiguous memory as there's no hardware page translation to
5687 	 * assemble apparent contiguous memory from discontiguous pages.
5688 	 *
5689 	 * Queueing large contiguous runs of pages for batching, however,
5690 	 * causes the pages to actually be freed in smaller chunks.  As there
5691 	 * can be a significant delay between the individual batches being
5692 	 * recycled, this leads to the once large chunks of space being
5693 	 * fragmented and becoming unavailable for high-order allocations.
5694 	 */
5695 	return 0;
5696 #endif
5697 }
5698 
5699 /*
5700  * pcp->high and pcp->batch values are related and dependent on one another:
5701  * ->batch must never be higher then ->high.
5702  * The following function updates them in a safe manner without read side
5703  * locking.
5704  *
5705  * Any new users of pcp->batch and pcp->high should ensure they can cope with
5706  * those fields changing asynchronously (acording the the above rule).
5707  *
5708  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5709  * outside of boot time (or some other assurance that no concurrent updaters
5710  * exist).
5711  */
5712 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5713 		unsigned long batch)
5714 {
5715        /* start with a fail safe value for batch */
5716 	pcp->batch = 1;
5717 	smp_wmb();
5718 
5719        /* Update high, then batch, in order */
5720 	pcp->high = high;
5721 	smp_wmb();
5722 
5723 	pcp->batch = batch;
5724 }
5725 
5726 /* a companion to pageset_set_high() */
5727 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5728 {
5729 	pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5730 }
5731 
5732 static void pageset_init(struct per_cpu_pageset *p)
5733 {
5734 	struct per_cpu_pages *pcp;
5735 	int migratetype;
5736 
5737 	memset(p, 0, sizeof(*p));
5738 
5739 	pcp = &p->pcp;
5740 	pcp->count = 0;
5741 	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5742 		INIT_LIST_HEAD(&pcp->lists[migratetype]);
5743 }
5744 
5745 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5746 {
5747 	pageset_init(p);
5748 	pageset_set_batch(p, batch);
5749 }
5750 
5751 /*
5752  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5753  * to the value high for the pageset p.
5754  */
5755 static void pageset_set_high(struct per_cpu_pageset *p,
5756 				unsigned long high)
5757 {
5758 	unsigned long batch = max(1UL, high / 4);
5759 	if ((high / 4) > (PAGE_SHIFT * 8))
5760 		batch = PAGE_SHIFT * 8;
5761 
5762 	pageset_update(&p->pcp, high, batch);
5763 }
5764 
5765 static void pageset_set_high_and_batch(struct zone *zone,
5766 				       struct per_cpu_pageset *pcp)
5767 {
5768 	if (percpu_pagelist_fraction)
5769 		pageset_set_high(pcp,
5770 			(zone->managed_pages /
5771 				percpu_pagelist_fraction));
5772 	else
5773 		pageset_set_batch(pcp, zone_batchsize(zone));
5774 }
5775 
5776 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5777 {
5778 	struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5779 
5780 	pageset_init(pcp);
5781 	pageset_set_high_and_batch(zone, pcp);
5782 }
5783 
5784 void __meminit setup_zone_pageset(struct zone *zone)
5785 {
5786 	int cpu;
5787 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
5788 	for_each_possible_cpu(cpu)
5789 		zone_pageset_init(zone, cpu);
5790 }
5791 
5792 /*
5793  * Allocate per cpu pagesets and initialize them.
5794  * Before this call only boot pagesets were available.
5795  */
5796 void __init setup_per_cpu_pageset(void)
5797 {
5798 	struct pglist_data *pgdat;
5799 	struct zone *zone;
5800 
5801 	for_each_populated_zone(zone)
5802 		setup_zone_pageset(zone);
5803 
5804 	for_each_online_pgdat(pgdat)
5805 		pgdat->per_cpu_nodestats =
5806 			alloc_percpu(struct per_cpu_nodestat);
5807 }
5808 
5809 static __meminit void zone_pcp_init(struct zone *zone)
5810 {
5811 	/*
5812 	 * per cpu subsystem is not up at this point. The following code
5813 	 * relies on the ability of the linker to provide the
5814 	 * offset of a (static) per cpu variable into the per cpu area.
5815 	 */
5816 	zone->pageset = &boot_pageset;
5817 
5818 	if (populated_zone(zone))
5819 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
5820 			zone->name, zone->present_pages,
5821 					 zone_batchsize(zone));
5822 }
5823 
5824 void __meminit init_currently_empty_zone(struct zone *zone,
5825 					unsigned long zone_start_pfn,
5826 					unsigned long size)
5827 {
5828 	struct pglist_data *pgdat = zone->zone_pgdat;
5829 	int zone_idx = zone_idx(zone) + 1;
5830 
5831 	if (zone_idx > pgdat->nr_zones)
5832 		pgdat->nr_zones = zone_idx;
5833 
5834 	zone->zone_start_pfn = zone_start_pfn;
5835 
5836 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
5837 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
5838 			pgdat->node_id,
5839 			(unsigned long)zone_idx(zone),
5840 			zone_start_pfn, (zone_start_pfn + size));
5841 
5842 	zone_init_free_lists(zone);
5843 	zone->initialized = 1;
5844 }
5845 
5846 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5847 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5848 
5849 /*
5850  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5851  */
5852 int __meminit __early_pfn_to_nid(unsigned long pfn,
5853 					struct mminit_pfnnid_cache *state)
5854 {
5855 	unsigned long start_pfn, end_pfn;
5856 	int nid;
5857 
5858 	if (state->last_start <= pfn && pfn < state->last_end)
5859 		return state->last_nid;
5860 
5861 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5862 	if (nid != -1) {
5863 		state->last_start = start_pfn;
5864 		state->last_end = end_pfn;
5865 		state->last_nid = nid;
5866 	}
5867 
5868 	return nid;
5869 }
5870 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5871 
5872 /**
5873  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5874  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5875  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5876  *
5877  * If an architecture guarantees that all ranges registered contain no holes
5878  * and may be freed, this this function may be used instead of calling
5879  * memblock_free_early_nid() manually.
5880  */
5881 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5882 {
5883 	unsigned long start_pfn, end_pfn;
5884 	int i, this_nid;
5885 
5886 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5887 		start_pfn = min(start_pfn, max_low_pfn);
5888 		end_pfn = min(end_pfn, max_low_pfn);
5889 
5890 		if (start_pfn < end_pfn)
5891 			memblock_free_early_nid(PFN_PHYS(start_pfn),
5892 					(end_pfn - start_pfn) << PAGE_SHIFT,
5893 					this_nid);
5894 	}
5895 }
5896 
5897 /**
5898  * sparse_memory_present_with_active_regions - Call memory_present for each active range
5899  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5900  *
5901  * If an architecture guarantees that all ranges registered contain no holes and may
5902  * be freed, this function may be used instead of calling memory_present() manually.
5903  */
5904 void __init sparse_memory_present_with_active_regions(int nid)
5905 {
5906 	unsigned long start_pfn, end_pfn;
5907 	int i, this_nid;
5908 
5909 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5910 		memory_present(this_nid, start_pfn, end_pfn);
5911 }
5912 
5913 /**
5914  * get_pfn_range_for_nid - Return the start and end page frames for a node
5915  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5916  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5917  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5918  *
5919  * It returns the start and end page frame of a node based on information
5920  * provided by memblock_set_node(). If called for a node
5921  * with no available memory, a warning is printed and the start and end
5922  * PFNs will be 0.
5923  */
5924 void __meminit get_pfn_range_for_nid(unsigned int nid,
5925 			unsigned long *start_pfn, unsigned long *end_pfn)
5926 {
5927 	unsigned long this_start_pfn, this_end_pfn;
5928 	int i;
5929 
5930 	*start_pfn = -1UL;
5931 	*end_pfn = 0;
5932 
5933 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5934 		*start_pfn = min(*start_pfn, this_start_pfn);
5935 		*end_pfn = max(*end_pfn, this_end_pfn);
5936 	}
5937 
5938 	if (*start_pfn == -1UL)
5939 		*start_pfn = 0;
5940 }
5941 
5942 /*
5943  * This finds a zone that can be used for ZONE_MOVABLE pages. The
5944  * assumption is made that zones within a node are ordered in monotonic
5945  * increasing memory addresses so that the "highest" populated zone is used
5946  */
5947 static void __init find_usable_zone_for_movable(void)
5948 {
5949 	int zone_index;
5950 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5951 		if (zone_index == ZONE_MOVABLE)
5952 			continue;
5953 
5954 		if (arch_zone_highest_possible_pfn[zone_index] >
5955 				arch_zone_lowest_possible_pfn[zone_index])
5956 			break;
5957 	}
5958 
5959 	VM_BUG_ON(zone_index == -1);
5960 	movable_zone = zone_index;
5961 }
5962 
5963 /*
5964  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5965  * because it is sized independent of architecture. Unlike the other zones,
5966  * the starting point for ZONE_MOVABLE is not fixed. It may be different
5967  * in each node depending on the size of each node and how evenly kernelcore
5968  * is distributed. This helper function adjusts the zone ranges
5969  * provided by the architecture for a given node by using the end of the
5970  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5971  * zones within a node are in order of monotonic increases memory addresses
5972  */
5973 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5974 					unsigned long zone_type,
5975 					unsigned long node_start_pfn,
5976 					unsigned long node_end_pfn,
5977 					unsigned long *zone_start_pfn,
5978 					unsigned long *zone_end_pfn)
5979 {
5980 	/* Only adjust if ZONE_MOVABLE is on this node */
5981 	if (zone_movable_pfn[nid]) {
5982 		/* Size ZONE_MOVABLE */
5983 		if (zone_type == ZONE_MOVABLE) {
5984 			*zone_start_pfn = zone_movable_pfn[nid];
5985 			*zone_end_pfn = min(node_end_pfn,
5986 				arch_zone_highest_possible_pfn[movable_zone]);
5987 
5988 		/* Adjust for ZONE_MOVABLE starting within this range */
5989 		} else if (!mirrored_kernelcore &&
5990 			*zone_start_pfn < zone_movable_pfn[nid] &&
5991 			*zone_end_pfn > zone_movable_pfn[nid]) {
5992 			*zone_end_pfn = zone_movable_pfn[nid];
5993 
5994 		/* Check if this whole range is within ZONE_MOVABLE */
5995 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
5996 			*zone_start_pfn = *zone_end_pfn;
5997 	}
5998 }
5999 
6000 /*
6001  * Return the number of pages a zone spans in a node, including holes
6002  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
6003  */
6004 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
6005 					unsigned long zone_type,
6006 					unsigned long node_start_pfn,
6007 					unsigned long node_end_pfn,
6008 					unsigned long *zone_start_pfn,
6009 					unsigned long *zone_end_pfn,
6010 					unsigned long *ignored)
6011 {
6012 	/* When hotadd a new node from cpu_up(), the node should be empty */
6013 	if (!node_start_pfn && !node_end_pfn)
6014 		return 0;
6015 
6016 	/* Get the start and end of the zone */
6017 	*zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
6018 	*zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
6019 	adjust_zone_range_for_zone_movable(nid, zone_type,
6020 				node_start_pfn, node_end_pfn,
6021 				zone_start_pfn, zone_end_pfn);
6022 
6023 	/* Check that this node has pages within the zone's required range */
6024 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6025 		return 0;
6026 
6027 	/* Move the zone boundaries inside the node if necessary */
6028 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6029 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6030 
6031 	/* Return the spanned pages */
6032 	return *zone_end_pfn - *zone_start_pfn;
6033 }
6034 
6035 /*
6036  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
6037  * then all holes in the requested range will be accounted for.
6038  */
6039 unsigned long __meminit __absent_pages_in_range(int nid,
6040 				unsigned long range_start_pfn,
6041 				unsigned long range_end_pfn)
6042 {
6043 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
6044 	unsigned long start_pfn, end_pfn;
6045 	int i;
6046 
6047 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6048 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6049 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6050 		nr_absent -= end_pfn - start_pfn;
6051 	}
6052 	return nr_absent;
6053 }
6054 
6055 /**
6056  * absent_pages_in_range - Return number of page frames in holes within a range
6057  * @start_pfn: The start PFN to start searching for holes
6058  * @end_pfn: The end PFN to stop searching for holes
6059  *
6060  * It returns the number of pages frames in memory holes within a range.
6061  */
6062 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6063 							unsigned long end_pfn)
6064 {
6065 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6066 }
6067 
6068 /* Return the number of page frames in holes in a zone on a node */
6069 static unsigned long __meminit zone_absent_pages_in_node(int nid,
6070 					unsigned long zone_type,
6071 					unsigned long node_start_pfn,
6072 					unsigned long node_end_pfn,
6073 					unsigned long *ignored)
6074 {
6075 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6076 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6077 	unsigned long zone_start_pfn, zone_end_pfn;
6078 	unsigned long nr_absent;
6079 
6080 	/* When hotadd a new node from cpu_up(), the node should be empty */
6081 	if (!node_start_pfn && !node_end_pfn)
6082 		return 0;
6083 
6084 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6085 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6086 
6087 	adjust_zone_range_for_zone_movable(nid, zone_type,
6088 			node_start_pfn, node_end_pfn,
6089 			&zone_start_pfn, &zone_end_pfn);
6090 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6091 
6092 	/*
6093 	 * ZONE_MOVABLE handling.
6094 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6095 	 * and vice versa.
6096 	 */
6097 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6098 		unsigned long start_pfn, end_pfn;
6099 		struct memblock_region *r;
6100 
6101 		for_each_memblock(memory, r) {
6102 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
6103 					  zone_start_pfn, zone_end_pfn);
6104 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
6105 					zone_start_pfn, zone_end_pfn);
6106 
6107 			if (zone_type == ZONE_MOVABLE &&
6108 			    memblock_is_mirror(r))
6109 				nr_absent += end_pfn - start_pfn;
6110 
6111 			if (zone_type == ZONE_NORMAL &&
6112 			    !memblock_is_mirror(r))
6113 				nr_absent += end_pfn - start_pfn;
6114 		}
6115 	}
6116 
6117 	return nr_absent;
6118 }
6119 
6120 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6121 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
6122 					unsigned long zone_type,
6123 					unsigned long node_start_pfn,
6124 					unsigned long node_end_pfn,
6125 					unsigned long *zone_start_pfn,
6126 					unsigned long *zone_end_pfn,
6127 					unsigned long *zones_size)
6128 {
6129 	unsigned int zone;
6130 
6131 	*zone_start_pfn = node_start_pfn;
6132 	for (zone = 0; zone < zone_type; zone++)
6133 		*zone_start_pfn += zones_size[zone];
6134 
6135 	*zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
6136 
6137 	return zones_size[zone_type];
6138 }
6139 
6140 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
6141 						unsigned long zone_type,
6142 						unsigned long node_start_pfn,
6143 						unsigned long node_end_pfn,
6144 						unsigned long *zholes_size)
6145 {
6146 	if (!zholes_size)
6147 		return 0;
6148 
6149 	return zholes_size[zone_type];
6150 }
6151 
6152 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6153 
6154 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
6155 						unsigned long node_start_pfn,
6156 						unsigned long node_end_pfn,
6157 						unsigned long *zones_size,
6158 						unsigned long *zholes_size)
6159 {
6160 	unsigned long realtotalpages = 0, totalpages = 0;
6161 	enum zone_type i;
6162 
6163 	for (i = 0; i < MAX_NR_ZONES; i++) {
6164 		struct zone *zone = pgdat->node_zones + i;
6165 		unsigned long zone_start_pfn, zone_end_pfn;
6166 		unsigned long size, real_size;
6167 
6168 		size = zone_spanned_pages_in_node(pgdat->node_id, i,
6169 						  node_start_pfn,
6170 						  node_end_pfn,
6171 						  &zone_start_pfn,
6172 						  &zone_end_pfn,
6173 						  zones_size);
6174 		real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
6175 						  node_start_pfn, node_end_pfn,
6176 						  zholes_size);
6177 		if (size)
6178 			zone->zone_start_pfn = zone_start_pfn;
6179 		else
6180 			zone->zone_start_pfn = 0;
6181 		zone->spanned_pages = size;
6182 		zone->present_pages = real_size;
6183 
6184 		totalpages += size;
6185 		realtotalpages += real_size;
6186 	}
6187 
6188 	pgdat->node_spanned_pages = totalpages;
6189 	pgdat->node_present_pages = realtotalpages;
6190 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6191 							realtotalpages);
6192 }
6193 
6194 #ifndef CONFIG_SPARSEMEM
6195 /*
6196  * Calculate the size of the zone->blockflags rounded to an unsigned long
6197  * Start by making sure zonesize is a multiple of pageblock_order by rounding
6198  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6199  * round what is now in bits to nearest long in bits, then return it in
6200  * bytes.
6201  */
6202 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6203 {
6204 	unsigned long usemapsize;
6205 
6206 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6207 	usemapsize = roundup(zonesize, pageblock_nr_pages);
6208 	usemapsize = usemapsize >> pageblock_order;
6209 	usemapsize *= NR_PAGEBLOCK_BITS;
6210 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6211 
6212 	return usemapsize / 8;
6213 }
6214 
6215 static void __ref setup_usemap(struct pglist_data *pgdat,
6216 				struct zone *zone,
6217 				unsigned long zone_start_pfn,
6218 				unsigned long zonesize)
6219 {
6220 	unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6221 	zone->pageblock_flags = NULL;
6222 	if (usemapsize)
6223 		zone->pageblock_flags =
6224 			memblock_alloc_node_nopanic(usemapsize,
6225 							 pgdat->node_id);
6226 }
6227 #else
6228 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6229 				unsigned long zone_start_pfn, unsigned long zonesize) {}
6230 #endif /* CONFIG_SPARSEMEM */
6231 
6232 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6233 
6234 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6235 void __init set_pageblock_order(void)
6236 {
6237 	unsigned int order;
6238 
6239 	/* Check that pageblock_nr_pages has not already been setup */
6240 	if (pageblock_order)
6241 		return;
6242 
6243 	if (HPAGE_SHIFT > PAGE_SHIFT)
6244 		order = HUGETLB_PAGE_ORDER;
6245 	else
6246 		order = MAX_ORDER - 1;
6247 
6248 	/*
6249 	 * Assume the largest contiguous order of interest is a huge page.
6250 	 * This value may be variable depending on boot parameters on IA64 and
6251 	 * powerpc.
6252 	 */
6253 	pageblock_order = order;
6254 }
6255 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6256 
6257 /*
6258  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6259  * is unused as pageblock_order is set at compile-time. See
6260  * include/linux/pageblock-flags.h for the values of pageblock_order based on
6261  * the kernel config
6262  */
6263 void __init set_pageblock_order(void)
6264 {
6265 }
6266 
6267 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6268 
6269 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6270 						unsigned long present_pages)
6271 {
6272 	unsigned long pages = spanned_pages;
6273 
6274 	/*
6275 	 * Provide a more accurate estimation if there are holes within
6276 	 * the zone and SPARSEMEM is in use. If there are holes within the
6277 	 * zone, each populated memory region may cost us one or two extra
6278 	 * memmap pages due to alignment because memmap pages for each
6279 	 * populated regions may not be naturally aligned on page boundary.
6280 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6281 	 */
6282 	if (spanned_pages > present_pages + (present_pages >> 4) &&
6283 	    IS_ENABLED(CONFIG_SPARSEMEM))
6284 		pages = present_pages;
6285 
6286 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6287 }
6288 
6289 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6290 static void pgdat_init_split_queue(struct pglist_data *pgdat)
6291 {
6292 	spin_lock_init(&pgdat->split_queue_lock);
6293 	INIT_LIST_HEAD(&pgdat->split_queue);
6294 	pgdat->split_queue_len = 0;
6295 }
6296 #else
6297 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6298 #endif
6299 
6300 #ifdef CONFIG_COMPACTION
6301 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6302 {
6303 	init_waitqueue_head(&pgdat->kcompactd_wait);
6304 }
6305 #else
6306 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6307 #endif
6308 
6309 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6310 {
6311 	pgdat_resize_init(pgdat);
6312 
6313 	pgdat_init_split_queue(pgdat);
6314 	pgdat_init_kcompactd(pgdat);
6315 
6316 	init_waitqueue_head(&pgdat->kswapd_wait);
6317 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
6318 
6319 	pgdat_page_ext_init(pgdat);
6320 	spin_lock_init(&pgdat->lru_lock);
6321 	lruvec_init(node_lruvec(pgdat));
6322 }
6323 
6324 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6325 							unsigned long remaining_pages)
6326 {
6327 	zone->managed_pages = remaining_pages;
6328 	zone_set_nid(zone, nid);
6329 	zone->name = zone_names[idx];
6330 	zone->zone_pgdat = NODE_DATA(nid);
6331 	spin_lock_init(&zone->lock);
6332 	zone_seqlock_init(zone);
6333 	zone_pcp_init(zone);
6334 }
6335 
6336 /*
6337  * Set up the zone data structures
6338  * - init pgdat internals
6339  * - init all zones belonging to this node
6340  *
6341  * NOTE: this function is only called during memory hotplug
6342  */
6343 #ifdef CONFIG_MEMORY_HOTPLUG
6344 void __ref free_area_init_core_hotplug(int nid)
6345 {
6346 	enum zone_type z;
6347 	pg_data_t *pgdat = NODE_DATA(nid);
6348 
6349 	pgdat_init_internals(pgdat);
6350 	for (z = 0; z < MAX_NR_ZONES; z++)
6351 		zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6352 }
6353 #endif
6354 
6355 /*
6356  * Set up the zone data structures:
6357  *   - mark all pages reserved
6358  *   - mark all memory queues empty
6359  *   - clear the memory bitmaps
6360  *
6361  * NOTE: pgdat should get zeroed by caller.
6362  * NOTE: this function is only called during early init.
6363  */
6364 static void __init free_area_init_core(struct pglist_data *pgdat)
6365 {
6366 	enum zone_type j;
6367 	int nid = pgdat->node_id;
6368 
6369 	pgdat_init_internals(pgdat);
6370 	pgdat->per_cpu_nodestats = &boot_nodestats;
6371 
6372 	for (j = 0; j < MAX_NR_ZONES; j++) {
6373 		struct zone *zone = pgdat->node_zones + j;
6374 		unsigned long size, freesize, memmap_pages;
6375 		unsigned long zone_start_pfn = zone->zone_start_pfn;
6376 
6377 		size = zone->spanned_pages;
6378 		freesize = zone->present_pages;
6379 
6380 		/*
6381 		 * Adjust freesize so that it accounts for how much memory
6382 		 * is used by this zone for memmap. This affects the watermark
6383 		 * and per-cpu initialisations
6384 		 */
6385 		memmap_pages = calc_memmap_size(size, freesize);
6386 		if (!is_highmem_idx(j)) {
6387 			if (freesize >= memmap_pages) {
6388 				freesize -= memmap_pages;
6389 				if (memmap_pages)
6390 					printk(KERN_DEBUG
6391 					       "  %s zone: %lu pages used for memmap\n",
6392 					       zone_names[j], memmap_pages);
6393 			} else
6394 				pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
6395 					zone_names[j], memmap_pages, freesize);
6396 		}
6397 
6398 		/* Account for reserved pages */
6399 		if (j == 0 && freesize > dma_reserve) {
6400 			freesize -= dma_reserve;
6401 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
6402 					zone_names[0], dma_reserve);
6403 		}
6404 
6405 		if (!is_highmem_idx(j))
6406 			nr_kernel_pages += freesize;
6407 		/* Charge for highmem memmap if there are enough kernel pages */
6408 		else if (nr_kernel_pages > memmap_pages * 2)
6409 			nr_kernel_pages -= memmap_pages;
6410 		nr_all_pages += freesize;
6411 
6412 		/*
6413 		 * Set an approximate value for lowmem here, it will be adjusted
6414 		 * when the bootmem allocator frees pages into the buddy system.
6415 		 * And all highmem pages will be managed by the buddy system.
6416 		 */
6417 		zone_init_internals(zone, j, nid, freesize);
6418 
6419 		if (!size)
6420 			continue;
6421 
6422 		set_pageblock_order();
6423 		setup_usemap(pgdat, zone, zone_start_pfn, size);
6424 		init_currently_empty_zone(zone, zone_start_pfn, size);
6425 		memmap_init(size, nid, j, zone_start_pfn);
6426 	}
6427 }
6428 
6429 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6430 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6431 {
6432 	unsigned long __maybe_unused start = 0;
6433 	unsigned long __maybe_unused offset = 0;
6434 
6435 	/* Skip empty nodes */
6436 	if (!pgdat->node_spanned_pages)
6437 		return;
6438 
6439 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6440 	offset = pgdat->node_start_pfn - start;
6441 	/* ia64 gets its own node_mem_map, before this, without bootmem */
6442 	if (!pgdat->node_mem_map) {
6443 		unsigned long size, end;
6444 		struct page *map;
6445 
6446 		/*
6447 		 * The zone's endpoints aren't required to be MAX_ORDER
6448 		 * aligned but the node_mem_map endpoints must be in order
6449 		 * for the buddy allocator to function correctly.
6450 		 */
6451 		end = pgdat_end_pfn(pgdat);
6452 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
6453 		size =  (end - start) * sizeof(struct page);
6454 		map = memblock_alloc_node_nopanic(size, pgdat->node_id);
6455 		pgdat->node_mem_map = map + offset;
6456 	}
6457 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6458 				__func__, pgdat->node_id, (unsigned long)pgdat,
6459 				(unsigned long)pgdat->node_mem_map);
6460 #ifndef CONFIG_NEED_MULTIPLE_NODES
6461 	/*
6462 	 * With no DISCONTIG, the global mem_map is just set as node 0's
6463 	 */
6464 	if (pgdat == NODE_DATA(0)) {
6465 		mem_map = NODE_DATA(0)->node_mem_map;
6466 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6467 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6468 			mem_map -= offset;
6469 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6470 	}
6471 #endif
6472 }
6473 #else
6474 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6475 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6476 
6477 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6478 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
6479 {
6480 	/*
6481 	 * We start only with one section of pages, more pages are added as
6482 	 * needed until the rest of deferred pages are initialized.
6483 	 */
6484 	pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION,
6485 						pgdat->node_spanned_pages);
6486 	pgdat->first_deferred_pfn = ULONG_MAX;
6487 }
6488 #else
6489 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
6490 #endif
6491 
6492 void __init free_area_init_node(int nid, unsigned long *zones_size,
6493 				   unsigned long node_start_pfn,
6494 				   unsigned long *zholes_size)
6495 {
6496 	pg_data_t *pgdat = NODE_DATA(nid);
6497 	unsigned long start_pfn = 0;
6498 	unsigned long end_pfn = 0;
6499 
6500 	/* pg_data_t should be reset to zero when it's allocated */
6501 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
6502 
6503 	pgdat->node_id = nid;
6504 	pgdat->node_start_pfn = node_start_pfn;
6505 	pgdat->per_cpu_nodestats = NULL;
6506 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6507 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6508 	pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6509 		(u64)start_pfn << PAGE_SHIFT,
6510 		end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6511 #else
6512 	start_pfn = node_start_pfn;
6513 #endif
6514 	calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6515 				  zones_size, zholes_size);
6516 
6517 	alloc_node_mem_map(pgdat);
6518 	pgdat_set_deferred_range(pgdat);
6519 
6520 	free_area_init_core(pgdat);
6521 }
6522 
6523 #if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6524 /*
6525  * Zero all valid struct pages in range [spfn, epfn), return number of struct
6526  * pages zeroed
6527  */
6528 static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
6529 {
6530 	unsigned long pfn;
6531 	u64 pgcnt = 0;
6532 
6533 	for (pfn = spfn; pfn < epfn; pfn++) {
6534 		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6535 			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6536 				+ pageblock_nr_pages - 1;
6537 			continue;
6538 		}
6539 		mm_zero_struct_page(pfn_to_page(pfn));
6540 		pgcnt++;
6541 	}
6542 
6543 	return pgcnt;
6544 }
6545 
6546 /*
6547  * Only struct pages that are backed by physical memory are zeroed and
6548  * initialized by going through __init_single_page(). But, there are some
6549  * struct pages which are reserved in memblock allocator and their fields
6550  * may be accessed (for example page_to_pfn() on some configuration accesses
6551  * flags). We must explicitly zero those struct pages.
6552  *
6553  * This function also addresses a similar issue where struct pages are left
6554  * uninitialized because the physical address range is not covered by
6555  * memblock.memory or memblock.reserved. That could happen when memblock
6556  * layout is manually configured via memmap=.
6557  */
6558 void __init zero_resv_unavail(void)
6559 {
6560 	phys_addr_t start, end;
6561 	u64 i, pgcnt;
6562 	phys_addr_t next = 0;
6563 
6564 	/*
6565 	 * Loop through unavailable ranges not covered by memblock.memory.
6566 	 */
6567 	pgcnt = 0;
6568 	for_each_mem_range(i, &memblock.memory, NULL,
6569 			NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
6570 		if (next < start)
6571 			pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
6572 		next = end;
6573 	}
6574 	pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
6575 
6576 	/*
6577 	 * Struct pages that do not have backing memory. This could be because
6578 	 * firmware is using some of this memory, or for some other reasons.
6579 	 */
6580 	if (pgcnt)
6581 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
6582 }
6583 #endif /* !CONFIG_FLAT_NODE_MEM_MAP */
6584 
6585 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6586 
6587 #if MAX_NUMNODES > 1
6588 /*
6589  * Figure out the number of possible node ids.
6590  */
6591 void __init setup_nr_node_ids(void)
6592 {
6593 	unsigned int highest;
6594 
6595 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
6596 	nr_node_ids = highest + 1;
6597 }
6598 #endif
6599 
6600 /**
6601  * node_map_pfn_alignment - determine the maximum internode alignment
6602  *
6603  * This function should be called after node map is populated and sorted.
6604  * It calculates the maximum power of two alignment which can distinguish
6605  * all the nodes.
6606  *
6607  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6608  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
6609  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
6610  * shifted, 1GiB is enough and this function will indicate so.
6611  *
6612  * This is used to test whether pfn -> nid mapping of the chosen memory
6613  * model has fine enough granularity to avoid incorrect mapping for the
6614  * populated node map.
6615  *
6616  * Returns the determined alignment in pfn's.  0 if there is no alignment
6617  * requirement (single node).
6618  */
6619 unsigned long __init node_map_pfn_alignment(void)
6620 {
6621 	unsigned long accl_mask = 0, last_end = 0;
6622 	unsigned long start, end, mask;
6623 	int last_nid = -1;
6624 	int i, nid;
6625 
6626 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
6627 		if (!start || last_nid < 0 || last_nid == nid) {
6628 			last_nid = nid;
6629 			last_end = end;
6630 			continue;
6631 		}
6632 
6633 		/*
6634 		 * Start with a mask granular enough to pin-point to the
6635 		 * start pfn and tick off bits one-by-one until it becomes
6636 		 * too coarse to separate the current node from the last.
6637 		 */
6638 		mask = ~((1 << __ffs(start)) - 1);
6639 		while (mask && last_end <= (start & (mask << 1)))
6640 			mask <<= 1;
6641 
6642 		/* accumulate all internode masks */
6643 		accl_mask |= mask;
6644 	}
6645 
6646 	/* convert mask to number of pages */
6647 	return ~accl_mask + 1;
6648 }
6649 
6650 /* Find the lowest pfn for a node */
6651 static unsigned long __init find_min_pfn_for_node(int nid)
6652 {
6653 	unsigned long min_pfn = ULONG_MAX;
6654 	unsigned long start_pfn;
6655 	int i;
6656 
6657 	for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6658 		min_pfn = min(min_pfn, start_pfn);
6659 
6660 	if (min_pfn == ULONG_MAX) {
6661 		pr_warn("Could not find start_pfn for node %d\n", nid);
6662 		return 0;
6663 	}
6664 
6665 	return min_pfn;
6666 }
6667 
6668 /**
6669  * find_min_pfn_with_active_regions - Find the minimum PFN registered
6670  *
6671  * It returns the minimum PFN based on information provided via
6672  * memblock_set_node().
6673  */
6674 unsigned long __init find_min_pfn_with_active_regions(void)
6675 {
6676 	return find_min_pfn_for_node(MAX_NUMNODES);
6677 }
6678 
6679 /*
6680  * early_calculate_totalpages()
6681  * Sum pages in active regions for movable zone.
6682  * Populate N_MEMORY for calculating usable_nodes.
6683  */
6684 static unsigned long __init early_calculate_totalpages(void)
6685 {
6686 	unsigned long totalpages = 0;
6687 	unsigned long start_pfn, end_pfn;
6688 	int i, nid;
6689 
6690 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6691 		unsigned long pages = end_pfn - start_pfn;
6692 
6693 		totalpages += pages;
6694 		if (pages)
6695 			node_set_state(nid, N_MEMORY);
6696 	}
6697 	return totalpages;
6698 }
6699 
6700 /*
6701  * Find the PFN the Movable zone begins in each node. Kernel memory
6702  * is spread evenly between nodes as long as the nodes have enough
6703  * memory. When they don't, some nodes will have more kernelcore than
6704  * others
6705  */
6706 static void __init find_zone_movable_pfns_for_nodes(void)
6707 {
6708 	int i, nid;
6709 	unsigned long usable_startpfn;
6710 	unsigned long kernelcore_node, kernelcore_remaining;
6711 	/* save the state before borrow the nodemask */
6712 	nodemask_t saved_node_state = node_states[N_MEMORY];
6713 	unsigned long totalpages = early_calculate_totalpages();
6714 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
6715 	struct memblock_region *r;
6716 
6717 	/* Need to find movable_zone earlier when movable_node is specified. */
6718 	find_usable_zone_for_movable();
6719 
6720 	/*
6721 	 * If movable_node is specified, ignore kernelcore and movablecore
6722 	 * options.
6723 	 */
6724 	if (movable_node_is_enabled()) {
6725 		for_each_memblock(memory, r) {
6726 			if (!memblock_is_hotpluggable(r))
6727 				continue;
6728 
6729 			nid = r->nid;
6730 
6731 			usable_startpfn = PFN_DOWN(r->base);
6732 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6733 				min(usable_startpfn, zone_movable_pfn[nid]) :
6734 				usable_startpfn;
6735 		}
6736 
6737 		goto out2;
6738 	}
6739 
6740 	/*
6741 	 * If kernelcore=mirror is specified, ignore movablecore option
6742 	 */
6743 	if (mirrored_kernelcore) {
6744 		bool mem_below_4gb_not_mirrored = false;
6745 
6746 		for_each_memblock(memory, r) {
6747 			if (memblock_is_mirror(r))
6748 				continue;
6749 
6750 			nid = r->nid;
6751 
6752 			usable_startpfn = memblock_region_memory_base_pfn(r);
6753 
6754 			if (usable_startpfn < 0x100000) {
6755 				mem_below_4gb_not_mirrored = true;
6756 				continue;
6757 			}
6758 
6759 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6760 				min(usable_startpfn, zone_movable_pfn[nid]) :
6761 				usable_startpfn;
6762 		}
6763 
6764 		if (mem_below_4gb_not_mirrored)
6765 			pr_warn("This configuration results in unmirrored kernel memory.");
6766 
6767 		goto out2;
6768 	}
6769 
6770 	/*
6771 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
6772 	 * amount of necessary memory.
6773 	 */
6774 	if (required_kernelcore_percent)
6775 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
6776 				       10000UL;
6777 	if (required_movablecore_percent)
6778 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
6779 					10000UL;
6780 
6781 	/*
6782 	 * If movablecore= was specified, calculate what size of
6783 	 * kernelcore that corresponds so that memory usable for
6784 	 * any allocation type is evenly spread. If both kernelcore
6785 	 * and movablecore are specified, then the value of kernelcore
6786 	 * will be used for required_kernelcore if it's greater than
6787 	 * what movablecore would have allowed.
6788 	 */
6789 	if (required_movablecore) {
6790 		unsigned long corepages;
6791 
6792 		/*
6793 		 * Round-up so that ZONE_MOVABLE is at least as large as what
6794 		 * was requested by the user
6795 		 */
6796 		required_movablecore =
6797 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
6798 		required_movablecore = min(totalpages, required_movablecore);
6799 		corepages = totalpages - required_movablecore;
6800 
6801 		required_kernelcore = max(required_kernelcore, corepages);
6802 	}
6803 
6804 	/*
6805 	 * If kernelcore was not specified or kernelcore size is larger
6806 	 * than totalpages, there is no ZONE_MOVABLE.
6807 	 */
6808 	if (!required_kernelcore || required_kernelcore >= totalpages)
6809 		goto out;
6810 
6811 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6812 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6813 
6814 restart:
6815 	/* Spread kernelcore memory as evenly as possible throughout nodes */
6816 	kernelcore_node = required_kernelcore / usable_nodes;
6817 	for_each_node_state(nid, N_MEMORY) {
6818 		unsigned long start_pfn, end_pfn;
6819 
6820 		/*
6821 		 * Recalculate kernelcore_node if the division per node
6822 		 * now exceeds what is necessary to satisfy the requested
6823 		 * amount of memory for the kernel
6824 		 */
6825 		if (required_kernelcore < kernelcore_node)
6826 			kernelcore_node = required_kernelcore / usable_nodes;
6827 
6828 		/*
6829 		 * As the map is walked, we track how much memory is usable
6830 		 * by the kernel using kernelcore_remaining. When it is
6831 		 * 0, the rest of the node is usable by ZONE_MOVABLE
6832 		 */
6833 		kernelcore_remaining = kernelcore_node;
6834 
6835 		/* Go through each range of PFNs within this node */
6836 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6837 			unsigned long size_pages;
6838 
6839 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
6840 			if (start_pfn >= end_pfn)
6841 				continue;
6842 
6843 			/* Account for what is only usable for kernelcore */
6844 			if (start_pfn < usable_startpfn) {
6845 				unsigned long kernel_pages;
6846 				kernel_pages = min(end_pfn, usable_startpfn)
6847 								- start_pfn;
6848 
6849 				kernelcore_remaining -= min(kernel_pages,
6850 							kernelcore_remaining);
6851 				required_kernelcore -= min(kernel_pages,
6852 							required_kernelcore);
6853 
6854 				/* Continue if range is now fully accounted */
6855 				if (end_pfn <= usable_startpfn) {
6856 
6857 					/*
6858 					 * Push zone_movable_pfn to the end so
6859 					 * that if we have to rebalance
6860 					 * kernelcore across nodes, we will
6861 					 * not double account here
6862 					 */
6863 					zone_movable_pfn[nid] = end_pfn;
6864 					continue;
6865 				}
6866 				start_pfn = usable_startpfn;
6867 			}
6868 
6869 			/*
6870 			 * The usable PFN range for ZONE_MOVABLE is from
6871 			 * start_pfn->end_pfn. Calculate size_pages as the
6872 			 * number of pages used as kernelcore
6873 			 */
6874 			size_pages = end_pfn - start_pfn;
6875 			if (size_pages > kernelcore_remaining)
6876 				size_pages = kernelcore_remaining;
6877 			zone_movable_pfn[nid] = start_pfn + size_pages;
6878 
6879 			/*
6880 			 * Some kernelcore has been met, update counts and
6881 			 * break if the kernelcore for this node has been
6882 			 * satisfied
6883 			 */
6884 			required_kernelcore -= min(required_kernelcore,
6885 								size_pages);
6886 			kernelcore_remaining -= size_pages;
6887 			if (!kernelcore_remaining)
6888 				break;
6889 		}
6890 	}
6891 
6892 	/*
6893 	 * If there is still required_kernelcore, we do another pass with one
6894 	 * less node in the count. This will push zone_movable_pfn[nid] further
6895 	 * along on the nodes that still have memory until kernelcore is
6896 	 * satisfied
6897 	 */
6898 	usable_nodes--;
6899 	if (usable_nodes && required_kernelcore > usable_nodes)
6900 		goto restart;
6901 
6902 out2:
6903 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6904 	for (nid = 0; nid < MAX_NUMNODES; nid++)
6905 		zone_movable_pfn[nid] =
6906 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
6907 
6908 out:
6909 	/* restore the node_state */
6910 	node_states[N_MEMORY] = saved_node_state;
6911 }
6912 
6913 /* Any regular or high memory on that node ? */
6914 static void check_for_memory(pg_data_t *pgdat, int nid)
6915 {
6916 	enum zone_type zone_type;
6917 
6918 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
6919 		struct zone *zone = &pgdat->node_zones[zone_type];
6920 		if (populated_zone(zone)) {
6921 			if (IS_ENABLED(CONFIG_HIGHMEM))
6922 				node_set_state(nid, N_HIGH_MEMORY);
6923 			if (zone_type <= ZONE_NORMAL)
6924 				node_set_state(nid, N_NORMAL_MEMORY);
6925 			break;
6926 		}
6927 	}
6928 }
6929 
6930 /**
6931  * free_area_init_nodes - Initialise all pg_data_t and zone data
6932  * @max_zone_pfn: an array of max PFNs for each zone
6933  *
6934  * This will call free_area_init_node() for each active node in the system.
6935  * Using the page ranges provided by memblock_set_node(), the size of each
6936  * zone in each node and their holes is calculated. If the maximum PFN
6937  * between two adjacent zones match, it is assumed that the zone is empty.
6938  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6939  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6940  * starts where the previous one ended. For example, ZONE_DMA32 starts
6941  * at arch_max_dma_pfn.
6942  */
6943 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6944 {
6945 	unsigned long start_pfn, end_pfn;
6946 	int i, nid;
6947 
6948 	/* Record where the zone boundaries are */
6949 	memset(arch_zone_lowest_possible_pfn, 0,
6950 				sizeof(arch_zone_lowest_possible_pfn));
6951 	memset(arch_zone_highest_possible_pfn, 0,
6952 				sizeof(arch_zone_highest_possible_pfn));
6953 
6954 	start_pfn = find_min_pfn_with_active_regions();
6955 
6956 	for (i = 0; i < MAX_NR_ZONES; i++) {
6957 		if (i == ZONE_MOVABLE)
6958 			continue;
6959 
6960 		end_pfn = max(max_zone_pfn[i], start_pfn);
6961 		arch_zone_lowest_possible_pfn[i] = start_pfn;
6962 		arch_zone_highest_possible_pfn[i] = end_pfn;
6963 
6964 		start_pfn = end_pfn;
6965 	}
6966 
6967 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
6968 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
6969 	find_zone_movable_pfns_for_nodes();
6970 
6971 	/* Print out the zone ranges */
6972 	pr_info("Zone ranges:\n");
6973 	for (i = 0; i < MAX_NR_ZONES; i++) {
6974 		if (i == ZONE_MOVABLE)
6975 			continue;
6976 		pr_info("  %-8s ", zone_names[i]);
6977 		if (arch_zone_lowest_possible_pfn[i] ==
6978 				arch_zone_highest_possible_pfn[i])
6979 			pr_cont("empty\n");
6980 		else
6981 			pr_cont("[mem %#018Lx-%#018Lx]\n",
6982 				(u64)arch_zone_lowest_possible_pfn[i]
6983 					<< PAGE_SHIFT,
6984 				((u64)arch_zone_highest_possible_pfn[i]
6985 					<< PAGE_SHIFT) - 1);
6986 	}
6987 
6988 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
6989 	pr_info("Movable zone start for each node\n");
6990 	for (i = 0; i < MAX_NUMNODES; i++) {
6991 		if (zone_movable_pfn[i])
6992 			pr_info("  Node %d: %#018Lx\n", i,
6993 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6994 	}
6995 
6996 	/* Print out the early node map */
6997 	pr_info("Early memory node ranges\n");
6998 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6999 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7000 			(u64)start_pfn << PAGE_SHIFT,
7001 			((u64)end_pfn << PAGE_SHIFT) - 1);
7002 
7003 	/* Initialise every node */
7004 	mminit_verify_pageflags_layout();
7005 	setup_nr_node_ids();
7006 	zero_resv_unavail();
7007 	for_each_online_node(nid) {
7008 		pg_data_t *pgdat = NODE_DATA(nid);
7009 		free_area_init_node(nid, NULL,
7010 				find_min_pfn_for_node(nid), NULL);
7011 
7012 		/* Any memory on that node */
7013 		if (pgdat->node_present_pages)
7014 			node_set_state(nid, N_MEMORY);
7015 		check_for_memory(pgdat, nid);
7016 	}
7017 }
7018 
7019 static int __init cmdline_parse_core(char *p, unsigned long *core,
7020 				     unsigned long *percent)
7021 {
7022 	unsigned long long coremem;
7023 	char *endptr;
7024 
7025 	if (!p)
7026 		return -EINVAL;
7027 
7028 	/* Value may be a percentage of total memory, otherwise bytes */
7029 	coremem = simple_strtoull(p, &endptr, 0);
7030 	if (*endptr == '%') {
7031 		/* Paranoid check for percent values greater than 100 */
7032 		WARN_ON(coremem > 100);
7033 
7034 		*percent = coremem;
7035 	} else {
7036 		coremem = memparse(p, &p);
7037 		/* Paranoid check that UL is enough for the coremem value */
7038 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7039 
7040 		*core = coremem >> PAGE_SHIFT;
7041 		*percent = 0UL;
7042 	}
7043 	return 0;
7044 }
7045 
7046 /*
7047  * kernelcore=size sets the amount of memory for use for allocations that
7048  * cannot be reclaimed or migrated.
7049  */
7050 static int __init cmdline_parse_kernelcore(char *p)
7051 {
7052 	/* parse kernelcore=mirror */
7053 	if (parse_option_str(p, "mirror")) {
7054 		mirrored_kernelcore = true;
7055 		return 0;
7056 	}
7057 
7058 	return cmdline_parse_core(p, &required_kernelcore,
7059 				  &required_kernelcore_percent);
7060 }
7061 
7062 /*
7063  * movablecore=size sets the amount of memory for use for allocations that
7064  * can be reclaimed or migrated.
7065  */
7066 static int __init cmdline_parse_movablecore(char *p)
7067 {
7068 	return cmdline_parse_core(p, &required_movablecore,
7069 				  &required_movablecore_percent);
7070 }
7071 
7072 early_param("kernelcore", cmdline_parse_kernelcore);
7073 early_param("movablecore", cmdline_parse_movablecore);
7074 
7075 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
7076 
7077 void adjust_managed_page_count(struct page *page, long count)
7078 {
7079 	spin_lock(&managed_page_count_lock);
7080 	page_zone(page)->managed_pages += count;
7081 	totalram_pages += count;
7082 #ifdef CONFIG_HIGHMEM
7083 	if (PageHighMem(page))
7084 		totalhigh_pages += count;
7085 #endif
7086 	spin_unlock(&managed_page_count_lock);
7087 }
7088 EXPORT_SYMBOL(adjust_managed_page_count);
7089 
7090 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
7091 {
7092 	void *pos;
7093 	unsigned long pages = 0;
7094 
7095 	start = (void *)PAGE_ALIGN((unsigned long)start);
7096 	end = (void *)((unsigned long)end & PAGE_MASK);
7097 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7098 		struct page *page = virt_to_page(pos);
7099 		void *direct_map_addr;
7100 
7101 		/*
7102 		 * 'direct_map_addr' might be different from 'pos'
7103 		 * because some architectures' virt_to_page()
7104 		 * work with aliases.  Getting the direct map
7105 		 * address ensures that we get a _writeable_
7106 		 * alias for the memset().
7107 		 */
7108 		direct_map_addr = page_address(page);
7109 		if ((unsigned int)poison <= 0xFF)
7110 			memset(direct_map_addr, poison, PAGE_SIZE);
7111 
7112 		free_reserved_page(page);
7113 	}
7114 
7115 	if (pages && s)
7116 		pr_info("Freeing %s memory: %ldK\n",
7117 			s, pages << (PAGE_SHIFT - 10));
7118 
7119 	return pages;
7120 }
7121 EXPORT_SYMBOL(free_reserved_area);
7122 
7123 #ifdef	CONFIG_HIGHMEM
7124 void free_highmem_page(struct page *page)
7125 {
7126 	__free_reserved_page(page);
7127 	totalram_pages++;
7128 	page_zone(page)->managed_pages++;
7129 	totalhigh_pages++;
7130 }
7131 #endif
7132 
7133 
7134 void __init mem_init_print_info(const char *str)
7135 {
7136 	unsigned long physpages, codesize, datasize, rosize, bss_size;
7137 	unsigned long init_code_size, init_data_size;
7138 
7139 	physpages = get_num_physpages();
7140 	codesize = _etext - _stext;
7141 	datasize = _edata - _sdata;
7142 	rosize = __end_rodata - __start_rodata;
7143 	bss_size = __bss_stop - __bss_start;
7144 	init_data_size = __init_end - __init_begin;
7145 	init_code_size = _einittext - _sinittext;
7146 
7147 	/*
7148 	 * Detect special cases and adjust section sizes accordingly:
7149 	 * 1) .init.* may be embedded into .data sections
7150 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
7151 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
7152 	 * 3) .rodata.* may be embedded into .text or .data sections.
7153 	 */
7154 #define adj_init_size(start, end, size, pos, adj) \
7155 	do { \
7156 		if (start <= pos && pos < end && size > adj) \
7157 			size -= adj; \
7158 	} while (0)
7159 
7160 	adj_init_size(__init_begin, __init_end, init_data_size,
7161 		     _sinittext, init_code_size);
7162 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7163 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7164 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7165 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7166 
7167 #undef	adj_init_size
7168 
7169 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7170 #ifdef	CONFIG_HIGHMEM
7171 		", %luK highmem"
7172 #endif
7173 		"%s%s)\n",
7174 		nr_free_pages() << (PAGE_SHIFT - 10),
7175 		physpages << (PAGE_SHIFT - 10),
7176 		codesize >> 10, datasize >> 10, rosize >> 10,
7177 		(init_data_size + init_code_size) >> 10, bss_size >> 10,
7178 		(physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
7179 		totalcma_pages << (PAGE_SHIFT - 10),
7180 #ifdef	CONFIG_HIGHMEM
7181 		totalhigh_pages << (PAGE_SHIFT - 10),
7182 #endif
7183 		str ? ", " : "", str ? str : "");
7184 }
7185 
7186 /**
7187  * set_dma_reserve - set the specified number of pages reserved in the first zone
7188  * @new_dma_reserve: The number of pages to mark reserved
7189  *
7190  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7191  * In the DMA zone, a significant percentage may be consumed by kernel image
7192  * and other unfreeable allocations which can skew the watermarks badly. This
7193  * function may optionally be used to account for unfreeable pages in the
7194  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7195  * smaller per-cpu batchsize.
7196  */
7197 void __init set_dma_reserve(unsigned long new_dma_reserve)
7198 {
7199 	dma_reserve = new_dma_reserve;
7200 }
7201 
7202 void __init free_area_init(unsigned long *zones_size)
7203 {
7204 	zero_resv_unavail();
7205 	free_area_init_node(0, zones_size,
7206 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
7207 }
7208 
7209 static int page_alloc_cpu_dead(unsigned int cpu)
7210 {
7211 
7212 	lru_add_drain_cpu(cpu);
7213 	drain_pages(cpu);
7214 
7215 	/*
7216 	 * Spill the event counters of the dead processor
7217 	 * into the current processors event counters.
7218 	 * This artificially elevates the count of the current
7219 	 * processor.
7220 	 */
7221 	vm_events_fold_cpu(cpu);
7222 
7223 	/*
7224 	 * Zero the differential counters of the dead processor
7225 	 * so that the vm statistics are consistent.
7226 	 *
7227 	 * This is only okay since the processor is dead and cannot
7228 	 * race with what we are doing.
7229 	 */
7230 	cpu_vm_stats_fold(cpu);
7231 	return 0;
7232 }
7233 
7234 void __init page_alloc_init(void)
7235 {
7236 	int ret;
7237 
7238 	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7239 					"mm/page_alloc:dead", NULL,
7240 					page_alloc_cpu_dead);
7241 	WARN_ON(ret < 0);
7242 }
7243 
7244 /*
7245  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7246  *	or min_free_kbytes changes.
7247  */
7248 static void calculate_totalreserve_pages(void)
7249 {
7250 	struct pglist_data *pgdat;
7251 	unsigned long reserve_pages = 0;
7252 	enum zone_type i, j;
7253 
7254 	for_each_online_pgdat(pgdat) {
7255 
7256 		pgdat->totalreserve_pages = 0;
7257 
7258 		for (i = 0; i < MAX_NR_ZONES; i++) {
7259 			struct zone *zone = pgdat->node_zones + i;
7260 			long max = 0;
7261 
7262 			/* Find valid and maximum lowmem_reserve in the zone */
7263 			for (j = i; j < MAX_NR_ZONES; j++) {
7264 				if (zone->lowmem_reserve[j] > max)
7265 					max = zone->lowmem_reserve[j];
7266 			}
7267 
7268 			/* we treat the high watermark as reserved pages. */
7269 			max += high_wmark_pages(zone);
7270 
7271 			if (max > zone->managed_pages)
7272 				max = zone->managed_pages;
7273 
7274 			pgdat->totalreserve_pages += max;
7275 
7276 			reserve_pages += max;
7277 		}
7278 	}
7279 	totalreserve_pages = reserve_pages;
7280 }
7281 
7282 /*
7283  * setup_per_zone_lowmem_reserve - called whenever
7284  *	sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
7285  *	has a correct pages reserved value, so an adequate number of
7286  *	pages are left in the zone after a successful __alloc_pages().
7287  */
7288 static void setup_per_zone_lowmem_reserve(void)
7289 {
7290 	struct pglist_data *pgdat;
7291 	enum zone_type j, idx;
7292 
7293 	for_each_online_pgdat(pgdat) {
7294 		for (j = 0; j < MAX_NR_ZONES; j++) {
7295 			struct zone *zone = pgdat->node_zones + j;
7296 			unsigned long managed_pages = zone->managed_pages;
7297 
7298 			zone->lowmem_reserve[j] = 0;
7299 
7300 			idx = j;
7301 			while (idx) {
7302 				struct zone *lower_zone;
7303 
7304 				idx--;
7305 				lower_zone = pgdat->node_zones + idx;
7306 
7307 				if (sysctl_lowmem_reserve_ratio[idx] < 1) {
7308 					sysctl_lowmem_reserve_ratio[idx] = 0;
7309 					lower_zone->lowmem_reserve[j] = 0;
7310 				} else {
7311 					lower_zone->lowmem_reserve[j] =
7312 						managed_pages / sysctl_lowmem_reserve_ratio[idx];
7313 				}
7314 				managed_pages += lower_zone->managed_pages;
7315 			}
7316 		}
7317 	}
7318 
7319 	/* update totalreserve_pages */
7320 	calculate_totalreserve_pages();
7321 }
7322 
7323 static void __setup_per_zone_wmarks(void)
7324 {
7325 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7326 	unsigned long lowmem_pages = 0;
7327 	struct zone *zone;
7328 	unsigned long flags;
7329 
7330 	/* Calculate total number of !ZONE_HIGHMEM pages */
7331 	for_each_zone(zone) {
7332 		if (!is_highmem(zone))
7333 			lowmem_pages += zone->managed_pages;
7334 	}
7335 
7336 	for_each_zone(zone) {
7337 		u64 tmp;
7338 
7339 		spin_lock_irqsave(&zone->lock, flags);
7340 		tmp = (u64)pages_min * zone->managed_pages;
7341 		do_div(tmp, lowmem_pages);
7342 		if (is_highmem(zone)) {
7343 			/*
7344 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7345 			 * need highmem pages, so cap pages_min to a small
7346 			 * value here.
7347 			 *
7348 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7349 			 * deltas control asynch page reclaim, and so should
7350 			 * not be capped for highmem.
7351 			 */
7352 			unsigned long min_pages;
7353 
7354 			min_pages = zone->managed_pages / 1024;
7355 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7356 			zone->watermark[WMARK_MIN] = min_pages;
7357 		} else {
7358 			/*
7359 			 * If it's a lowmem zone, reserve a number of pages
7360 			 * proportionate to the zone's size.
7361 			 */
7362 			zone->watermark[WMARK_MIN] = tmp;
7363 		}
7364 
7365 		/*
7366 		 * Set the kswapd watermarks distance according to the
7367 		 * scale factor in proportion to available memory, but
7368 		 * ensure a minimum size on small systems.
7369 		 */
7370 		tmp = max_t(u64, tmp >> 2,
7371 			    mult_frac(zone->managed_pages,
7372 				      watermark_scale_factor, 10000));
7373 
7374 		zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
7375 		zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7376 
7377 		spin_unlock_irqrestore(&zone->lock, flags);
7378 	}
7379 
7380 	/* update totalreserve_pages */
7381 	calculate_totalreserve_pages();
7382 }
7383 
7384 /**
7385  * setup_per_zone_wmarks - called when min_free_kbytes changes
7386  * or when memory is hot-{added|removed}
7387  *
7388  * Ensures that the watermark[min,low,high] values for each zone are set
7389  * correctly with respect to min_free_kbytes.
7390  */
7391 void setup_per_zone_wmarks(void)
7392 {
7393 	static DEFINE_SPINLOCK(lock);
7394 
7395 	spin_lock(&lock);
7396 	__setup_per_zone_wmarks();
7397 	spin_unlock(&lock);
7398 }
7399 
7400 /*
7401  * Initialise min_free_kbytes.
7402  *
7403  * For small machines we want it small (128k min).  For large machines
7404  * we want it large (64MB max).  But it is not linear, because network
7405  * bandwidth does not increase linearly with machine size.  We use
7406  *
7407  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7408  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
7409  *
7410  * which yields
7411  *
7412  * 16MB:	512k
7413  * 32MB:	724k
7414  * 64MB:	1024k
7415  * 128MB:	1448k
7416  * 256MB:	2048k
7417  * 512MB:	2896k
7418  * 1024MB:	4096k
7419  * 2048MB:	5792k
7420  * 4096MB:	8192k
7421  * 8192MB:	11584k
7422  * 16384MB:	16384k
7423  */
7424 int __meminit init_per_zone_wmark_min(void)
7425 {
7426 	unsigned long lowmem_kbytes;
7427 	int new_min_free_kbytes;
7428 
7429 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7430 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7431 
7432 	if (new_min_free_kbytes > user_min_free_kbytes) {
7433 		min_free_kbytes = new_min_free_kbytes;
7434 		if (min_free_kbytes < 128)
7435 			min_free_kbytes = 128;
7436 		if (min_free_kbytes > 65536)
7437 			min_free_kbytes = 65536;
7438 	} else {
7439 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7440 				new_min_free_kbytes, user_min_free_kbytes);
7441 	}
7442 	setup_per_zone_wmarks();
7443 	refresh_zone_stat_thresholds();
7444 	setup_per_zone_lowmem_reserve();
7445 
7446 #ifdef CONFIG_NUMA
7447 	setup_min_unmapped_ratio();
7448 	setup_min_slab_ratio();
7449 #endif
7450 
7451 	return 0;
7452 }
7453 core_initcall(init_per_zone_wmark_min)
7454 
7455 /*
7456  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
7457  *	that we can call two helper functions whenever min_free_kbytes
7458  *	changes.
7459  */
7460 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7461 	void __user *buffer, size_t *length, loff_t *ppos)
7462 {
7463 	int rc;
7464 
7465 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7466 	if (rc)
7467 		return rc;
7468 
7469 	if (write) {
7470 		user_min_free_kbytes = min_free_kbytes;
7471 		setup_per_zone_wmarks();
7472 	}
7473 	return 0;
7474 }
7475 
7476 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7477 	void __user *buffer, size_t *length, loff_t *ppos)
7478 {
7479 	int rc;
7480 
7481 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7482 	if (rc)
7483 		return rc;
7484 
7485 	if (write)
7486 		setup_per_zone_wmarks();
7487 
7488 	return 0;
7489 }
7490 
7491 #ifdef CONFIG_NUMA
7492 static void setup_min_unmapped_ratio(void)
7493 {
7494 	pg_data_t *pgdat;
7495 	struct zone *zone;
7496 
7497 	for_each_online_pgdat(pgdat)
7498 		pgdat->min_unmapped_pages = 0;
7499 
7500 	for_each_zone(zone)
7501 		zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
7502 				sysctl_min_unmapped_ratio) / 100;
7503 }
7504 
7505 
7506 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7507 	void __user *buffer, size_t *length, loff_t *ppos)
7508 {
7509 	int rc;
7510 
7511 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7512 	if (rc)
7513 		return rc;
7514 
7515 	setup_min_unmapped_ratio();
7516 
7517 	return 0;
7518 }
7519 
7520 static void setup_min_slab_ratio(void)
7521 {
7522 	pg_data_t *pgdat;
7523 	struct zone *zone;
7524 
7525 	for_each_online_pgdat(pgdat)
7526 		pgdat->min_slab_pages = 0;
7527 
7528 	for_each_zone(zone)
7529 		zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
7530 				sysctl_min_slab_ratio) / 100;
7531 }
7532 
7533 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7534 	void __user *buffer, size_t *length, loff_t *ppos)
7535 {
7536 	int rc;
7537 
7538 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7539 	if (rc)
7540 		return rc;
7541 
7542 	setup_min_slab_ratio();
7543 
7544 	return 0;
7545 }
7546 #endif
7547 
7548 /*
7549  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7550  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7551  *	whenever sysctl_lowmem_reserve_ratio changes.
7552  *
7553  * The reserve ratio obviously has absolutely no relation with the
7554  * minimum watermarks. The lowmem reserve ratio can only make sense
7555  * if in function of the boot time zone sizes.
7556  */
7557 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
7558 	void __user *buffer, size_t *length, loff_t *ppos)
7559 {
7560 	proc_dointvec_minmax(table, write, buffer, length, ppos);
7561 	setup_per_zone_lowmem_reserve();
7562 	return 0;
7563 }
7564 
7565 /*
7566  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7567  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
7568  * pagelist can have before it gets flushed back to buddy allocator.
7569  */
7570 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
7571 	void __user *buffer, size_t *length, loff_t *ppos)
7572 {
7573 	struct zone *zone;
7574 	int old_percpu_pagelist_fraction;
7575 	int ret;
7576 
7577 	mutex_lock(&pcp_batch_high_lock);
7578 	old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7579 
7580 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7581 	if (!write || ret < 0)
7582 		goto out;
7583 
7584 	/* Sanity checking to avoid pcp imbalance */
7585 	if (percpu_pagelist_fraction &&
7586 	    percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7587 		percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7588 		ret = -EINVAL;
7589 		goto out;
7590 	}
7591 
7592 	/* No change? */
7593 	if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7594 		goto out;
7595 
7596 	for_each_populated_zone(zone) {
7597 		unsigned int cpu;
7598 
7599 		for_each_possible_cpu(cpu)
7600 			pageset_set_high_and_batch(zone,
7601 					per_cpu_ptr(zone->pageset, cpu));
7602 	}
7603 out:
7604 	mutex_unlock(&pcp_batch_high_lock);
7605 	return ret;
7606 }
7607 
7608 #ifdef CONFIG_NUMA
7609 int hashdist = HASHDIST_DEFAULT;
7610 
7611 static int __init set_hashdist(char *str)
7612 {
7613 	if (!str)
7614 		return 0;
7615 	hashdist = simple_strtoul(str, &str, 0);
7616 	return 1;
7617 }
7618 __setup("hashdist=", set_hashdist);
7619 #endif
7620 
7621 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7622 /*
7623  * Returns the number of pages that arch has reserved but
7624  * is not known to alloc_large_system_hash().
7625  */
7626 static unsigned long __init arch_reserved_kernel_pages(void)
7627 {
7628 	return 0;
7629 }
7630 #endif
7631 
7632 /*
7633  * Adaptive scale is meant to reduce sizes of hash tables on large memory
7634  * machines. As memory size is increased the scale is also increased but at
7635  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
7636  * quadruples the scale is increased by one, which means the size of hash table
7637  * only doubles, instead of quadrupling as well.
7638  * Because 32-bit systems cannot have large physical memory, where this scaling
7639  * makes sense, it is disabled on such platforms.
7640  */
7641 #if __BITS_PER_LONG > 32
7642 #define ADAPT_SCALE_BASE	(64ul << 30)
7643 #define ADAPT_SCALE_SHIFT	2
7644 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
7645 #endif
7646 
7647 /*
7648  * allocate a large system hash table from bootmem
7649  * - it is assumed that the hash table must contain an exact power-of-2
7650  *   quantity of entries
7651  * - limit is the number of hash buckets, not the total allocation size
7652  */
7653 void *__init alloc_large_system_hash(const char *tablename,
7654 				     unsigned long bucketsize,
7655 				     unsigned long numentries,
7656 				     int scale,
7657 				     int flags,
7658 				     unsigned int *_hash_shift,
7659 				     unsigned int *_hash_mask,
7660 				     unsigned long low_limit,
7661 				     unsigned long high_limit)
7662 {
7663 	unsigned long long max = high_limit;
7664 	unsigned long log2qty, size;
7665 	void *table = NULL;
7666 	gfp_t gfp_flags;
7667 
7668 	/* allow the kernel cmdline to have a say */
7669 	if (!numentries) {
7670 		/* round applicable memory size up to nearest megabyte */
7671 		numentries = nr_kernel_pages;
7672 		numentries -= arch_reserved_kernel_pages();
7673 
7674 		/* It isn't necessary when PAGE_SIZE >= 1MB */
7675 		if (PAGE_SHIFT < 20)
7676 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
7677 
7678 #if __BITS_PER_LONG > 32
7679 		if (!high_limit) {
7680 			unsigned long adapt;
7681 
7682 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
7683 			     adapt <<= ADAPT_SCALE_SHIFT)
7684 				scale++;
7685 		}
7686 #endif
7687 
7688 		/* limit to 1 bucket per 2^scale bytes of low memory */
7689 		if (scale > PAGE_SHIFT)
7690 			numentries >>= (scale - PAGE_SHIFT);
7691 		else
7692 			numentries <<= (PAGE_SHIFT - scale);
7693 
7694 		/* Make sure we've got at least a 0-order allocation.. */
7695 		if (unlikely(flags & HASH_SMALL)) {
7696 			/* Makes no sense without HASH_EARLY */
7697 			WARN_ON(!(flags & HASH_EARLY));
7698 			if (!(numentries >> *_hash_shift)) {
7699 				numentries = 1UL << *_hash_shift;
7700 				BUG_ON(!numentries);
7701 			}
7702 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
7703 			numentries = PAGE_SIZE / bucketsize;
7704 	}
7705 	numentries = roundup_pow_of_two(numentries);
7706 
7707 	/* limit allocation size to 1/16 total memory by default */
7708 	if (max == 0) {
7709 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7710 		do_div(max, bucketsize);
7711 	}
7712 	max = min(max, 0x80000000ULL);
7713 
7714 	if (numentries < low_limit)
7715 		numentries = low_limit;
7716 	if (numentries > max)
7717 		numentries = max;
7718 
7719 	log2qty = ilog2(numentries);
7720 
7721 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
7722 	do {
7723 		size = bucketsize << log2qty;
7724 		if (flags & HASH_EARLY) {
7725 			if (flags & HASH_ZERO)
7726 				table = memblock_alloc_nopanic(size,
7727 							       SMP_CACHE_BYTES);
7728 			else
7729 				table = memblock_alloc_raw(size,
7730 							   SMP_CACHE_BYTES);
7731 		} else if (hashdist) {
7732 			table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
7733 		} else {
7734 			/*
7735 			 * If bucketsize is not a power-of-two, we may free
7736 			 * some pages at the end of hash table which
7737 			 * alloc_pages_exact() automatically does
7738 			 */
7739 			if (get_order(size) < MAX_ORDER) {
7740 				table = alloc_pages_exact(size, gfp_flags);
7741 				kmemleak_alloc(table, size, 1, gfp_flags);
7742 			}
7743 		}
7744 	} while (!table && size > PAGE_SIZE && --log2qty);
7745 
7746 	if (!table)
7747 		panic("Failed to allocate %s hash table\n", tablename);
7748 
7749 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7750 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
7751 
7752 	if (_hash_shift)
7753 		*_hash_shift = log2qty;
7754 	if (_hash_mask)
7755 		*_hash_mask = (1 << log2qty) - 1;
7756 
7757 	return table;
7758 }
7759 
7760 /*
7761  * This function checks whether pageblock includes unmovable pages or not.
7762  * If @count is not zero, it is okay to include less @count unmovable pages
7763  *
7764  * PageLRU check without isolation or lru_lock could race so that
7765  * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7766  * check without lock_page also may miss some movable non-lru pages at
7767  * race condition. So you can't expect this function should be exact.
7768  */
7769 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7770 			 int migratetype,
7771 			 bool skip_hwpoisoned_pages)
7772 {
7773 	unsigned long pfn, iter, found;
7774 
7775 	/*
7776 	 * TODO we could make this much more efficient by not checking every
7777 	 * page in the range if we know all of them are in MOVABLE_ZONE and
7778 	 * that the movable zone guarantees that pages are migratable but
7779 	 * the later is not the case right now unfortunatelly. E.g. movablecore
7780 	 * can still lead to having bootmem allocations in zone_movable.
7781 	 */
7782 
7783 	/*
7784 	 * CMA allocations (alloc_contig_range) really need to mark isolate
7785 	 * CMA pageblocks even when they are not movable in fact so consider
7786 	 * them movable here.
7787 	 */
7788 	if (is_migrate_cma(migratetype) &&
7789 			is_migrate_cma(get_pageblock_migratetype(page)))
7790 		return false;
7791 
7792 	pfn = page_to_pfn(page);
7793 	for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7794 		unsigned long check = pfn + iter;
7795 
7796 		if (!pfn_valid_within(check))
7797 			continue;
7798 
7799 		page = pfn_to_page(check);
7800 
7801 		if (PageReserved(page))
7802 			goto unmovable;
7803 
7804 		/*
7805 		 * If the zone is movable and we have ruled out all reserved
7806 		 * pages then it should be reasonably safe to assume the rest
7807 		 * is movable.
7808 		 */
7809 		if (zone_idx(zone) == ZONE_MOVABLE)
7810 			continue;
7811 
7812 		/*
7813 		 * Hugepages are not in LRU lists, but they're movable.
7814 		 * We need not scan over tail pages bacause we don't
7815 		 * handle each tail page individually in migration.
7816 		 */
7817 		if (PageHuge(page)) {
7818 			struct page *head = compound_head(page);
7819 			unsigned int skip_pages;
7820 
7821 			if (!hugepage_migration_supported(page_hstate(head)))
7822 				goto unmovable;
7823 
7824 			skip_pages = (1 << compound_order(head)) - (page - head);
7825 			iter += skip_pages - 1;
7826 			continue;
7827 		}
7828 
7829 		/*
7830 		 * We can't use page_count without pin a page
7831 		 * because another CPU can free compound page.
7832 		 * This check already skips compound tails of THP
7833 		 * because their page->_refcount is zero at all time.
7834 		 */
7835 		if (!page_ref_count(page)) {
7836 			if (PageBuddy(page))
7837 				iter += (1 << page_order(page)) - 1;
7838 			continue;
7839 		}
7840 
7841 		/*
7842 		 * The HWPoisoned page may be not in buddy system, and
7843 		 * page_count() is not 0.
7844 		 */
7845 		if (skip_hwpoisoned_pages && PageHWPoison(page))
7846 			continue;
7847 
7848 		if (__PageMovable(page))
7849 			continue;
7850 
7851 		if (!PageLRU(page))
7852 			found++;
7853 		/*
7854 		 * If there are RECLAIMABLE pages, we need to check
7855 		 * it.  But now, memory offline itself doesn't call
7856 		 * shrink_node_slabs() and it still to be fixed.
7857 		 */
7858 		/*
7859 		 * If the page is not RAM, page_count()should be 0.
7860 		 * we don't need more check. This is an _used_ not-movable page.
7861 		 *
7862 		 * The problematic thing here is PG_reserved pages. PG_reserved
7863 		 * is set to both of a memory hole page and a _used_ kernel
7864 		 * page at boot.
7865 		 */
7866 		if (found > count)
7867 			goto unmovable;
7868 	}
7869 	return false;
7870 unmovable:
7871 	WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
7872 	return true;
7873 }
7874 
7875 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7876 
7877 static unsigned long pfn_max_align_down(unsigned long pfn)
7878 {
7879 	return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7880 			     pageblock_nr_pages) - 1);
7881 }
7882 
7883 static unsigned long pfn_max_align_up(unsigned long pfn)
7884 {
7885 	return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7886 				pageblock_nr_pages));
7887 }
7888 
7889 /* [start, end) must belong to a single zone. */
7890 static int __alloc_contig_migrate_range(struct compact_control *cc,
7891 					unsigned long start, unsigned long end)
7892 {
7893 	/* This function is based on compact_zone() from compaction.c. */
7894 	unsigned long nr_reclaimed;
7895 	unsigned long pfn = start;
7896 	unsigned int tries = 0;
7897 	int ret = 0;
7898 
7899 	migrate_prep();
7900 
7901 	while (pfn < end || !list_empty(&cc->migratepages)) {
7902 		if (fatal_signal_pending(current)) {
7903 			ret = -EINTR;
7904 			break;
7905 		}
7906 
7907 		if (list_empty(&cc->migratepages)) {
7908 			cc->nr_migratepages = 0;
7909 			pfn = isolate_migratepages_range(cc, pfn, end);
7910 			if (!pfn) {
7911 				ret = -EINTR;
7912 				break;
7913 			}
7914 			tries = 0;
7915 		} else if (++tries == 5) {
7916 			ret = ret < 0 ? ret : -EBUSY;
7917 			break;
7918 		}
7919 
7920 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7921 							&cc->migratepages);
7922 		cc->nr_migratepages -= nr_reclaimed;
7923 
7924 		ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
7925 				    NULL, 0, cc->mode, MR_CONTIG_RANGE);
7926 	}
7927 	if (ret < 0) {
7928 		putback_movable_pages(&cc->migratepages);
7929 		return ret;
7930 	}
7931 	return 0;
7932 }
7933 
7934 /**
7935  * alloc_contig_range() -- tries to allocate given range of pages
7936  * @start:	start PFN to allocate
7937  * @end:	one-past-the-last PFN to allocate
7938  * @migratetype:	migratetype of the underlaying pageblocks (either
7939  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
7940  *			in range must have the same migratetype and it must
7941  *			be either of the two.
7942  * @gfp_mask:	GFP mask to use during compaction
7943  *
7944  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7945  * aligned.  The PFN range must belong to a single zone.
7946  *
7947  * The first thing this routine does is attempt to MIGRATE_ISOLATE all
7948  * pageblocks in the range.  Once isolated, the pageblocks should not
7949  * be modified by others.
7950  *
7951  * Returns zero on success or negative error code.  On success all
7952  * pages which PFN is in [start, end) are allocated for the caller and
7953  * need to be freed with free_contig_range().
7954  */
7955 int alloc_contig_range(unsigned long start, unsigned long end,
7956 		       unsigned migratetype, gfp_t gfp_mask)
7957 {
7958 	unsigned long outer_start, outer_end;
7959 	unsigned int order;
7960 	int ret = 0;
7961 
7962 	struct compact_control cc = {
7963 		.nr_migratepages = 0,
7964 		.order = -1,
7965 		.zone = page_zone(pfn_to_page(start)),
7966 		.mode = MIGRATE_SYNC,
7967 		.ignore_skip_hint = true,
7968 		.no_set_skip_hint = true,
7969 		.gfp_mask = current_gfp_context(gfp_mask),
7970 	};
7971 	INIT_LIST_HEAD(&cc.migratepages);
7972 
7973 	/*
7974 	 * What we do here is we mark all pageblocks in range as
7975 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
7976 	 * have different sizes, and due to the way page allocator
7977 	 * work, we align the range to biggest of the two pages so
7978 	 * that page allocator won't try to merge buddies from
7979 	 * different pageblocks and change MIGRATE_ISOLATE to some
7980 	 * other migration type.
7981 	 *
7982 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7983 	 * migrate the pages from an unaligned range (ie. pages that
7984 	 * we are interested in).  This will put all the pages in
7985 	 * range back to page allocator as MIGRATE_ISOLATE.
7986 	 *
7987 	 * When this is done, we take the pages in range from page
7988 	 * allocator removing them from the buddy system.  This way
7989 	 * page allocator will never consider using them.
7990 	 *
7991 	 * This lets us mark the pageblocks back as
7992 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7993 	 * aligned range but not in the unaligned, original range are
7994 	 * put back to page allocator so that buddy can use them.
7995 	 */
7996 
7997 	ret = start_isolate_page_range(pfn_max_align_down(start),
7998 				       pfn_max_align_up(end), migratetype,
7999 				       false);
8000 	if (ret)
8001 		return ret;
8002 
8003 	/*
8004 	 * In case of -EBUSY, we'd like to know which page causes problem.
8005 	 * So, just fall through. test_pages_isolated() has a tracepoint
8006 	 * which will report the busy page.
8007 	 *
8008 	 * It is possible that busy pages could become available before
8009 	 * the call to test_pages_isolated, and the range will actually be
8010 	 * allocated.  So, if we fall through be sure to clear ret so that
8011 	 * -EBUSY is not accidentally used or returned to caller.
8012 	 */
8013 	ret = __alloc_contig_migrate_range(&cc, start, end);
8014 	if (ret && ret != -EBUSY)
8015 		goto done;
8016 	ret =0;
8017 
8018 	/*
8019 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
8020 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
8021 	 * more, all pages in [start, end) are free in page allocator.
8022 	 * What we are going to do is to allocate all pages from
8023 	 * [start, end) (that is remove them from page allocator).
8024 	 *
8025 	 * The only problem is that pages at the beginning and at the
8026 	 * end of interesting range may be not aligned with pages that
8027 	 * page allocator holds, ie. they can be part of higher order
8028 	 * pages.  Because of this, we reserve the bigger range and
8029 	 * once this is done free the pages we are not interested in.
8030 	 *
8031 	 * We don't have to hold zone->lock here because the pages are
8032 	 * isolated thus they won't get removed from buddy.
8033 	 */
8034 
8035 	lru_add_drain_all();
8036 	drain_all_pages(cc.zone);
8037 
8038 	order = 0;
8039 	outer_start = start;
8040 	while (!PageBuddy(pfn_to_page(outer_start))) {
8041 		if (++order >= MAX_ORDER) {
8042 			outer_start = start;
8043 			break;
8044 		}
8045 		outer_start &= ~0UL << order;
8046 	}
8047 
8048 	if (outer_start != start) {
8049 		order = page_order(pfn_to_page(outer_start));
8050 
8051 		/*
8052 		 * outer_start page could be small order buddy page and
8053 		 * it doesn't include start page. Adjust outer_start
8054 		 * in this case to report failed page properly
8055 		 * on tracepoint in test_pages_isolated()
8056 		 */
8057 		if (outer_start + (1UL << order) <= start)
8058 			outer_start = start;
8059 	}
8060 
8061 	/* Make sure the range is really isolated. */
8062 	if (test_pages_isolated(outer_start, end, false)) {
8063 		pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8064 			__func__, outer_start, end);
8065 		ret = -EBUSY;
8066 		goto done;
8067 	}
8068 
8069 	/* Grab isolated pages from freelists. */
8070 	outer_end = isolate_freepages_range(&cc, outer_start, end);
8071 	if (!outer_end) {
8072 		ret = -EBUSY;
8073 		goto done;
8074 	}
8075 
8076 	/* Free head and tail (if any) */
8077 	if (start != outer_start)
8078 		free_contig_range(outer_start, start - outer_start);
8079 	if (end != outer_end)
8080 		free_contig_range(end, outer_end - end);
8081 
8082 done:
8083 	undo_isolate_page_range(pfn_max_align_down(start),
8084 				pfn_max_align_up(end), migratetype);
8085 	return ret;
8086 }
8087 
8088 void free_contig_range(unsigned long pfn, unsigned nr_pages)
8089 {
8090 	unsigned int count = 0;
8091 
8092 	for (; nr_pages--; pfn++) {
8093 		struct page *page = pfn_to_page(pfn);
8094 
8095 		count += page_count(page) != 1;
8096 		__free_page(page);
8097 	}
8098 	WARN(count != 0, "%d pages are still in use!\n", count);
8099 }
8100 #endif
8101 
8102 #ifdef CONFIG_MEMORY_HOTPLUG
8103 /*
8104  * The zone indicated has a new number of managed_pages; batch sizes and percpu
8105  * page high values need to be recalulated.
8106  */
8107 void __meminit zone_pcp_update(struct zone *zone)
8108 {
8109 	unsigned cpu;
8110 	mutex_lock(&pcp_batch_high_lock);
8111 	for_each_possible_cpu(cpu)
8112 		pageset_set_high_and_batch(zone,
8113 				per_cpu_ptr(zone->pageset, cpu));
8114 	mutex_unlock(&pcp_batch_high_lock);
8115 }
8116 #endif
8117 
8118 void zone_pcp_reset(struct zone *zone)
8119 {
8120 	unsigned long flags;
8121 	int cpu;
8122 	struct per_cpu_pageset *pset;
8123 
8124 	/* avoid races with drain_pages()  */
8125 	local_irq_save(flags);
8126 	if (zone->pageset != &boot_pageset) {
8127 		for_each_online_cpu(cpu) {
8128 			pset = per_cpu_ptr(zone->pageset, cpu);
8129 			drain_zonestat(zone, pset);
8130 		}
8131 		free_percpu(zone->pageset);
8132 		zone->pageset = &boot_pageset;
8133 	}
8134 	local_irq_restore(flags);
8135 }
8136 
8137 #ifdef CONFIG_MEMORY_HOTREMOVE
8138 /*
8139  * All pages in the range must be in a single zone and isolated
8140  * before calling this.
8141  */
8142 void
8143 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8144 {
8145 	struct page *page;
8146 	struct zone *zone;
8147 	unsigned int order, i;
8148 	unsigned long pfn;
8149 	unsigned long flags;
8150 	/* find the first valid pfn */
8151 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
8152 		if (pfn_valid(pfn))
8153 			break;
8154 	if (pfn == end_pfn)
8155 		return;
8156 	offline_mem_sections(pfn, end_pfn);
8157 	zone = page_zone(pfn_to_page(pfn));
8158 	spin_lock_irqsave(&zone->lock, flags);
8159 	pfn = start_pfn;
8160 	while (pfn < end_pfn) {
8161 		if (!pfn_valid(pfn)) {
8162 			pfn++;
8163 			continue;
8164 		}
8165 		page = pfn_to_page(pfn);
8166 		/*
8167 		 * The HWPoisoned page may be not in buddy system, and
8168 		 * page_count() is not 0.
8169 		 */
8170 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8171 			pfn++;
8172 			SetPageReserved(page);
8173 			continue;
8174 		}
8175 
8176 		BUG_ON(page_count(page));
8177 		BUG_ON(!PageBuddy(page));
8178 		order = page_order(page);
8179 #ifdef CONFIG_DEBUG_VM
8180 		pr_info("remove from free list %lx %d %lx\n",
8181 			pfn, 1 << order, end_pfn);
8182 #endif
8183 		list_del(&page->lru);
8184 		rmv_page_order(page);
8185 		zone->free_area[order].nr_free--;
8186 		for (i = 0; i < (1 << order); i++)
8187 			SetPageReserved((page+i));
8188 		pfn += (1 << order);
8189 	}
8190 	spin_unlock_irqrestore(&zone->lock, flags);
8191 }
8192 #endif
8193 
8194 bool is_free_buddy_page(struct page *page)
8195 {
8196 	struct zone *zone = page_zone(page);
8197 	unsigned long pfn = page_to_pfn(page);
8198 	unsigned long flags;
8199 	unsigned int order;
8200 
8201 	spin_lock_irqsave(&zone->lock, flags);
8202 	for (order = 0; order < MAX_ORDER; order++) {
8203 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8204 
8205 		if (PageBuddy(page_head) && page_order(page_head) >= order)
8206 			break;
8207 	}
8208 	spin_unlock_irqrestore(&zone->lock, flags);
8209 
8210 	return order < MAX_ORDER;
8211 }
8212 
8213 #ifdef CONFIG_MEMORY_FAILURE
8214 /*
8215  * Set PG_hwpoison flag if a given page is confirmed to be a free page.  This
8216  * test is performed under the zone lock to prevent a race against page
8217  * allocation.
8218  */
8219 bool set_hwpoison_free_buddy_page(struct page *page)
8220 {
8221 	struct zone *zone = page_zone(page);
8222 	unsigned long pfn = page_to_pfn(page);
8223 	unsigned long flags;
8224 	unsigned int order;
8225 	bool hwpoisoned = false;
8226 
8227 	spin_lock_irqsave(&zone->lock, flags);
8228 	for (order = 0; order < MAX_ORDER; order++) {
8229 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8230 
8231 		if (PageBuddy(page_head) && page_order(page_head) >= order) {
8232 			if (!TestSetPageHWPoison(page))
8233 				hwpoisoned = true;
8234 			break;
8235 		}
8236 	}
8237 	spin_unlock_irqrestore(&zone->lock, flags);
8238 
8239 	return hwpoisoned;
8240 }
8241 #endif
8242