]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - mm/page_alloc.c
userfaultfd: hugetlbfs: prevent UFFDIO_COPY to fill beyond the end of i_size
[mirror_ubuntu-artful-kernel.git] / mm / page_alloc.c
CommitLineData
1da177e4
LT
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
1da177e4
LT
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>
10ed273f 22#include <linux/jiffies.h>
1da177e4 23#include <linux/bootmem.h>
edbe7d23 24#include <linux/memblock.h>
1da177e4 25#include <linux/compiler.h>
9f158333 26#include <linux/kernel.h>
b1eeab67 27#include <linux/kmemcheck.h>
b8c73fc2 28#include <linux/kasan.h>
1da177e4
LT
29#include <linux/module.h>
30#include <linux/suspend.h>
31#include <linux/pagevec.h>
32#include <linux/blkdev.h>
33#include <linux/slab.h>
a238ab5b 34#include <linux/ratelimit.h>
5a3135c2 35#include <linux/oom.h>
1da177e4
LT
36#include <linux/notifier.h>
37#include <linux/topology.h>
38#include <linux/sysctl.h>
39#include <linux/cpu.h>
40#include <linux/cpuset.h>
bdc8cb98 41#include <linux/memory_hotplug.h>
1da177e4
LT
42#include <linux/nodemask.h>
43#include <linux/vmalloc.h>
a6cccdc3 44#include <linux/vmstat.h>
4be38e35 45#include <linux/mempolicy.h>
4b94ffdc 46#include <linux/memremap.h>
6811378e 47#include <linux/stop_machine.h>
c713216d
MG
48#include <linux/sort.h>
49#include <linux/pfn.h>
3fcfab16 50#include <linux/backing-dev.h>
933e312e 51#include <linux/fault-inject.h>
a5d76b54 52#include <linux/page-isolation.h>
eefa864b 53#include <linux/page_ext.h>
3ac7fe5a 54#include <linux/debugobjects.h>
dbb1f81c 55#include <linux/kmemleak.h>
56de7263 56#include <linux/compaction.h>
0d3d062a 57#include <trace/events/kmem.h>
d379f01d 58#include <trace/events/oom.h>
268bb0ce 59#include <linux/prefetch.h>
6e543d57 60#include <linux/mm_inline.h>
041d3a8c 61#include <linux/migrate.h>
949f7ec5 62#include <linux/hugetlb.h>
8bd75c77 63#include <linux/sched/rt.h>
5b3cc15a 64#include <linux/sched/mm.h>
48c96a36 65#include <linux/page_owner.h>
0e1cc95b 66#include <linux/kthread.h>
4949148a 67#include <linux/memcontrol.h>
42c269c8 68#include <linux/ftrace.h>
556b969a 69#include <linux/nmi.h>
1da177e4 70
7ee3d4e8 71#include <asm/sections.h>
1da177e4 72#include <asm/tlbflush.h>
ac924c60 73#include <asm/div64.h>
1da177e4
LT
74#include "internal.h"
75
c8e251fa
CS
76/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
77static DEFINE_MUTEX(pcp_batch_high_lock);
7cd2b0a3 78#define MIN_PERCPU_PAGELIST_FRACTION (8)
c8e251fa 79
72812019
LS
80#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
81DEFINE_PER_CPU(int, numa_node);
82EXPORT_PER_CPU_SYMBOL(numa_node);
83#endif
84
7aac7898
LS
85#ifdef CONFIG_HAVE_MEMORYLESS_NODES
86/*
87 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
88 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
89 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
90 * defined in <linux/topology.h>.
91 */
92DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
93EXPORT_PER_CPU_SYMBOL(_numa_mem_);
ad2c8144 94int _node_numa_mem_[MAX_NUMNODES];
7aac7898
LS
95#endif
96
bd233f53
MG
97/* work_structs for global per-cpu drains */
98DEFINE_MUTEX(pcpu_drain_mutex);
99DEFINE_PER_CPU(struct work_struct, pcpu_drain);
100
38addce8 101#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
58bea414 102volatile unsigned long latent_entropy __latent_entropy;
38addce8
ER
103EXPORT_SYMBOL(latent_entropy);
104#endif
105
1da177e4 106/*
13808910 107 * Array of node states.
1da177e4 108 */
13808910
CL
109nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
110 [N_POSSIBLE] = NODE_MASK_ALL,
111 [N_ONLINE] = { { [0] = 1UL } },
112#ifndef CONFIG_NUMA
113 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
114#ifdef CONFIG_HIGHMEM
115 [N_HIGH_MEMORY] = { { [0] = 1UL } },
20b2f52b 116#endif
20b2f52b 117 [N_MEMORY] = { { [0] = 1UL } },
13808910
CL
118 [N_CPU] = { { [0] = 1UL } },
119#endif /* NUMA */
120};
121EXPORT_SYMBOL(node_states);
122
c3d5f5f0
JL
123/* Protect totalram_pages and zone->managed_pages */
124static DEFINE_SPINLOCK(managed_page_count_lock);
125
6c231b7b 126unsigned long totalram_pages __read_mostly;
cb45b0e9 127unsigned long totalreserve_pages __read_mostly;
e48322ab 128unsigned long totalcma_pages __read_mostly;
ab8fabd4 129
1b76b02f 130int percpu_pagelist_fraction;
dcce284a 131gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
1da177e4 132
bb14c2c7
VB
133/*
134 * A cached value of the page's pageblock's migratetype, used when the page is
135 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
136 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
137 * Also the migratetype set in the page does not necessarily match the pcplist
138 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
139 * other index - this ensures that it will be put on the correct CMA freelist.
140 */
141static inline int get_pcppage_migratetype(struct page *page)
142{
143 return page->index;
144}
145
146static inline void set_pcppage_migratetype(struct page *page, int migratetype)
147{
148 page->index = migratetype;
149}
150
452aa699
RW
151#ifdef CONFIG_PM_SLEEP
152/*
153 * The following functions are used by the suspend/hibernate code to temporarily
154 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
155 * while devices are suspended. To avoid races with the suspend/hibernate code,
156 * they should always be called with pm_mutex held (gfp_allowed_mask also should
157 * only be modified with pm_mutex held, unless the suspend/hibernate code is
158 * guaranteed not to run in parallel with that modification).
159 */
c9e664f1
RW
160
161static gfp_t saved_gfp_mask;
162
163void pm_restore_gfp_mask(void)
452aa699
RW
164{
165 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
166 if (saved_gfp_mask) {
167 gfp_allowed_mask = saved_gfp_mask;
168 saved_gfp_mask = 0;
169 }
452aa699
RW
170}
171
c9e664f1 172void pm_restrict_gfp_mask(void)
452aa699 173{
452aa699 174 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
175 WARN_ON(saved_gfp_mask);
176 saved_gfp_mask = gfp_allowed_mask;
d0164adc 177 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
452aa699 178}
f90ac398
MG
179
180bool pm_suspended_storage(void)
181{
d0164adc 182 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
f90ac398
MG
183 return false;
184 return true;
185}
452aa699
RW
186#endif /* CONFIG_PM_SLEEP */
187
d9c23400 188#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
d00181b9 189unsigned int pageblock_order __read_mostly;
d9c23400
MG
190#endif
191
d98c7a09 192static void __free_pages_ok(struct page *page, unsigned int order);
a226f6c8 193
1da177e4
LT
194/*
195 * results with 256, 32 in the lowmem_reserve sysctl:
196 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
197 * 1G machine -> (16M dma, 784M normal, 224M high)
198 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
199 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
84109e15 200 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
a2f1b424
AK
201 *
202 * TBD: should special case ZONE_DMA32 machines here - in those we normally
203 * don't need any ZONE_NORMAL reservation
1da177e4 204 */
2f1b6248 205int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
4b51d669 206#ifdef CONFIG_ZONE_DMA
2f1b6248 207 256,
4b51d669 208#endif
fb0e7942 209#ifdef CONFIG_ZONE_DMA32
2f1b6248 210 256,
fb0e7942 211#endif
e53ef38d 212#ifdef CONFIG_HIGHMEM
2a1e274a 213 32,
e53ef38d 214#endif
2a1e274a 215 32,
2f1b6248 216};
1da177e4
LT
217
218EXPORT_SYMBOL(totalram_pages);
1da177e4 219
15ad7cdc 220static char * const zone_names[MAX_NR_ZONES] = {
4b51d669 221#ifdef CONFIG_ZONE_DMA
2f1b6248 222 "DMA",
4b51d669 223#endif
fb0e7942 224#ifdef CONFIG_ZONE_DMA32
2f1b6248 225 "DMA32",
fb0e7942 226#endif
2f1b6248 227 "Normal",
e53ef38d 228#ifdef CONFIG_HIGHMEM
2a1e274a 229 "HighMem",
e53ef38d 230#endif
2a1e274a 231 "Movable",
033fbae9
DW
232#ifdef CONFIG_ZONE_DEVICE
233 "Device",
234#endif
2f1b6248
CL
235};
236
60f30350
VB
237char * const migratetype_names[MIGRATE_TYPES] = {
238 "Unmovable",
239 "Movable",
240 "Reclaimable",
241 "HighAtomic",
242#ifdef CONFIG_CMA
243 "CMA",
244#endif
245#ifdef CONFIG_MEMORY_ISOLATION
246 "Isolate",
247#endif
248};
249
f1e61557
KS
250compound_page_dtor * const compound_page_dtors[] = {
251 NULL,
252 free_compound_page,
253#ifdef CONFIG_HUGETLB_PAGE
254 free_huge_page,
255#endif
9a982250
KS
256#ifdef CONFIG_TRANSPARENT_HUGEPAGE
257 free_transhuge_page,
258#endif
f1e61557
KS
259};
260
1da177e4 261int min_free_kbytes = 1024;
42aa83cb 262int user_min_free_kbytes = -1;
795ae7a0 263int watermark_scale_factor = 10;
1da177e4 264
2c85f51d
JB
265static unsigned long __meminitdata nr_kernel_pages;
266static unsigned long __meminitdata nr_all_pages;
a3142c8e 267static unsigned long __meminitdata dma_reserve;
1da177e4 268
0ee332c1
TH
269#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
270static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
271static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
272static unsigned long __initdata required_kernelcore;
273static unsigned long __initdata required_movablecore;
274static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
342332e6 275static bool mirrored_kernelcore;
0ee332c1
TH
276
277/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
278int movable_zone;
279EXPORT_SYMBOL(movable_zone);
280#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 281
418508c1
MS
282#if MAX_NUMNODES > 1
283int nr_node_ids __read_mostly = MAX_NUMNODES;
62bc62a8 284int nr_online_nodes __read_mostly = 1;
418508c1 285EXPORT_SYMBOL(nr_node_ids);
62bc62a8 286EXPORT_SYMBOL(nr_online_nodes);
418508c1
MS
287#endif
288
9ef9acb0
MG
289int page_group_by_mobility_disabled __read_mostly;
290
3a80a7fa
MG
291#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
292static inline void reset_deferred_meminit(pg_data_t *pgdat)
293{
864b9a39
MH
294 unsigned long max_initialise;
295 unsigned long reserved_lowmem;
296
297 /*
298 * Initialise at least 2G of a node but also take into account that
299 * two large system hashes that can take up 1GB for 0.25TB/node.
300 */
301 max_initialise = max(2UL << (30 - PAGE_SHIFT),
302 (pgdat->node_spanned_pages >> 8));
303
304 /*
305 * Compensate the all the memblock reservations (e.g. crash kernel)
306 * from the initial estimation to make sure we will initialize enough
307 * memory to boot.
308 */
309 reserved_lowmem = memblock_reserved_memory_within(pgdat->node_start_pfn,
310 pgdat->node_start_pfn + max_initialise);
311 max_initialise += reserved_lowmem;
312
313 pgdat->static_init_size = min(max_initialise, pgdat->node_spanned_pages);
3a80a7fa
MG
314 pgdat->first_deferred_pfn = ULONG_MAX;
315}
316
317/* Returns true if the struct page for the pfn is uninitialised */
0e1cc95b 318static inline bool __meminit early_page_uninitialised(unsigned long pfn)
3a80a7fa 319{
ef70b6f4
MG
320 int nid = early_pfn_to_nid(pfn);
321
322 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
3a80a7fa
MG
323 return true;
324
325 return false;
326}
327
328/*
329 * Returns false when the remaining initialisation should be deferred until
330 * later in the boot cycle when it can be parallelised.
331 */
332static inline bool update_defer_init(pg_data_t *pgdat,
333 unsigned long pfn, unsigned long zone_end,
334 unsigned long *nr_initialised)
335{
336 /* Always populate low zones for address-contrained allocations */
337 if (zone_end < pgdat_end_pfn(pgdat))
338 return true;
3a80a7fa 339 (*nr_initialised)++;
864b9a39 340 if ((*nr_initialised > pgdat->static_init_size) &&
3a80a7fa
MG
341 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
342 pgdat->first_deferred_pfn = pfn;
343 return false;
344 }
345
346 return true;
347}
348#else
349static inline void reset_deferred_meminit(pg_data_t *pgdat)
350{
351}
352
353static inline bool early_page_uninitialised(unsigned long pfn)
354{
355 return false;
356}
357
358static inline bool update_defer_init(pg_data_t *pgdat,
359 unsigned long pfn, unsigned long zone_end,
360 unsigned long *nr_initialised)
361{
362 return true;
363}
364#endif
365
0b423ca2
MG
366/* Return a pointer to the bitmap storing bits affecting a block of pages */
367static inline unsigned long *get_pageblock_bitmap(struct page *page,
368 unsigned long pfn)
369{
370#ifdef CONFIG_SPARSEMEM
371 return __pfn_to_section(pfn)->pageblock_flags;
372#else
373 return page_zone(page)->pageblock_flags;
374#endif /* CONFIG_SPARSEMEM */
375}
376
377static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
378{
379#ifdef CONFIG_SPARSEMEM
380 pfn &= (PAGES_PER_SECTION-1);
381 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
382#else
383 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
384 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
385#endif /* CONFIG_SPARSEMEM */
386}
387
388/**
389 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
390 * @page: The page within the block of interest
391 * @pfn: The target page frame number
392 * @end_bitidx: The last bit of interest to retrieve
393 * @mask: mask of bits that the caller is interested in
394 *
395 * Return: pageblock_bits flags
396 */
397static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
398 unsigned long pfn,
399 unsigned long end_bitidx,
400 unsigned long mask)
401{
402 unsigned long *bitmap;
403 unsigned long bitidx, word_bitidx;
404 unsigned long word;
405
406 bitmap = get_pageblock_bitmap(page, pfn);
407 bitidx = pfn_to_bitidx(page, pfn);
408 word_bitidx = bitidx / BITS_PER_LONG;
409 bitidx &= (BITS_PER_LONG-1);
410
411 word = bitmap[word_bitidx];
412 bitidx += end_bitidx;
413 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
414}
415
416unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
417 unsigned long end_bitidx,
418 unsigned long mask)
419{
420 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
421}
422
423static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
424{
425 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
426}
427
428/**
429 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
430 * @page: The page within the block of interest
431 * @flags: The flags to set
432 * @pfn: The target page frame number
433 * @end_bitidx: The last bit of interest
434 * @mask: mask of bits that the caller is interested in
435 */
436void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
437 unsigned long pfn,
438 unsigned long end_bitidx,
439 unsigned long mask)
440{
441 unsigned long *bitmap;
442 unsigned long bitidx, word_bitidx;
443 unsigned long old_word, word;
444
445 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
446
447 bitmap = get_pageblock_bitmap(page, pfn);
448 bitidx = pfn_to_bitidx(page, pfn);
449 word_bitidx = bitidx / BITS_PER_LONG;
450 bitidx &= (BITS_PER_LONG-1);
451
452 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
453
454 bitidx += end_bitidx;
455 mask <<= (BITS_PER_LONG - bitidx - 1);
456 flags <<= (BITS_PER_LONG - bitidx - 1);
457
458 word = READ_ONCE(bitmap[word_bitidx]);
459 for (;;) {
460 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
461 if (word == old_word)
462 break;
463 word = old_word;
464 }
465}
3a80a7fa 466
ee6f509c 467void set_pageblock_migratetype(struct page *page, int migratetype)
b2a0ac88 468{
5d0f3f72
KM
469 if (unlikely(page_group_by_mobility_disabled &&
470 migratetype < MIGRATE_PCPTYPES))
49255c61
MG
471 migratetype = MIGRATE_UNMOVABLE;
472
b2a0ac88
MG
473 set_pageblock_flags_group(page, (unsigned long)migratetype,
474 PB_migrate, PB_migrate_end);
475}
476
13e7444b 477#ifdef CONFIG_DEBUG_VM
c6a57e19 478static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
1da177e4 479{
bdc8cb98
DH
480 int ret = 0;
481 unsigned seq;
482 unsigned long pfn = page_to_pfn(page);
b5e6a5a2 483 unsigned long sp, start_pfn;
c6a57e19 484
bdc8cb98
DH
485 do {
486 seq = zone_span_seqbegin(zone);
b5e6a5a2
CS
487 start_pfn = zone->zone_start_pfn;
488 sp = zone->spanned_pages;
108bcc96 489 if (!zone_spans_pfn(zone, pfn))
bdc8cb98
DH
490 ret = 1;
491 } while (zone_span_seqretry(zone, seq));
492
b5e6a5a2 493 if (ret)
613813e8
DH
494 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
495 pfn, zone_to_nid(zone), zone->name,
496 start_pfn, start_pfn + sp);
b5e6a5a2 497
bdc8cb98 498 return ret;
c6a57e19
DH
499}
500
501static int page_is_consistent(struct zone *zone, struct page *page)
502{
14e07298 503 if (!pfn_valid_within(page_to_pfn(page)))
c6a57e19 504 return 0;
1da177e4 505 if (zone != page_zone(page))
c6a57e19
DH
506 return 0;
507
508 return 1;
509}
510/*
511 * Temporary debugging check for pages not lying within a given zone.
512 */
d73d3c9f 513static int __maybe_unused bad_range(struct zone *zone, struct page *page)
c6a57e19
DH
514{
515 if (page_outside_zone_boundaries(zone, page))
1da177e4 516 return 1;
c6a57e19
DH
517 if (!page_is_consistent(zone, page))
518 return 1;
519
1da177e4
LT
520 return 0;
521}
13e7444b 522#else
d73d3c9f 523static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
13e7444b
NP
524{
525 return 0;
526}
527#endif
528
d230dec1
KS
529static void bad_page(struct page *page, const char *reason,
530 unsigned long bad_flags)
1da177e4 531{
d936cf9b
HD
532 static unsigned long resume;
533 static unsigned long nr_shown;
534 static unsigned long nr_unshown;
535
536 /*
537 * Allow a burst of 60 reports, then keep quiet for that minute;
538 * or allow a steady drip of one report per second.
539 */
540 if (nr_shown == 60) {
541 if (time_before(jiffies, resume)) {
542 nr_unshown++;
543 goto out;
544 }
545 if (nr_unshown) {
ff8e8116 546 pr_alert(
1e9e6365 547 "BUG: Bad page state: %lu messages suppressed\n",
d936cf9b
HD
548 nr_unshown);
549 nr_unshown = 0;
550 }
551 nr_shown = 0;
552 }
553 if (nr_shown++ == 0)
554 resume = jiffies + 60 * HZ;
555
ff8e8116 556 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
3dc14741 557 current->comm, page_to_pfn(page));
ff8e8116
VB
558 __dump_page(page, reason);
559 bad_flags &= page->flags;
560 if (bad_flags)
561 pr_alert("bad because of flags: %#lx(%pGp)\n",
562 bad_flags, &bad_flags);
4e462112 563 dump_page_owner(page);
3dc14741 564
4f31888c 565 print_modules();
1da177e4 566 dump_stack();
d936cf9b 567out:
8cc3b392 568 /* Leave bad fields for debug, except PageBuddy could make trouble */
22b751c3 569 page_mapcount_reset(page); /* remove PageBuddy */
373d4d09 570 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1da177e4
LT
571}
572
1da177e4
LT
573/*
574 * Higher-order pages are called "compound pages". They are structured thusly:
575 *
1d798ca3 576 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
1da177e4 577 *
1d798ca3
KS
578 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
579 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
1da177e4 580 *
1d798ca3
KS
581 * The first tail page's ->compound_dtor holds the offset in array of compound
582 * page destructors. See compound_page_dtors.
1da177e4 583 *
1d798ca3 584 * The first tail page's ->compound_order holds the order of allocation.
41d78ba5 585 * This usage means that zero-order pages may not be compound.
1da177e4 586 */
d98c7a09 587
9a982250 588void free_compound_page(struct page *page)
d98c7a09 589{
d85f3385 590 __free_pages_ok(page, compound_order(page));
d98c7a09
HD
591}
592
d00181b9 593void prep_compound_page(struct page *page, unsigned int order)
18229df5
AW
594{
595 int i;
596 int nr_pages = 1 << order;
597
f1e61557 598 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
18229df5
AW
599 set_compound_order(page, order);
600 __SetPageHead(page);
601 for (i = 1; i < nr_pages; i++) {
602 struct page *p = page + i;
58a84aa9 603 set_page_count(p, 0);
1c290f64 604 p->mapping = TAIL_MAPPING;
1d798ca3 605 set_compound_head(p, page);
18229df5 606 }
53f9263b 607 atomic_set(compound_mapcount_ptr(page), -1);
18229df5
AW
608}
609
c0a32fc5
SG
610#ifdef CONFIG_DEBUG_PAGEALLOC
611unsigned int _debug_guardpage_minorder;
ea6eabb0
CB
612bool _debug_pagealloc_enabled __read_mostly
613 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
505f6d22 614EXPORT_SYMBOL(_debug_pagealloc_enabled);
e30825f1
JK
615bool _debug_guardpage_enabled __read_mostly;
616
031bc574
JK
617static int __init early_debug_pagealloc(char *buf)
618{
619 if (!buf)
620 return -EINVAL;
2a138dc7 621 return kstrtobool(buf, &_debug_pagealloc_enabled);
031bc574
JK
622}
623early_param("debug_pagealloc", early_debug_pagealloc);
624
e30825f1
JK
625static bool need_debug_guardpage(void)
626{
031bc574
JK
627 /* If we don't use debug_pagealloc, we don't need guard page */
628 if (!debug_pagealloc_enabled())
629 return false;
630
f1c1e9f7
JK
631 if (!debug_guardpage_minorder())
632 return false;
633
e30825f1
JK
634 return true;
635}
636
637static void init_debug_guardpage(void)
638{
031bc574
JK
639 if (!debug_pagealloc_enabled())
640 return;
641
f1c1e9f7
JK
642 if (!debug_guardpage_minorder())
643 return;
644
e30825f1
JK
645 _debug_guardpage_enabled = true;
646}
647
648struct page_ext_operations debug_guardpage_ops = {
649 .need = need_debug_guardpage,
650 .init = init_debug_guardpage,
651};
c0a32fc5
SG
652
653static int __init debug_guardpage_minorder_setup(char *buf)
654{
655 unsigned long res;
656
657 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
1170532b 658 pr_err("Bad debug_guardpage_minorder value\n");
c0a32fc5
SG
659 return 0;
660 }
661 _debug_guardpage_minorder = res;
1170532b 662 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
c0a32fc5
SG
663 return 0;
664}
f1c1e9f7 665early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
c0a32fc5 666
acbc15a4 667static inline bool set_page_guard(struct zone *zone, struct page *page,
2847cf95 668 unsigned int order, int migratetype)
c0a32fc5 669{
e30825f1
JK
670 struct page_ext *page_ext;
671
672 if (!debug_guardpage_enabled())
acbc15a4
JK
673 return false;
674
675 if (order >= debug_guardpage_minorder())
676 return false;
e30825f1
JK
677
678 page_ext = lookup_page_ext(page);
f86e4271 679 if (unlikely(!page_ext))
acbc15a4 680 return false;
f86e4271 681
e30825f1
JK
682 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
683
2847cf95
JK
684 INIT_LIST_HEAD(&page->lru);
685 set_page_private(page, order);
686 /* Guard pages are not available for any usage */
687 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
acbc15a4
JK
688
689 return true;
c0a32fc5
SG
690}
691
2847cf95
JK
692static inline void clear_page_guard(struct zone *zone, struct page *page,
693 unsigned int order, int migratetype)
c0a32fc5 694{
e30825f1
JK
695 struct page_ext *page_ext;
696
697 if (!debug_guardpage_enabled())
698 return;
699
700 page_ext = lookup_page_ext(page);
f86e4271
YS
701 if (unlikely(!page_ext))
702 return;
703
e30825f1
JK
704 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
705
2847cf95
JK
706 set_page_private(page, 0);
707 if (!is_migrate_isolate(migratetype))
708 __mod_zone_freepage_state(zone, (1 << order), migratetype);
c0a32fc5
SG
709}
710#else
980ac167 711struct page_ext_operations debug_guardpage_ops;
acbc15a4
JK
712static inline bool set_page_guard(struct zone *zone, struct page *page,
713 unsigned int order, int migratetype) { return false; }
2847cf95
JK
714static inline void clear_page_guard(struct zone *zone, struct page *page,
715 unsigned int order, int migratetype) {}
c0a32fc5
SG
716#endif
717
7aeb09f9 718static inline void set_page_order(struct page *page, unsigned int order)
6aa3001b 719{
4c21e2f2 720 set_page_private(page, order);
676165a8 721 __SetPageBuddy(page);
1da177e4
LT
722}
723
724static inline void rmv_page_order(struct page *page)
725{
676165a8 726 __ClearPageBuddy(page);
4c21e2f2 727 set_page_private(page, 0);
1da177e4
LT
728}
729
1da177e4
LT
730/*
731 * This function checks whether a page is free && is the buddy
732 * we can do coalesce a page and its buddy if
13ad59df 733 * (a) the buddy is not in a hole (check before calling!) &&
676165a8 734 * (b) the buddy is in the buddy system &&
cb2b95e1
AW
735 * (c) a page and its buddy have the same order &&
736 * (d) a page and its buddy are in the same zone.
676165a8 737 *
cf6fe945
WSH
738 * For recording whether a page is in the buddy system, we set ->_mapcount
739 * PAGE_BUDDY_MAPCOUNT_VALUE.
740 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
741 * serialized by zone->lock.
1da177e4 742 *
676165a8 743 * For recording page's order, we use page_private(page).
1da177e4 744 */
cb2b95e1 745static inline int page_is_buddy(struct page *page, struct page *buddy,
7aeb09f9 746 unsigned int order)
1da177e4 747{
c0a32fc5 748 if (page_is_guard(buddy) && page_order(buddy) == order) {
d34c5fa0
MG
749 if (page_zone_id(page) != page_zone_id(buddy))
750 return 0;
751
4c5018ce
WY
752 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
753
c0a32fc5
SG
754 return 1;
755 }
756
cb2b95e1 757 if (PageBuddy(buddy) && page_order(buddy) == order) {
d34c5fa0
MG
758 /*
759 * zone check is done late to avoid uselessly
760 * calculating zone/node ids for pages that could
761 * never merge.
762 */
763 if (page_zone_id(page) != page_zone_id(buddy))
764 return 0;
765
4c5018ce
WY
766 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
767
6aa3001b 768 return 1;
676165a8 769 }
6aa3001b 770 return 0;
1da177e4
LT
771}
772
773/*
774 * Freeing function for a buddy system allocator.
775 *
776 * The concept of a buddy system is to maintain direct-mapped table
777 * (containing bit values) for memory blocks of various "orders".
778 * The bottom level table contains the map for the smallest allocatable
779 * units of memory (here, pages), and each level above it describes
780 * pairs of units from the levels below, hence, "buddies".
781 * At a high level, all that happens here is marking the table entry
782 * at the bottom level available, and propagating the changes upward
783 * as necessary, plus some accounting needed to play nicely with other
784 * parts of the VM system.
785 * At each level, we keep a list of pages, which are heads of continuous
cf6fe945
WSH
786 * free pages of length of (1 << order) and marked with _mapcount
787 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
788 * field.
1da177e4 789 * So when we are allocating or freeing one, we can derive the state of the
5f63b720
MN
790 * other. That is, if we allocate a small block, and both were
791 * free, the remainder of the region must be split into blocks.
1da177e4 792 * If a block is freed, and its buddy is also free, then this
5f63b720 793 * triggers coalescing into a block of larger size.
1da177e4 794 *
6d49e352 795 * -- nyc
1da177e4
LT
796 */
797
48db57f8 798static inline void __free_one_page(struct page *page,
dc4b0caf 799 unsigned long pfn,
ed0ae21d
MG
800 struct zone *zone, unsigned int order,
801 int migratetype)
1da177e4 802{
76741e77
VB
803 unsigned long combined_pfn;
804 unsigned long uninitialized_var(buddy_pfn);
6dda9d55 805 struct page *buddy;
d9dddbf5
VB
806 unsigned int max_order;
807
808 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
1da177e4 809
d29bb978 810 VM_BUG_ON(!zone_is_initialized(zone));
6e9f0d58 811 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1da177e4 812
ed0ae21d 813 VM_BUG_ON(migratetype == -1);
d9dddbf5 814 if (likely(!is_migrate_isolate(migratetype)))
8f82b55d 815 __mod_zone_freepage_state(zone, 1 << order, migratetype);
ed0ae21d 816
76741e77 817 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
309381fe 818 VM_BUG_ON_PAGE(bad_range(zone, page), page);
1da177e4 819
d9dddbf5 820continue_merging:
3c605096 821 while (order < max_order - 1) {
76741e77
VB
822 buddy_pfn = __find_buddy_pfn(pfn, order);
823 buddy = page + (buddy_pfn - pfn);
13ad59df
VB
824
825 if (!pfn_valid_within(buddy_pfn))
826 goto done_merging;
cb2b95e1 827 if (!page_is_buddy(page, buddy, order))
d9dddbf5 828 goto done_merging;
c0a32fc5
SG
829 /*
830 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
831 * merge with it and move up one order.
832 */
833 if (page_is_guard(buddy)) {
2847cf95 834 clear_page_guard(zone, buddy, order, migratetype);
c0a32fc5
SG
835 } else {
836 list_del(&buddy->lru);
837 zone->free_area[order].nr_free--;
838 rmv_page_order(buddy);
839 }
76741e77
VB
840 combined_pfn = buddy_pfn & pfn;
841 page = page + (combined_pfn - pfn);
842 pfn = combined_pfn;
1da177e4
LT
843 order++;
844 }
d9dddbf5
VB
845 if (max_order < MAX_ORDER) {
846 /* If we are here, it means order is >= pageblock_order.
847 * We want to prevent merge between freepages on isolate
848 * pageblock and normal pageblock. Without this, pageblock
849 * isolation could cause incorrect freepage or CMA accounting.
850 *
851 * We don't want to hit this code for the more frequent
852 * low-order merging.
853 */
854 if (unlikely(has_isolate_pageblock(zone))) {
855 int buddy_mt;
856
76741e77
VB
857 buddy_pfn = __find_buddy_pfn(pfn, order);
858 buddy = page + (buddy_pfn - pfn);
d9dddbf5
VB
859 buddy_mt = get_pageblock_migratetype(buddy);
860
861 if (migratetype != buddy_mt
862 && (is_migrate_isolate(migratetype) ||
863 is_migrate_isolate(buddy_mt)))
864 goto done_merging;
865 }
866 max_order++;
867 goto continue_merging;
868 }
869
870done_merging:
1da177e4 871 set_page_order(page, order);
6dda9d55
CZ
872
873 /*
874 * If this is not the largest possible page, check if the buddy
875 * of the next-highest order is free. If it is, it's possible
876 * that pages are being freed that will coalesce soon. In case,
877 * that is happening, add the free page to the tail of the list
878 * so it's less likely to be used soon and more likely to be merged
879 * as a higher order page
880 */
13ad59df 881 if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
6dda9d55 882 struct page *higher_page, *higher_buddy;
76741e77
VB
883 combined_pfn = buddy_pfn & pfn;
884 higher_page = page + (combined_pfn - pfn);
885 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
886 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
b4fb8f66
TL
887 if (pfn_valid_within(buddy_pfn) &&
888 page_is_buddy(higher_page, higher_buddy, order + 1)) {
6dda9d55
CZ
889 list_add_tail(&page->lru,
890 &zone->free_area[order].free_list[migratetype]);
891 goto out;
892 }
893 }
894
895 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
896out:
1da177e4
LT
897 zone->free_area[order].nr_free++;
898}
899
7bfec6f4
MG
900/*
901 * A bad page could be due to a number of fields. Instead of multiple branches,
902 * try and check multiple fields with one check. The caller must do a detailed
903 * check if necessary.
904 */
905static inline bool page_expected_state(struct page *page,
906 unsigned long check_flags)
907{
908 if (unlikely(atomic_read(&page->_mapcount) != -1))
909 return false;
910
911 if (unlikely((unsigned long)page->mapping |
912 page_ref_count(page) |
913#ifdef CONFIG_MEMCG
914 (unsigned long)page->mem_cgroup |
915#endif
916 (page->flags & check_flags)))
917 return false;
918
919 return true;
920}
921
bb552ac6 922static void free_pages_check_bad(struct page *page)
1da177e4 923{
7bfec6f4
MG
924 const char *bad_reason;
925 unsigned long bad_flags;
926
7bfec6f4
MG
927 bad_reason = NULL;
928 bad_flags = 0;
f0b791a3 929
53f9263b 930 if (unlikely(atomic_read(&page->_mapcount) != -1))
f0b791a3
DH
931 bad_reason = "nonzero mapcount";
932 if (unlikely(page->mapping != NULL))
933 bad_reason = "non-NULL mapping";
fe896d18 934 if (unlikely(page_ref_count(page) != 0))
0139aa7b 935 bad_reason = "nonzero _refcount";
f0b791a3
DH
936 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
937 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
938 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
939 }
9edad6ea
JW
940#ifdef CONFIG_MEMCG
941 if (unlikely(page->mem_cgroup))
942 bad_reason = "page still charged to cgroup";
943#endif
7bfec6f4 944 bad_page(page, bad_reason, bad_flags);
bb552ac6
MG
945}
946
947static inline int free_pages_check(struct page *page)
948{
da838d4f 949 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
bb552ac6 950 return 0;
bb552ac6
MG
951
952 /* Something has gone sideways, find it */
953 free_pages_check_bad(page);
7bfec6f4 954 return 1;
1da177e4
LT
955}
956
4db7548c
MG
957static int free_tail_pages_check(struct page *head_page, struct page *page)
958{
959 int ret = 1;
960
961 /*
962 * We rely page->lru.next never has bit 0 set, unless the page
963 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
964 */
965 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
966
967 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
968 ret = 0;
969 goto out;
970 }
971 switch (page - head_page) {
972 case 1:
973 /* the first tail page: ->mapping is compound_mapcount() */
974 if (unlikely(compound_mapcount(page))) {
975 bad_page(page, "nonzero compound_mapcount", 0);
976 goto out;
977 }
978 break;
979 case 2:
980 /*
981 * the second tail page: ->mapping is
982 * page_deferred_list().next -- ignore value.
983 */
984 break;
985 default:
986 if (page->mapping != TAIL_MAPPING) {
987 bad_page(page, "corrupted mapping in tail page", 0);
988 goto out;
989 }
990 break;
991 }
992 if (unlikely(!PageTail(page))) {
993 bad_page(page, "PageTail not set", 0);
994 goto out;
995 }
996 if (unlikely(compound_head(page) != head_page)) {
997 bad_page(page, "compound_head not consistent", 0);
998 goto out;
999 }
1000 ret = 0;
1001out:
1002 page->mapping = NULL;
1003 clear_compound_head(page);
1004 return ret;
1005}
1006
e2769dbd
MG
1007static __always_inline bool free_pages_prepare(struct page *page,
1008 unsigned int order, bool check_free)
4db7548c 1009{
e2769dbd 1010 int bad = 0;
4db7548c 1011
4db7548c
MG
1012 VM_BUG_ON_PAGE(PageTail(page), page);
1013
e2769dbd
MG
1014 trace_mm_page_free(page, order);
1015 kmemcheck_free_shadow(page, order);
e2769dbd
MG
1016
1017 /*
1018 * Check tail pages before head page information is cleared to
1019 * avoid checking PageCompound for order-0 pages.
1020 */
1021 if (unlikely(order)) {
1022 bool compound = PageCompound(page);
1023 int i;
1024
1025 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
4db7548c 1026
9a73f61b
KS
1027 if (compound)
1028 ClearPageDoubleMap(page);
e2769dbd
MG
1029 for (i = 1; i < (1 << order); i++) {
1030 if (compound)
1031 bad += free_tail_pages_check(page, page + i);
1032 if (unlikely(free_pages_check(page + i))) {
1033 bad++;
1034 continue;
1035 }
1036 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1037 }
1038 }
bda807d4 1039 if (PageMappingFlags(page))
4db7548c 1040 page->mapping = NULL;
c4159a75 1041 if (memcg_kmem_enabled() && PageKmemcg(page))
4949148a 1042 memcg_kmem_uncharge(page, order);
e2769dbd
MG
1043 if (check_free)
1044 bad += free_pages_check(page);
1045 if (bad)
1046 return false;
4db7548c 1047
e2769dbd
MG
1048 page_cpupid_reset_last(page);
1049 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1050 reset_page_owner(page, order);
4db7548c
MG
1051
1052 if (!PageHighMem(page)) {
1053 debug_check_no_locks_freed(page_address(page),
e2769dbd 1054 PAGE_SIZE << order);
4db7548c 1055 debug_check_no_obj_freed(page_address(page),
e2769dbd 1056 PAGE_SIZE << order);
4db7548c 1057 }
e2769dbd
MG
1058 arch_free_page(page, order);
1059 kernel_poison_pages(page, 1 << order, 0);
1060 kernel_map_pages(page, 1 << order, 0);
29b52de1 1061 kasan_free_pages(page, order);
4db7548c 1062
4db7548c
MG
1063 return true;
1064}
1065
e2769dbd
MG
1066#ifdef CONFIG_DEBUG_VM
1067static inline bool free_pcp_prepare(struct page *page)
1068{
1069 return free_pages_prepare(page, 0, true);
1070}
1071
1072static inline bool bulkfree_pcp_prepare(struct page *page)
1073{
1074 return false;
1075}
1076#else
1077static bool free_pcp_prepare(struct page *page)
1078{
1079 return free_pages_prepare(page, 0, false);
1080}
1081
4db7548c
MG
1082static bool bulkfree_pcp_prepare(struct page *page)
1083{
1084 return free_pages_check(page);
1085}
1086#endif /* CONFIG_DEBUG_VM */
1087
1da177e4 1088/*
5f8dcc21 1089 * Frees a number of pages from the PCP lists
1da177e4 1090 * Assumes all pages on list are in same zone, and of same order.
207f36ee 1091 * count is the number of pages to free.
1da177e4
LT
1092 *
1093 * If the zone was previously in an "all pages pinned" state then look to
1094 * see if this freeing clears that state.
1095 *
1096 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1097 * pinned" detection logic.
1098 */
5f8dcc21
MG
1099static void free_pcppages_bulk(struct zone *zone, int count,
1100 struct per_cpu_pages *pcp)
1da177e4 1101{
5f8dcc21 1102 int migratetype = 0;
a6f9edd6 1103 int batch_free = 0;
3777999d 1104 bool isolated_pageblocks;
5f8dcc21 1105
d34b0733 1106 spin_lock(&zone->lock);
3777999d 1107 isolated_pageblocks = has_isolate_pageblock(zone);
f2260e6b 1108
e5b31ac2 1109 while (count) {
48db57f8 1110 struct page *page;
5f8dcc21
MG
1111 struct list_head *list;
1112
1113 /*
a6f9edd6
MG
1114 * Remove pages from lists in a round-robin fashion. A
1115 * batch_free count is maintained that is incremented when an
1116 * empty list is encountered. This is so more pages are freed
1117 * off fuller lists instead of spinning excessively around empty
1118 * lists
5f8dcc21
MG
1119 */
1120 do {
a6f9edd6 1121 batch_free++;
5f8dcc21
MG
1122 if (++migratetype == MIGRATE_PCPTYPES)
1123 migratetype = 0;
1124 list = &pcp->lists[migratetype];
1125 } while (list_empty(list));
48db57f8 1126
1d16871d
NK
1127 /* This is the only non-empty list. Free them all. */
1128 if (batch_free == MIGRATE_PCPTYPES)
e5b31ac2 1129 batch_free = count;
1d16871d 1130
a6f9edd6 1131 do {
770c8aaa
BZ
1132 int mt; /* migratetype of the to-be-freed page */
1133
a16601c5 1134 page = list_last_entry(list, struct page, lru);
a6f9edd6
MG
1135 /* must delete as __free_one_page list manipulates */
1136 list_del(&page->lru);
aa016d14 1137
bb14c2c7 1138 mt = get_pcppage_migratetype(page);
aa016d14
VB
1139 /* MIGRATE_ISOLATE page should not go to pcplists */
1140 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1141 /* Pageblock could have been isolated meanwhile */
3777999d 1142 if (unlikely(isolated_pageblocks))
51bb1a40 1143 mt = get_pageblock_migratetype(page);
51bb1a40 1144
4db7548c
MG
1145 if (bulkfree_pcp_prepare(page))
1146 continue;
1147
dc4b0caf 1148 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
770c8aaa 1149 trace_mm_page_pcpu_drain(page, 0, mt);
e5b31ac2 1150 } while (--count && --batch_free && !list_empty(list));
1da177e4 1151 }
d34b0733 1152 spin_unlock(&zone->lock);
1da177e4
LT
1153}
1154
dc4b0caf
MG
1155static void free_one_page(struct zone *zone,
1156 struct page *page, unsigned long pfn,
7aeb09f9 1157 unsigned int order,
ed0ae21d 1158 int migratetype)
1da177e4 1159{
d34b0733 1160 spin_lock(&zone->lock);
ad53f92e
JK
1161 if (unlikely(has_isolate_pageblock(zone) ||
1162 is_migrate_isolate(migratetype))) {
1163 migratetype = get_pfnblock_migratetype(page, pfn);
ad53f92e 1164 }
dc4b0caf 1165 __free_one_page(page, pfn, zone, order, migratetype);
d34b0733 1166 spin_unlock(&zone->lock);
48db57f8
NP
1167}
1168
1e8ce83c
RH
1169static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1170 unsigned long zone, int nid)
1171{
1e8ce83c 1172 set_page_links(page, zone, nid, pfn);
1e8ce83c
RH
1173 init_page_count(page);
1174 page_mapcount_reset(page);
1175 page_cpupid_reset_last(page);
1e8ce83c 1176
1e8ce83c
RH
1177 INIT_LIST_HEAD(&page->lru);
1178#ifdef WANT_PAGE_VIRTUAL
1179 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1180 if (!is_highmem_idx(zone))
1181 set_page_address(page, __va(pfn << PAGE_SHIFT));
1182#endif
1183}
1184
1185static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1186 int nid)
1187{
1188 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1189}
1190
7e18adb4
MG
1191#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1192static void init_reserved_page(unsigned long pfn)
1193{
1194 pg_data_t *pgdat;
1195 int nid, zid;
1196
1197 if (!early_page_uninitialised(pfn))
1198 return;
1199
1200 nid = early_pfn_to_nid(pfn);
1201 pgdat = NODE_DATA(nid);
1202
1203 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1204 struct zone *zone = &pgdat->node_zones[zid];
1205
1206 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1207 break;
1208 }
1209 __init_single_pfn(pfn, zid, nid);
1210}
1211#else
1212static inline void init_reserved_page(unsigned long pfn)
1213{
1214}
1215#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1216
92923ca3
NZ
1217/*
1218 * Initialised pages do not have PageReserved set. This function is
1219 * called for each range allocated by the bootmem allocator and
1220 * marks the pages PageReserved. The remaining valid pages are later
1221 * sent to the buddy page allocator.
1222 */
4b50bcc7 1223void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
92923ca3
NZ
1224{
1225 unsigned long start_pfn = PFN_DOWN(start);
1226 unsigned long end_pfn = PFN_UP(end);
1227
7e18adb4
MG
1228 for (; start_pfn < end_pfn; start_pfn++) {
1229 if (pfn_valid(start_pfn)) {
1230 struct page *page = pfn_to_page(start_pfn);
1231
1232 init_reserved_page(start_pfn);
1d798ca3
KS
1233
1234 /* Avoid false-positive PageTail() */
1235 INIT_LIST_HEAD(&page->lru);
1236
7e18adb4
MG
1237 SetPageReserved(page);
1238 }
1239 }
92923ca3
NZ
1240}
1241
ec95f53a
KM
1242static void __free_pages_ok(struct page *page, unsigned int order)
1243{
d34b0733 1244 unsigned long flags;
95e34412 1245 int migratetype;
dc4b0caf 1246 unsigned long pfn = page_to_pfn(page);
ec95f53a 1247
e2769dbd 1248 if (!free_pages_prepare(page, order, true))
ec95f53a
KM
1249 return;
1250
cfc47a28 1251 migratetype = get_pfnblock_migratetype(page, pfn);
d34b0733
MG
1252 local_irq_save(flags);
1253 __count_vm_events(PGFREE, 1 << order);
dc4b0caf 1254 free_one_page(page_zone(page), page, pfn, order, migratetype);
d34b0733 1255 local_irq_restore(flags);
1da177e4
LT
1256}
1257
949698a3 1258static void __init __free_pages_boot_core(struct page *page, unsigned int order)
a226f6c8 1259{
c3993076 1260 unsigned int nr_pages = 1 << order;
e2d0bd2b 1261 struct page *p = page;
c3993076 1262 unsigned int loop;
a226f6c8 1263
e2d0bd2b
YL
1264 prefetchw(p);
1265 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1266 prefetchw(p + 1);
c3993076
JW
1267 __ClearPageReserved(p);
1268 set_page_count(p, 0);
a226f6c8 1269 }
e2d0bd2b
YL
1270 __ClearPageReserved(p);
1271 set_page_count(p, 0);
c3993076 1272
e2d0bd2b 1273 page_zone(page)->managed_pages += nr_pages;
c3993076
JW
1274 set_page_refcounted(page);
1275 __free_pages(page, order);
a226f6c8
DH
1276}
1277
75a592a4
MG
1278#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1279 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
7ace9917 1280
75a592a4
MG
1281static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1282
1283int __meminit early_pfn_to_nid(unsigned long pfn)
1284{
7ace9917 1285 static DEFINE_SPINLOCK(early_pfn_lock);
75a592a4
MG
1286 int nid;
1287
7ace9917 1288 spin_lock(&early_pfn_lock);
75a592a4 1289 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
7ace9917 1290 if (nid < 0)
e4568d38 1291 nid = first_online_node;
7ace9917
MG
1292 spin_unlock(&early_pfn_lock);
1293
1294 return nid;
75a592a4
MG
1295}
1296#endif
1297
1298#ifdef CONFIG_NODES_SPAN_OTHER_NODES
d73d3c9f
MK
1299static inline bool __meminit __maybe_unused
1300meminit_pfn_in_nid(unsigned long pfn, int node,
1301 struct mminit_pfnnid_cache *state)
75a592a4
MG
1302{
1303 int nid;
1304
1305 nid = __early_pfn_to_nid(pfn, state);
1306 if (nid >= 0 && nid != node)
1307 return false;
1308 return true;
1309}
1310
1311/* Only safe to use early in boot when initialisation is single-threaded */
1312static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1313{
1314 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1315}
1316
1317#else
1318
1319static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1320{
1321 return true;
1322}
d73d3c9f
MK
1323static inline bool __meminit __maybe_unused
1324meminit_pfn_in_nid(unsigned long pfn, int node,
1325 struct mminit_pfnnid_cache *state)
75a592a4
MG
1326{
1327 return true;
1328}
1329#endif
1330
1331
0e1cc95b 1332void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
3a80a7fa
MG
1333 unsigned int order)
1334{
1335 if (early_page_uninitialised(pfn))
1336 return;
949698a3 1337 return __free_pages_boot_core(page, order);
3a80a7fa
MG
1338}
1339
7cf91a98
JK
1340/*
1341 * Check that the whole (or subset of) a pageblock given by the interval of
1342 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1343 * with the migration of free compaction scanner. The scanners then need to
1344 * use only pfn_valid_within() check for arches that allow holes within
1345 * pageblocks.
1346 *
1347 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1348 *
1349 * It's possible on some configurations to have a setup like node0 node1 node0
1350 * i.e. it's possible that all pages within a zones range of pages do not
1351 * belong to a single zone. We assume that a border between node0 and node1
1352 * can occur within a single pageblock, but not a node0 node1 node0
1353 * interleaving within a single pageblock. It is therefore sufficient to check
1354 * the first and last page of a pageblock and avoid checking each individual
1355 * page in a pageblock.
1356 */
1357struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1358 unsigned long end_pfn, struct zone *zone)
1359{
1360 struct page *start_page;
1361 struct page *end_page;
1362
1363 /* end_pfn is one past the range we are checking */
1364 end_pfn--;
1365
1366 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1367 return NULL;
1368
2d070eab
MH
1369 start_page = pfn_to_online_page(start_pfn);
1370 if (!start_page)
1371 return NULL;
7cf91a98
JK
1372
1373 if (page_zone(start_page) != zone)
1374 return NULL;
1375
1376 end_page = pfn_to_page(end_pfn);
1377
1378 /* This gives a shorter code than deriving page_zone(end_page) */
1379 if (page_zone_id(start_page) != page_zone_id(end_page))
1380 return NULL;
1381
1382 return start_page;
1383}
1384
1385void set_zone_contiguous(struct zone *zone)
1386{
1387 unsigned long block_start_pfn = zone->zone_start_pfn;
1388 unsigned long block_end_pfn;
1389
1390 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1391 for (; block_start_pfn < zone_end_pfn(zone);
1392 block_start_pfn = block_end_pfn,
1393 block_end_pfn += pageblock_nr_pages) {
1394
1395 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1396
1397 if (!__pageblock_pfn_to_page(block_start_pfn,
1398 block_end_pfn, zone))
1399 return;
1400 }
1401
1402 /* We confirm that there is no hole */
1403 zone->contiguous = true;
1404}
1405
1406void clear_zone_contiguous(struct zone *zone)
1407{
1408 zone->contiguous = false;
1409}
1410
7e18adb4 1411#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
0e1cc95b 1412static void __init deferred_free_range(struct page *page,
a4de83dd
MG
1413 unsigned long pfn, int nr_pages)
1414{
1415 int i;
1416
1417 if (!page)
1418 return;
1419
1420 /* Free a large naturally-aligned chunk if possible */
e780149b
XQ
1421 if (nr_pages == pageblock_nr_pages &&
1422 (pfn & (pageblock_nr_pages - 1)) == 0) {
ac5d2539 1423 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
e780149b 1424 __free_pages_boot_core(page, pageblock_order);
a4de83dd
MG
1425 return;
1426 }
1427
e780149b
XQ
1428 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1429 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1430 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
949698a3 1431 __free_pages_boot_core(page, 0);
e780149b 1432 }
a4de83dd
MG
1433}
1434
d3cd131d
NS
1435/* Completion tracking for deferred_init_memmap() threads */
1436static atomic_t pgdat_init_n_undone __initdata;
1437static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1438
1439static inline void __init pgdat_init_report_one_done(void)
1440{
1441 if (atomic_dec_and_test(&pgdat_init_n_undone))
1442 complete(&pgdat_init_all_done_comp);
1443}
0e1cc95b 1444
7e18adb4 1445/* Initialise remaining memory on a node */
0e1cc95b 1446static int __init deferred_init_memmap(void *data)
7e18adb4 1447{
0e1cc95b
MG
1448 pg_data_t *pgdat = data;
1449 int nid = pgdat->node_id;
7e18adb4
MG
1450 struct mminit_pfnnid_cache nid_init_state = { };
1451 unsigned long start = jiffies;
1452 unsigned long nr_pages = 0;
1453 unsigned long walk_start, walk_end;
1454 int i, zid;
1455 struct zone *zone;
7e18adb4 1456 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
0e1cc95b 1457 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
7e18adb4 1458
0e1cc95b 1459 if (first_init_pfn == ULONG_MAX) {
d3cd131d 1460 pgdat_init_report_one_done();
0e1cc95b
MG
1461 return 0;
1462 }
1463
1464 /* Bind memory initialisation thread to a local node if possible */
1465 if (!cpumask_empty(cpumask))
1466 set_cpus_allowed_ptr(current, cpumask);
7e18adb4
MG
1467
1468 /* Sanity check boundaries */
1469 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1470 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1471 pgdat->first_deferred_pfn = ULONG_MAX;
1472
1473 /* Only the highest zone is deferred so find it */
1474 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1475 zone = pgdat->node_zones + zid;
1476 if (first_init_pfn < zone_end_pfn(zone))
1477 break;
1478 }
1479
1480 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1481 unsigned long pfn, end_pfn;
54608c3f 1482 struct page *page = NULL;
a4de83dd
MG
1483 struct page *free_base_page = NULL;
1484 unsigned long free_base_pfn = 0;
1485 int nr_to_free = 0;
7e18adb4
MG
1486
1487 end_pfn = min(walk_end, zone_end_pfn(zone));
1488 pfn = first_init_pfn;
1489 if (pfn < walk_start)
1490 pfn = walk_start;
1491 if (pfn < zone->zone_start_pfn)
1492 pfn = zone->zone_start_pfn;
1493
1494 for (; pfn < end_pfn; pfn++) {
54608c3f 1495 if (!pfn_valid_within(pfn))
a4de83dd 1496 goto free_range;
7e18adb4 1497
54608c3f
MG
1498 /*
1499 * Ensure pfn_valid is checked every
e780149b 1500 * pageblock_nr_pages for memory holes
54608c3f 1501 */
e780149b 1502 if ((pfn & (pageblock_nr_pages - 1)) == 0) {
54608c3f
MG
1503 if (!pfn_valid(pfn)) {
1504 page = NULL;
a4de83dd 1505 goto free_range;
54608c3f
MG
1506 }
1507 }
1508
1509 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1510 page = NULL;
a4de83dd 1511 goto free_range;
54608c3f
MG
1512 }
1513
1514 /* Minimise pfn page lookups and scheduler checks */
e780149b 1515 if (page && (pfn & (pageblock_nr_pages - 1)) != 0) {
54608c3f
MG
1516 page++;
1517 } else {
a4de83dd
MG
1518 nr_pages += nr_to_free;
1519 deferred_free_range(free_base_page,
1520 free_base_pfn, nr_to_free);
1521 free_base_page = NULL;
1522 free_base_pfn = nr_to_free = 0;
1523
54608c3f
MG
1524 page = pfn_to_page(pfn);
1525 cond_resched();
1526 }
7e18adb4
MG
1527
1528 if (page->flags) {
1529 VM_BUG_ON(page_zone(page) != zone);
a4de83dd 1530 goto free_range;
7e18adb4
MG
1531 }
1532
1533 __init_single_page(page, pfn, zid, nid);
a4de83dd
MG
1534 if (!free_base_page) {
1535 free_base_page = page;
1536 free_base_pfn = pfn;
1537 nr_to_free = 0;
1538 }
1539 nr_to_free++;
1540
1541 /* Where possible, batch up pages for a single free */
1542 continue;
1543free_range:
1544 /* Free the current block of pages to allocator */
1545 nr_pages += nr_to_free;
1546 deferred_free_range(free_base_page, free_base_pfn,
1547 nr_to_free);
1548 free_base_page = NULL;
1549 free_base_pfn = nr_to_free = 0;
7e18adb4 1550 }
e780149b
XQ
1551 /* Free the last block of pages to allocator */
1552 nr_pages += nr_to_free;
1553 deferred_free_range(free_base_page, free_base_pfn, nr_to_free);
a4de83dd 1554
7e18adb4
MG
1555 first_init_pfn = max(end_pfn, first_init_pfn);
1556 }
1557
1558 /* Sanity check that the next zone really is unpopulated */
1559 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1560
0e1cc95b 1561 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
7e18adb4 1562 jiffies_to_msecs(jiffies - start));
d3cd131d
NS
1563
1564 pgdat_init_report_one_done();
0e1cc95b
MG
1565 return 0;
1566}
7cf91a98 1567#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
0e1cc95b
MG
1568
1569void __init page_alloc_init_late(void)
1570{
7cf91a98
JK
1571 struct zone *zone;
1572
1573#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
0e1cc95b
MG
1574 int nid;
1575
d3cd131d
NS
1576 /* There will be num_node_state(N_MEMORY) threads */
1577 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
0e1cc95b 1578 for_each_node_state(nid, N_MEMORY) {
0e1cc95b
MG
1579 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1580 }
1581
1582 /* Block until all are initialised */
d3cd131d 1583 wait_for_completion(&pgdat_init_all_done_comp);
4248b0da
MG
1584
1585 /* Reinit limits that are based on free pages after the kernel is up */
1586 files_maxfiles_init();
7cf91a98 1587#endif
3010f876
PT
1588#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
1589 /* Discard memblock private memory */
1590 memblock_discard();
1591#endif
7cf91a98
JK
1592
1593 for_each_populated_zone(zone)
1594 set_zone_contiguous(zone);
7e18adb4 1595}
7e18adb4 1596
47118af0 1597#ifdef CONFIG_CMA
9cf510a5 1598/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
47118af0
MN
1599void __init init_cma_reserved_pageblock(struct page *page)
1600{
1601 unsigned i = pageblock_nr_pages;
1602 struct page *p = page;
1603
1604 do {
1605 __ClearPageReserved(p);
1606 set_page_count(p, 0);
1607 } while (++p, --i);
1608
47118af0 1609 set_pageblock_migratetype(page, MIGRATE_CMA);
dc78327c
MN
1610
1611 if (pageblock_order >= MAX_ORDER) {
1612 i = pageblock_nr_pages;
1613 p = page;
1614 do {
1615 set_page_refcounted(p);
1616 __free_pages(p, MAX_ORDER - 1);
1617 p += MAX_ORDER_NR_PAGES;
1618 } while (i -= MAX_ORDER_NR_PAGES);
1619 } else {
1620 set_page_refcounted(page);
1621 __free_pages(page, pageblock_order);
1622 }
1623
3dcc0571 1624 adjust_managed_page_count(page, pageblock_nr_pages);
47118af0
MN
1625}
1626#endif
1da177e4
LT
1627
1628/*
1629 * The order of subdivision here is critical for the IO subsystem.
1630 * Please do not alter this order without good reasons and regression
1631 * testing. Specifically, as large blocks of memory are subdivided,
1632 * the order in which smaller blocks are delivered depends on the order
1633 * they're subdivided in this function. This is the primary factor
1634 * influencing the order in which pages are delivered to the IO
1635 * subsystem according to empirical testing, and this is also justified
1636 * by considering the behavior of a buddy system containing a single
1637 * large block of memory acted on by a series of small allocations.
1638 * This behavior is a critical factor in sglist merging's success.
1639 *
6d49e352 1640 * -- nyc
1da177e4 1641 */
085cc7d5 1642static inline void expand(struct zone *zone, struct page *page,
b2a0ac88
MG
1643 int low, int high, struct free_area *area,
1644 int migratetype)
1da177e4
LT
1645{
1646 unsigned long size = 1 << high;
1647
1648 while (high > low) {
1649 area--;
1650 high--;
1651 size >>= 1;
309381fe 1652 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
c0a32fc5 1653
acbc15a4
JK
1654 /*
1655 * Mark as guard pages (or page), that will allow to
1656 * merge back to allocator when buddy will be freed.
1657 * Corresponding page table entries will not be touched,
1658 * pages will stay not present in virtual address space
1659 */
1660 if (set_page_guard(zone, &page[size], high, migratetype))
c0a32fc5 1661 continue;
acbc15a4 1662
b2a0ac88 1663 list_add(&page[size].lru, &area->free_list[migratetype]);
1da177e4
LT
1664 area->nr_free++;
1665 set_page_order(&page[size], high);
1666 }
1da177e4
LT
1667}
1668
4e611801 1669static void check_new_page_bad(struct page *page)
1da177e4 1670{
4e611801
VB
1671 const char *bad_reason = NULL;
1672 unsigned long bad_flags = 0;
7bfec6f4 1673
53f9263b 1674 if (unlikely(atomic_read(&page->_mapcount) != -1))
f0b791a3
DH
1675 bad_reason = "nonzero mapcount";
1676 if (unlikely(page->mapping != NULL))
1677 bad_reason = "non-NULL mapping";
fe896d18 1678 if (unlikely(page_ref_count(page) != 0))
f0b791a3 1679 bad_reason = "nonzero _count";
f4c18e6f
NH
1680 if (unlikely(page->flags & __PG_HWPOISON)) {
1681 bad_reason = "HWPoisoned (hardware-corrupted)";
1682 bad_flags = __PG_HWPOISON;
e570f56c
NH
1683 /* Don't complain about hwpoisoned pages */
1684 page_mapcount_reset(page); /* remove PageBuddy */
1685 return;
f4c18e6f 1686 }
f0b791a3
DH
1687 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1688 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1689 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1690 }
9edad6ea
JW
1691#ifdef CONFIG_MEMCG
1692 if (unlikely(page->mem_cgroup))
1693 bad_reason = "page still charged to cgroup";
1694#endif
4e611801
VB
1695 bad_page(page, bad_reason, bad_flags);
1696}
1697
1698/*
1699 * This page is about to be returned from the page allocator
1700 */
1701static inline int check_new_page(struct page *page)
1702{
1703 if (likely(page_expected_state(page,
1704 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1705 return 0;
1706
1707 check_new_page_bad(page);
1708 return 1;
2a7684a2
WF
1709}
1710
bd33ef36 1711static inline bool free_pages_prezeroed(void)
1414c7f4
LA
1712{
1713 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
bd33ef36 1714 page_poisoning_enabled();
1414c7f4
LA
1715}
1716
479f854a
MG
1717#ifdef CONFIG_DEBUG_VM
1718static bool check_pcp_refill(struct page *page)
1719{
1720 return false;
1721}
1722
1723static bool check_new_pcp(struct page *page)
1724{
1725 return check_new_page(page);
1726}
1727#else
1728static bool check_pcp_refill(struct page *page)
1729{
1730 return check_new_page(page);
1731}
1732static bool check_new_pcp(struct page *page)
1733{
1734 return false;
1735}
1736#endif /* CONFIG_DEBUG_VM */
1737
1738static bool check_new_pages(struct page *page, unsigned int order)
1739{
1740 int i;
1741 for (i = 0; i < (1 << order); i++) {
1742 struct page *p = page + i;
1743
1744 if (unlikely(check_new_page(p)))
1745 return true;
1746 }
1747
1748 return false;
1749}
1750
46f24fd8
JK
1751inline void post_alloc_hook(struct page *page, unsigned int order,
1752 gfp_t gfp_flags)
1753{
1754 set_page_private(page, 0);
1755 set_page_refcounted(page);
1756
1757 arch_alloc_page(page, order);
1758 kernel_map_pages(page, 1 << order, 1);
1759 kernel_poison_pages(page, 1 << order, 1);
1760 kasan_alloc_pages(page, order);
1761 set_page_owner(page, order, gfp_flags);
1762}
1763
479f854a 1764static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
c603844b 1765 unsigned int alloc_flags)
2a7684a2
WF
1766{
1767 int i;
689bcebf 1768
46f24fd8 1769 post_alloc_hook(page, order, gfp_flags);
17cf4406 1770
bd33ef36 1771 if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
f4d2897b
AA
1772 for (i = 0; i < (1 << order); i++)
1773 clear_highpage(page + i);
17cf4406
NP
1774
1775 if (order && (gfp_flags & __GFP_COMP))
1776 prep_compound_page(page, order);
1777
75379191 1778 /*
2f064f34 1779 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
75379191
VB
1780 * allocate the page. The expectation is that the caller is taking
1781 * steps that will free more memory. The caller should avoid the page
1782 * being used for !PFMEMALLOC purposes.
1783 */
2f064f34
MH
1784 if (alloc_flags & ALLOC_NO_WATERMARKS)
1785 set_page_pfmemalloc(page);
1786 else
1787 clear_page_pfmemalloc(page);
1da177e4
LT
1788}
1789
56fd56b8
MG
1790/*
1791 * Go through the free lists for the given migratetype and remove
1792 * the smallest available page from the freelists
1793 */
728ec980
MG
1794static inline
1795struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
56fd56b8
MG
1796 int migratetype)
1797{
1798 unsigned int current_order;
b8af2941 1799 struct free_area *area;
56fd56b8
MG
1800 struct page *page;
1801
1802 /* Find a page of the appropriate size in the preferred list */
1803 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1804 area = &(zone->free_area[current_order]);
a16601c5 1805 page = list_first_entry_or_null(&area->free_list[migratetype],
56fd56b8 1806 struct page, lru);
a16601c5
GT
1807 if (!page)
1808 continue;
56fd56b8
MG
1809 list_del(&page->lru);
1810 rmv_page_order(page);
1811 area->nr_free--;
56fd56b8 1812 expand(zone, page, order, current_order, area, migratetype);
bb14c2c7 1813 set_pcppage_migratetype(page, migratetype);
56fd56b8
MG
1814 return page;
1815 }
1816
1817 return NULL;
1818}
1819
1820
b2a0ac88
MG
1821/*
1822 * This array describes the order lists are fallen back to when
1823 * the free lists for the desirable migrate type are depleted
1824 */
47118af0 1825static int fallbacks[MIGRATE_TYPES][4] = {
974a786e
MG
1826 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1827 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1828 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
47118af0 1829#ifdef CONFIG_CMA
974a786e 1830 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
47118af0 1831#endif
194159fb 1832#ifdef CONFIG_MEMORY_ISOLATION
974a786e 1833 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
194159fb 1834#endif
b2a0ac88
MG
1835};
1836
dc67647b
JK
1837#ifdef CONFIG_CMA
1838static struct page *__rmqueue_cma_fallback(struct zone *zone,
1839 unsigned int order)
1840{
1841 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1842}
1843#else
1844static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1845 unsigned int order) { return NULL; }
1846#endif
1847
c361be55
MG
1848/*
1849 * Move the free pages in a range to the free lists of the requested type.
d9c23400 1850 * Note that start_page and end_pages are not aligned on a pageblock
c361be55
MG
1851 * boundary. If alignment is required, use move_freepages_block()
1852 */
02aa0cdd 1853static int move_freepages(struct zone *zone,
b69a7288 1854 struct page *start_page, struct page *end_page,
02aa0cdd 1855 int migratetype, int *num_movable)
c361be55
MG
1856{
1857 struct page *page;
d00181b9 1858 unsigned int order;
d100313f 1859 int pages_moved = 0;
c361be55
MG
1860
1861#ifndef CONFIG_HOLES_IN_ZONE
1862 /*
1863 * page_zone is not safe to call in this context when
1864 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1865 * anyway as we check zone boundaries in move_freepages_block().
1866 * Remove at a later date when no bug reports exist related to
ac0e5b7a 1867 * grouping pages by mobility
c361be55 1868 */
97ee4ba7 1869 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
c361be55
MG
1870#endif
1871
02aa0cdd
VB
1872 if (num_movable)
1873 *num_movable = 0;
1874
c361be55
MG
1875 for (page = start_page; page <= end_page;) {
1876 if (!pfn_valid_within(page_to_pfn(page))) {
1877 page++;
1878 continue;
1879 }
1880
f073bdc5
AB
1881 /* Make sure we are not inadvertently changing nodes */
1882 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1883
c361be55 1884 if (!PageBuddy(page)) {
02aa0cdd
VB
1885 /*
1886 * We assume that pages that could be isolated for
1887 * migration are movable. But we don't actually try
1888 * isolating, as that would be expensive.
1889 */
1890 if (num_movable &&
1891 (PageLRU(page) || __PageMovable(page)))
1892 (*num_movable)++;
1893
c361be55
MG
1894 page++;
1895 continue;
1896 }
1897
1898 order = page_order(page);
84be48d8
KS
1899 list_move(&page->lru,
1900 &zone->free_area[order].free_list[migratetype]);
c361be55 1901 page += 1 << order;
d100313f 1902 pages_moved += 1 << order;
c361be55
MG
1903 }
1904
d100313f 1905 return pages_moved;
c361be55
MG
1906}
1907
ee6f509c 1908int move_freepages_block(struct zone *zone, struct page *page,
02aa0cdd 1909 int migratetype, int *num_movable)
c361be55
MG
1910{
1911 unsigned long start_pfn, end_pfn;
1912 struct page *start_page, *end_page;
1913
1914 start_pfn = page_to_pfn(page);
d9c23400 1915 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
c361be55 1916 start_page = pfn_to_page(start_pfn);
d9c23400
MG
1917 end_page = start_page + pageblock_nr_pages - 1;
1918 end_pfn = start_pfn + pageblock_nr_pages - 1;
c361be55
MG
1919
1920 /* Do not cross zone boundaries */
108bcc96 1921 if (!zone_spans_pfn(zone, start_pfn))
c361be55 1922 start_page = page;
108bcc96 1923 if (!zone_spans_pfn(zone, end_pfn))
c361be55
MG
1924 return 0;
1925
02aa0cdd
VB
1926 return move_freepages(zone, start_page, end_page, migratetype,
1927 num_movable);
c361be55
MG
1928}
1929
2f66a68f
MG
1930static void change_pageblock_range(struct page *pageblock_page,
1931 int start_order, int migratetype)
1932{
1933 int nr_pageblocks = 1 << (start_order - pageblock_order);
1934
1935 while (nr_pageblocks--) {
1936 set_pageblock_migratetype(pageblock_page, migratetype);
1937 pageblock_page += pageblock_nr_pages;
1938 }
1939}
1940
fef903ef 1941/*
9c0415eb
VB
1942 * When we are falling back to another migratetype during allocation, try to
1943 * steal extra free pages from the same pageblocks to satisfy further
1944 * allocations, instead of polluting multiple pageblocks.
1945 *
1946 * If we are stealing a relatively large buddy page, it is likely there will
1947 * be more free pages in the pageblock, so try to steal them all. For
1948 * reclaimable and unmovable allocations, we steal regardless of page size,
1949 * as fragmentation caused by those allocations polluting movable pageblocks
1950 * is worse than movable allocations stealing from unmovable and reclaimable
1951 * pageblocks.
fef903ef 1952 */
4eb7dce6
JK
1953static bool can_steal_fallback(unsigned int order, int start_mt)
1954{
1955 /*
1956 * Leaving this order check is intended, although there is
1957 * relaxed order check in next check. The reason is that
1958 * we can actually steal whole pageblock if this condition met,
1959 * but, below check doesn't guarantee it and that is just heuristic
1960 * so could be changed anytime.
1961 */
1962 if (order >= pageblock_order)
1963 return true;
1964
1965 if (order >= pageblock_order / 2 ||
1966 start_mt == MIGRATE_RECLAIMABLE ||
1967 start_mt == MIGRATE_UNMOVABLE ||
1968 page_group_by_mobility_disabled)
1969 return true;
1970
1971 return false;
1972}
1973
1974/*
1975 * This function implements actual steal behaviour. If order is large enough,
1976 * we can steal whole pageblock. If not, we first move freepages in this
02aa0cdd
VB
1977 * pageblock to our migratetype and determine how many already-allocated pages
1978 * are there in the pageblock with a compatible migratetype. If at least half
1979 * of pages are free or compatible, we can change migratetype of the pageblock
1980 * itself, so pages freed in the future will be put on the correct free list.
4eb7dce6
JK
1981 */
1982static void steal_suitable_fallback(struct zone *zone, struct page *page,
3bc48f96 1983 int start_type, bool whole_block)
fef903ef 1984{
d00181b9 1985 unsigned int current_order = page_order(page);
3bc48f96 1986 struct free_area *area;
02aa0cdd
VB
1987 int free_pages, movable_pages, alike_pages;
1988 int old_block_type;
1989
1990 old_block_type = get_pageblock_migratetype(page);
fef903ef 1991
3bc48f96
VB
1992 /*
1993 * This can happen due to races and we want to prevent broken
1994 * highatomic accounting.
1995 */
02aa0cdd 1996 if (is_migrate_highatomic(old_block_type))
3bc48f96
VB
1997 goto single_page;
1998
fef903ef
SB
1999 /* Take ownership for orders >= pageblock_order */
2000 if (current_order >= pageblock_order) {
2001 change_pageblock_range(page, current_order, start_type);
3bc48f96 2002 goto single_page;
fef903ef
SB
2003 }
2004
3bc48f96
VB
2005 /* We are not allowed to try stealing from the whole block */
2006 if (!whole_block)
2007 goto single_page;
2008
02aa0cdd
VB
2009 free_pages = move_freepages_block(zone, page, start_type,
2010 &movable_pages);
2011 /*
2012 * Determine how many pages are compatible with our allocation.
2013 * For movable allocation, it's the number of movable pages which
2014 * we just obtained. For other types it's a bit more tricky.
2015 */
2016 if (start_type == MIGRATE_MOVABLE) {
2017 alike_pages = movable_pages;
2018 } else {
2019 /*
2020 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2021 * to MOVABLE pageblock, consider all non-movable pages as
2022 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2023 * vice versa, be conservative since we can't distinguish the
2024 * exact migratetype of non-movable pages.
2025 */
2026 if (old_block_type == MIGRATE_MOVABLE)
2027 alike_pages = pageblock_nr_pages
2028 - (free_pages + movable_pages);
2029 else
2030 alike_pages = 0;
2031 }
2032
3bc48f96 2033 /* moving whole block can fail due to zone boundary conditions */
02aa0cdd 2034 if (!free_pages)
3bc48f96 2035 goto single_page;
fef903ef 2036
02aa0cdd
VB
2037 /*
2038 * If a sufficient number of pages in the block are either free or of
2039 * comparable migratability as our allocation, claim the whole block.
2040 */
2041 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
4eb7dce6
JK
2042 page_group_by_mobility_disabled)
2043 set_pageblock_migratetype(page, start_type);
3bc48f96
VB
2044
2045 return;
2046
2047single_page:
2048 area = &zone->free_area[current_order];
2049 list_move(&page->lru, &area->free_list[start_type]);
4eb7dce6
JK
2050}
2051
2149cdae
JK
2052/*
2053 * Check whether there is a suitable fallback freepage with requested order.
2054 * If only_stealable is true, this function returns fallback_mt only if
2055 * we can steal other freepages all together. This would help to reduce
2056 * fragmentation due to mixed migratetype pages in one pageblock.
2057 */
2058int find_suitable_fallback(struct free_area *area, unsigned int order,
2059 int migratetype, bool only_stealable, bool *can_steal)
4eb7dce6
JK
2060{
2061 int i;
2062 int fallback_mt;
2063
2064 if (area->nr_free == 0)
2065 return -1;
2066
2067 *can_steal = false;
2068 for (i = 0;; i++) {
2069 fallback_mt = fallbacks[migratetype][i];
974a786e 2070 if (fallback_mt == MIGRATE_TYPES)
4eb7dce6
JK
2071 break;
2072
2073 if (list_empty(&area->free_list[fallback_mt]))
2074 continue;
fef903ef 2075
4eb7dce6
JK
2076 if (can_steal_fallback(order, migratetype))
2077 *can_steal = true;
2078
2149cdae
JK
2079 if (!only_stealable)
2080 return fallback_mt;
2081
2082 if (*can_steal)
2083 return fallback_mt;
fef903ef 2084 }
4eb7dce6
JK
2085
2086 return -1;
fef903ef
SB
2087}
2088
0aaa29a5
MG
2089/*
2090 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2091 * there are no empty page blocks that contain a page with a suitable order
2092 */
2093static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2094 unsigned int alloc_order)
2095{
2096 int mt;
2097 unsigned long max_managed, flags;
2098
2099 /*
2100 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2101 * Check is race-prone but harmless.
2102 */
2103 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2104 if (zone->nr_reserved_highatomic >= max_managed)
2105 return;
2106
2107 spin_lock_irqsave(&zone->lock, flags);
2108
2109 /* Recheck the nr_reserved_highatomic limit under the lock */
2110 if (zone->nr_reserved_highatomic >= max_managed)
2111 goto out_unlock;
2112
2113 /* Yoink! */
2114 mt = get_pageblock_migratetype(page);
a6ffdc07
XQ
2115 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2116 && !is_migrate_cma(mt)) {
0aaa29a5
MG
2117 zone->nr_reserved_highatomic += pageblock_nr_pages;
2118 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
02aa0cdd 2119 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
0aaa29a5
MG
2120 }
2121
2122out_unlock:
2123 spin_unlock_irqrestore(&zone->lock, flags);
2124}
2125
2126/*
2127 * Used when an allocation is about to fail under memory pressure. This
2128 * potentially hurts the reliability of high-order allocations when under
2129 * intense memory pressure but failed atomic allocations should be easier
2130 * to recover from than an OOM.
29fac03b
MK
2131 *
2132 * If @force is true, try to unreserve a pageblock even though highatomic
2133 * pageblock is exhausted.
0aaa29a5 2134 */
29fac03b
MK
2135static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2136 bool force)
0aaa29a5
MG
2137{
2138 struct zonelist *zonelist = ac->zonelist;
2139 unsigned long flags;
2140 struct zoneref *z;
2141 struct zone *zone;
2142 struct page *page;
2143 int order;
04c8716f 2144 bool ret;
0aaa29a5
MG
2145
2146 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2147 ac->nodemask) {
29fac03b
MK
2148 /*
2149 * Preserve at least one pageblock unless memory pressure
2150 * is really high.
2151 */
2152 if (!force && zone->nr_reserved_highatomic <=
2153 pageblock_nr_pages)
0aaa29a5
MG
2154 continue;
2155
2156 spin_lock_irqsave(&zone->lock, flags);
2157 for (order = 0; order < MAX_ORDER; order++) {
2158 struct free_area *area = &(zone->free_area[order]);
2159
a16601c5
GT
2160 page = list_first_entry_or_null(
2161 &area->free_list[MIGRATE_HIGHATOMIC],
2162 struct page, lru);
2163 if (!page)
0aaa29a5
MG
2164 continue;
2165
0aaa29a5 2166 /*
4855e4a7
MK
2167 * In page freeing path, migratetype change is racy so
2168 * we can counter several free pages in a pageblock
2169 * in this loop althoug we changed the pageblock type
2170 * from highatomic to ac->migratetype. So we should
2171 * adjust the count once.
0aaa29a5 2172 */
a6ffdc07 2173 if (is_migrate_highatomic_page(page)) {
4855e4a7
MK
2174 /*
2175 * It should never happen but changes to
2176 * locking could inadvertently allow a per-cpu
2177 * drain to add pages to MIGRATE_HIGHATOMIC
2178 * while unreserving so be safe and watch for
2179 * underflows.
2180 */
2181 zone->nr_reserved_highatomic -= min(
2182 pageblock_nr_pages,
2183 zone->nr_reserved_highatomic);
2184 }
0aaa29a5
MG
2185
2186 /*
2187 * Convert to ac->migratetype and avoid the normal
2188 * pageblock stealing heuristics. Minimally, the caller
2189 * is doing the work and needs the pages. More
2190 * importantly, if the block was always converted to
2191 * MIGRATE_UNMOVABLE or another type then the number
2192 * of pageblocks that cannot be completely freed
2193 * may increase.
2194 */
2195 set_pageblock_migratetype(page, ac->migratetype);
02aa0cdd
VB
2196 ret = move_freepages_block(zone, page, ac->migratetype,
2197 NULL);
29fac03b
MK
2198 if (ret) {
2199 spin_unlock_irqrestore(&zone->lock, flags);
2200 return ret;
2201 }
0aaa29a5
MG
2202 }
2203 spin_unlock_irqrestore(&zone->lock, flags);
2204 }
04c8716f
MK
2205
2206 return false;
0aaa29a5
MG
2207}
2208
3bc48f96
VB
2209/*
2210 * Try finding a free buddy page on the fallback list and put it on the free
2211 * list of requested migratetype, possibly along with other pages from the same
2212 * block, depending on fragmentation avoidance heuristics. Returns true if
2213 * fallback was found so that __rmqueue_smallest() can grab it.
b002529d
RV
2214 *
2215 * The use of signed ints for order and current_order is a deliberate
2216 * deviation from the rest of this file, to make the for loop
2217 * condition simpler.
3bc48f96
VB
2218 */
2219static inline bool
b002529d 2220__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
b2a0ac88 2221{
b8af2941 2222 struct free_area *area;
b002529d 2223 int current_order;
b2a0ac88 2224 struct page *page;
4eb7dce6
JK
2225 int fallback_mt;
2226 bool can_steal;
b2a0ac88 2227
7a8f58f3
VB
2228 /*
2229 * Find the largest available free page in the other list. This roughly
2230 * approximates finding the pageblock with the most free pages, which
2231 * would be too costly to do exactly.
2232 */
b002529d 2233 for (current_order = MAX_ORDER - 1; current_order >= order;
7aeb09f9 2234 --current_order) {
4eb7dce6
JK
2235 area = &(zone->free_area[current_order]);
2236 fallback_mt = find_suitable_fallback(area, current_order,
2149cdae 2237 start_migratetype, false, &can_steal);
4eb7dce6
JK
2238 if (fallback_mt == -1)
2239 continue;
b2a0ac88 2240
7a8f58f3
VB
2241 /*
2242 * We cannot steal all free pages from the pageblock and the
2243 * requested migratetype is movable. In that case it's better to
2244 * steal and split the smallest available page instead of the
2245 * largest available page, because even if the next movable
2246 * allocation falls back into a different pageblock than this
2247 * one, it won't cause permanent fragmentation.
2248 */
2249 if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2250 && current_order > order)
2251 goto find_smallest;
b2a0ac88 2252
7a8f58f3
VB
2253 goto do_steal;
2254 }
e0fff1bd 2255
7a8f58f3 2256 return false;
e0fff1bd 2257
7a8f58f3
VB
2258find_smallest:
2259 for (current_order = order; current_order < MAX_ORDER;
2260 current_order++) {
2261 area = &(zone->free_area[current_order]);
2262 fallback_mt = find_suitable_fallback(area, current_order,
2263 start_migratetype, false, &can_steal);
2264 if (fallback_mt != -1)
2265 break;
b2a0ac88
MG
2266 }
2267
7a8f58f3
VB
2268 /*
2269 * This should not happen - we already found a suitable fallback
2270 * when looking for the largest page.
2271 */
2272 VM_BUG_ON(current_order == MAX_ORDER);
2273
2274do_steal:
2275 page = list_first_entry(&area->free_list[fallback_mt],
2276 struct page, lru);
2277
2278 steal_suitable_fallback(zone, page, start_migratetype, can_steal);
2279
2280 trace_mm_page_alloc_extfrag(page, order, current_order,
2281 start_migratetype, fallback_mt);
2282
2283 return true;
2284
b2a0ac88
MG
2285}
2286
56fd56b8 2287/*
1da177e4
LT
2288 * Do the hard work of removing an element from the buddy allocator.
2289 * Call me with the zone->lock already held.
2290 */
b2a0ac88 2291static struct page *__rmqueue(struct zone *zone, unsigned int order,
6ac0206b 2292 int migratetype)
1da177e4 2293{
1da177e4
LT
2294 struct page *page;
2295
3bc48f96 2296retry:
56fd56b8 2297 page = __rmqueue_smallest(zone, order, migratetype);
974a786e 2298 if (unlikely(!page)) {
dc67647b
JK
2299 if (migratetype == MIGRATE_MOVABLE)
2300 page = __rmqueue_cma_fallback(zone, order);
2301
3bc48f96
VB
2302 if (!page && __rmqueue_fallback(zone, order, migratetype))
2303 goto retry;
728ec980
MG
2304 }
2305
0d3d062a 2306 trace_mm_page_alloc_zone_locked(page, order, migratetype);
b2a0ac88 2307 return page;
1da177e4
LT
2308}
2309
5f63b720 2310/*
1da177e4
LT
2311 * Obtain a specified number of elements from the buddy allocator, all under
2312 * a single hold of the lock, for efficiency. Add them to the supplied list.
2313 * Returns the number of new pages which were placed at *list.
2314 */
5f63b720 2315static int rmqueue_bulk(struct zone *zone, unsigned int order,
b2a0ac88 2316 unsigned long count, struct list_head *list,
b745bc85 2317 int migratetype, bool cold)
1da177e4 2318{
a6de734b 2319 int i, alloced = 0;
5f63b720 2320
d34b0733 2321 spin_lock(&zone->lock);
1da177e4 2322 for (i = 0; i < count; ++i) {
6ac0206b 2323 struct page *page = __rmqueue(zone, order, migratetype);
085cc7d5 2324 if (unlikely(page == NULL))
1da177e4 2325 break;
81eabcbe 2326
479f854a
MG
2327 if (unlikely(check_pcp_refill(page)))
2328 continue;
2329
81eabcbe
MG
2330 /*
2331 * Split buddy pages returned by expand() are received here
2332 * in physical page order. The page is added to the callers and
2333 * list and the list head then moves forward. From the callers
2334 * perspective, the linked list is ordered by page number in
2335 * some conditions. This is useful for IO devices that can
2336 * merge IO requests if the physical pages are ordered
2337 * properly.
2338 */
b745bc85 2339 if (likely(!cold))
e084b2d9
MG
2340 list_add(&page->lru, list);
2341 else
2342 list_add_tail(&page->lru, list);
81eabcbe 2343 list = &page->lru;
a6de734b 2344 alloced++;
bb14c2c7 2345 if (is_migrate_cma(get_pcppage_migratetype(page)))
d1ce749a
BZ
2346 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2347 -(1 << order));
1da177e4 2348 }
a6de734b
MG
2349
2350 /*
2351 * i pages were removed from the buddy list even if some leak due
2352 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2353 * on i. Do not confuse with 'alloced' which is the number of
2354 * pages added to the pcp list.
2355 */
f2260e6b 2356 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
d34b0733 2357 spin_unlock(&zone->lock);
a6de734b 2358 return alloced;
1da177e4
LT
2359}
2360
4ae7c039 2361#ifdef CONFIG_NUMA
8fce4d8e 2362/*
4037d452
CL
2363 * Called from the vmstat counter updater to drain pagesets of this
2364 * currently executing processor on remote nodes after they have
2365 * expired.
2366 *
879336c3
CL
2367 * Note that this function must be called with the thread pinned to
2368 * a single processor.
8fce4d8e 2369 */
4037d452 2370void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
4ae7c039 2371{
4ae7c039 2372 unsigned long flags;
7be12fc9 2373 int to_drain, batch;
4ae7c039 2374
4037d452 2375 local_irq_save(flags);
4db0c3c2 2376 batch = READ_ONCE(pcp->batch);
7be12fc9 2377 to_drain = min(pcp->count, batch);
2a13515c
KM
2378 if (to_drain > 0) {
2379 free_pcppages_bulk(zone, to_drain, pcp);
2380 pcp->count -= to_drain;
2381 }
4037d452 2382 local_irq_restore(flags);
4ae7c039
CL
2383}
2384#endif
2385
9f8f2172 2386/*
93481ff0 2387 * Drain pcplists of the indicated processor and zone.
9f8f2172
CL
2388 *
2389 * The processor must either be the current processor and the
2390 * thread pinned to the current processor or a processor that
2391 * is not online.
2392 */
93481ff0 2393static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1da177e4 2394{
c54ad30c 2395 unsigned long flags;
93481ff0
VB
2396 struct per_cpu_pageset *pset;
2397 struct per_cpu_pages *pcp;
1da177e4 2398
93481ff0
VB
2399 local_irq_save(flags);
2400 pset = per_cpu_ptr(zone->pageset, cpu);
1da177e4 2401
93481ff0
VB
2402 pcp = &pset->pcp;
2403 if (pcp->count) {
2404 free_pcppages_bulk(zone, pcp->count, pcp);
2405 pcp->count = 0;
2406 }
2407 local_irq_restore(flags);
2408}
3dfa5721 2409
93481ff0
VB
2410/*
2411 * Drain pcplists of all zones on the indicated processor.
2412 *
2413 * The processor must either be the current processor and the
2414 * thread pinned to the current processor or a processor that
2415 * is not online.
2416 */
2417static void drain_pages(unsigned int cpu)
2418{
2419 struct zone *zone;
2420
2421 for_each_populated_zone(zone) {
2422 drain_pages_zone(cpu, zone);
1da177e4
LT
2423 }
2424}
1da177e4 2425
9f8f2172
CL
2426/*
2427 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
93481ff0
VB
2428 *
2429 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2430 * the single zone's pages.
9f8f2172 2431 */
93481ff0 2432void drain_local_pages(struct zone *zone)
9f8f2172 2433{
93481ff0
VB
2434 int cpu = smp_processor_id();
2435
2436 if (zone)
2437 drain_pages_zone(cpu, zone);
2438 else
2439 drain_pages(cpu);
9f8f2172
CL
2440}
2441
0ccce3b9
MG
2442static void drain_local_pages_wq(struct work_struct *work)
2443{
a459eeb7
MH
2444 /*
2445 * drain_all_pages doesn't use proper cpu hotplug protection so
2446 * we can race with cpu offline when the WQ can move this from
2447 * a cpu pinned worker to an unbound one. We can operate on a different
2448 * cpu which is allright but we also have to make sure to not move to
2449 * a different one.
2450 */
2451 preempt_disable();
0ccce3b9 2452 drain_local_pages(NULL);
a459eeb7 2453 preempt_enable();
0ccce3b9
MG
2454}
2455
9f8f2172 2456/*
74046494
GBY
2457 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2458 *
93481ff0
VB
2459 * When zone parameter is non-NULL, spill just the single zone's pages.
2460 *
0ccce3b9 2461 * Note that this can be extremely slow as the draining happens in a workqueue.
9f8f2172 2462 */
93481ff0 2463void drain_all_pages(struct zone *zone)
9f8f2172 2464{
74046494 2465 int cpu;
74046494
GBY
2466
2467 /*
2468 * Allocate in the BSS so we wont require allocation in
2469 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2470 */
2471 static cpumask_t cpus_with_pcps;
2472
ce612879
MH
2473 /*
2474 * Make sure nobody triggers this path before mm_percpu_wq is fully
2475 * initialized.
2476 */
2477 if (WARN_ON_ONCE(!mm_percpu_wq))
2478 return;
2479
0ccce3b9
MG
2480 /* Workqueues cannot recurse */
2481 if (current->flags & PF_WQ_WORKER)
2482 return;
2483
bd233f53
MG
2484 /*
2485 * Do not drain if one is already in progress unless it's specific to
2486 * a zone. Such callers are primarily CMA and memory hotplug and need
2487 * the drain to be complete when the call returns.
2488 */
2489 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2490 if (!zone)
2491 return;
2492 mutex_lock(&pcpu_drain_mutex);
2493 }
0ccce3b9 2494
74046494
GBY
2495 /*
2496 * We don't care about racing with CPU hotplug event
2497 * as offline notification will cause the notified
2498 * cpu to drain that CPU pcps and on_each_cpu_mask
2499 * disables preemption as part of its processing
2500 */
2501 for_each_online_cpu(cpu) {
93481ff0
VB
2502 struct per_cpu_pageset *pcp;
2503 struct zone *z;
74046494 2504 bool has_pcps = false;
93481ff0
VB
2505
2506 if (zone) {
74046494 2507 pcp = per_cpu_ptr(zone->pageset, cpu);
93481ff0 2508 if (pcp->pcp.count)
74046494 2509 has_pcps = true;
93481ff0
VB
2510 } else {
2511 for_each_populated_zone(z) {
2512 pcp = per_cpu_ptr(z->pageset, cpu);
2513 if (pcp->pcp.count) {
2514 has_pcps = true;
2515 break;
2516 }
74046494
GBY
2517 }
2518 }
93481ff0 2519
74046494
GBY
2520 if (has_pcps)
2521 cpumask_set_cpu(cpu, &cpus_with_pcps);
2522 else
2523 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2524 }
0ccce3b9 2525
bd233f53
MG
2526 for_each_cpu(cpu, &cpus_with_pcps) {
2527 struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu);
2528 INIT_WORK(work, drain_local_pages_wq);
ce612879 2529 queue_work_on(cpu, mm_percpu_wq, work);
0ccce3b9 2530 }
bd233f53
MG
2531 for_each_cpu(cpu, &cpus_with_pcps)
2532 flush_work(per_cpu_ptr(&pcpu_drain, cpu));
2533
2534 mutex_unlock(&pcpu_drain_mutex);
9f8f2172
CL
2535}
2536
296699de 2537#ifdef CONFIG_HIBERNATION
1da177e4 2538
556b969a
CY
2539/*
2540 * Touch the watchdog for every WD_PAGE_COUNT pages.
2541 */
2542#define WD_PAGE_COUNT (128*1024)
2543
1da177e4
LT
2544void mark_free_pages(struct zone *zone)
2545{
556b969a 2546 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
f623f0db 2547 unsigned long flags;
7aeb09f9 2548 unsigned int order, t;
86760a2c 2549 struct page *page;
1da177e4 2550
8080fc03 2551 if (zone_is_empty(zone))
1da177e4
LT
2552 return;
2553
2554 spin_lock_irqsave(&zone->lock, flags);
f623f0db 2555
108bcc96 2556 max_zone_pfn = zone_end_pfn(zone);
f623f0db
RW
2557 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2558 if (pfn_valid(pfn)) {
86760a2c 2559 page = pfn_to_page(pfn);
ba6b0979 2560
556b969a
CY
2561 if (!--page_count) {
2562 touch_nmi_watchdog();
2563 page_count = WD_PAGE_COUNT;
2564 }
2565
ba6b0979
JK
2566 if (page_zone(page) != zone)
2567 continue;
2568
7be98234
RW
2569 if (!swsusp_page_is_forbidden(page))
2570 swsusp_unset_page_free(page);
f623f0db 2571 }
1da177e4 2572
b2a0ac88 2573 for_each_migratetype_order(order, t) {
86760a2c
GT
2574 list_for_each_entry(page,
2575 &zone->free_area[order].free_list[t], lru) {
f623f0db 2576 unsigned long i;
1da177e4 2577
86760a2c 2578 pfn = page_to_pfn(page);
556b969a
CY
2579 for (i = 0; i < (1UL << order); i++) {
2580 if (!--page_count) {
2581 touch_nmi_watchdog();
2582 page_count = WD_PAGE_COUNT;
2583 }
7be98234 2584 swsusp_set_page_free(pfn_to_page(pfn + i));
556b969a 2585 }
f623f0db 2586 }
b2a0ac88 2587 }
1da177e4
LT
2588 spin_unlock_irqrestore(&zone->lock, flags);
2589}
e2c55dc8 2590#endif /* CONFIG_PM */
1da177e4 2591
1da177e4
LT
2592/*
2593 * Free a 0-order page
b745bc85 2594 * cold == true ? free a cold page : free a hot page
1da177e4 2595 */
b745bc85 2596void free_hot_cold_page(struct page *page, bool cold)
1da177e4
LT
2597{
2598 struct zone *zone = page_zone(page);
2599 struct per_cpu_pages *pcp;
d34b0733 2600 unsigned long flags;
dc4b0caf 2601 unsigned long pfn = page_to_pfn(page);
5f8dcc21 2602 int migratetype;
1da177e4 2603
4db7548c 2604 if (!free_pcp_prepare(page))
689bcebf
HD
2605 return;
2606
dc4b0caf 2607 migratetype = get_pfnblock_migratetype(page, pfn);
bb14c2c7 2608 set_pcppage_migratetype(page, migratetype);
d34b0733
MG
2609 local_irq_save(flags);
2610 __count_vm_event(PGFREE);
da456f14 2611
5f8dcc21
MG
2612 /*
2613 * We only track unmovable, reclaimable and movable on pcp lists.
2614 * Free ISOLATE pages back to the allocator because they are being
a6ffdc07 2615 * offlined but treat HIGHATOMIC as movable pages so we can get those
5f8dcc21
MG
2616 * areas back if necessary. Otherwise, we may have to free
2617 * excessively into the page allocator
2618 */
2619 if (migratetype >= MIGRATE_PCPTYPES) {
194159fb 2620 if (unlikely(is_migrate_isolate(migratetype))) {
dc4b0caf 2621 free_one_page(zone, page, pfn, 0, migratetype);
5f8dcc21
MG
2622 goto out;
2623 }
2624 migratetype = MIGRATE_MOVABLE;
2625 }
2626
99dcc3e5 2627 pcp = &this_cpu_ptr(zone->pageset)->pcp;
b745bc85 2628 if (!cold)
5f8dcc21 2629 list_add(&page->lru, &pcp->lists[migratetype]);
b745bc85
MG
2630 else
2631 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1da177e4 2632 pcp->count++;
48db57f8 2633 if (pcp->count >= pcp->high) {
4db0c3c2 2634 unsigned long batch = READ_ONCE(pcp->batch);
998d39cb
CS
2635 free_pcppages_bulk(zone, batch, pcp);
2636 pcp->count -= batch;
48db57f8 2637 }
5f8dcc21
MG
2638
2639out:
d34b0733 2640 local_irq_restore(flags);
1da177e4
LT
2641}
2642
cc59850e
KK
2643/*
2644 * Free a list of 0-order pages
2645 */
b745bc85 2646void free_hot_cold_page_list(struct list_head *list, bool cold)
cc59850e
KK
2647{
2648 struct page *page, *next;
2649
2650 list_for_each_entry_safe(page, next, list, lru) {
b413d48a 2651 trace_mm_page_free_batched(page, cold);
cc59850e
KK
2652 free_hot_cold_page(page, cold);
2653 }
2654}
2655
8dfcc9ba
NP
2656/*
2657 * split_page takes a non-compound higher-order page, and splits it into
2658 * n (1<<order) sub-pages: page[0..n]
2659 * Each sub-page must be freed individually.
2660 *
2661 * Note: this is probably too low level an operation for use in drivers.
2662 * Please consult with lkml before using this in your driver.
2663 */
2664void split_page(struct page *page, unsigned int order)
2665{
2666 int i;
2667
309381fe
SL
2668 VM_BUG_ON_PAGE(PageCompound(page), page);
2669 VM_BUG_ON_PAGE(!page_count(page), page);
b1eeab67
VN
2670
2671#ifdef CONFIG_KMEMCHECK
2672 /*
2673 * Split shadow pages too, because free(page[0]) would
2674 * otherwise free the whole shadow.
2675 */
2676 if (kmemcheck_page_is_tracked(page))
2677 split_page(virt_to_page(page[0].shadow), order);
2678#endif
2679
a9627bc5 2680 for (i = 1; i < (1 << order); i++)
7835e98b 2681 set_page_refcounted(page + i);
a9627bc5 2682 split_page_owner(page, order);
8dfcc9ba 2683}
5853ff23 2684EXPORT_SYMBOL_GPL(split_page);
8dfcc9ba 2685
3c605096 2686int __isolate_free_page(struct page *page, unsigned int order)
748446bb 2687{
748446bb
MG
2688 unsigned long watermark;
2689 struct zone *zone;
2139cbe6 2690 int mt;
748446bb
MG
2691
2692 BUG_ON(!PageBuddy(page));
2693
2694 zone = page_zone(page);
2e30abd1 2695 mt = get_pageblock_migratetype(page);
748446bb 2696
194159fb 2697 if (!is_migrate_isolate(mt)) {
8348faf9
VB
2698 /*
2699 * Obey watermarks as if the page was being allocated. We can
2700 * emulate a high-order watermark check with a raised order-0
2701 * watermark, because we already know our high-order page
2702 * exists.
2703 */
2704 watermark = min_wmark_pages(zone) + (1UL << order);
984fdba6 2705 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
2e30abd1
MS
2706 return 0;
2707
8fb74b9f 2708 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2e30abd1 2709 }
748446bb
MG
2710
2711 /* Remove page from free list */
2712 list_del(&page->lru);
2713 zone->free_area[order].nr_free--;
2714 rmv_page_order(page);
2139cbe6 2715
400bc7fd 2716 /*
2717 * Set the pageblock if the isolated page is at least half of a
2718 * pageblock
2719 */
748446bb
MG
2720 if (order >= pageblock_order - 1) {
2721 struct page *endpage = page + (1 << order) - 1;
47118af0
MN
2722 for (; page < endpage; page += pageblock_nr_pages) {
2723 int mt = get_pageblock_migratetype(page);
88ed365e 2724 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
a6ffdc07 2725 && !is_migrate_highatomic(mt))
47118af0
MN
2726 set_pageblock_migratetype(page,
2727 MIGRATE_MOVABLE);
2728 }
748446bb
MG
2729 }
2730
f3a14ced 2731
8fb74b9f 2732 return 1UL << order;
1fb3f8ca
MG
2733}
2734
060e7417
MG
2735/*
2736 * Update NUMA hit/miss statistics
2737 *
2738 * Must be called with interrupts disabled.
060e7417 2739 */
41b6167e 2740static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
060e7417
MG
2741{
2742#ifdef CONFIG_NUMA
060e7417
MG
2743 enum zone_stat_item local_stat = NUMA_LOCAL;
2744
2df26639 2745 if (z->node != numa_node_id())
060e7417 2746 local_stat = NUMA_OTHER;
060e7417 2747
2df26639 2748 if (z->node == preferred_zone->node)
060e7417 2749 __inc_zone_state(z, NUMA_HIT);
2df26639 2750 else {
060e7417
MG
2751 __inc_zone_state(z, NUMA_MISS);
2752 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2753 }
2df26639 2754 __inc_zone_state(z, local_stat);
060e7417
MG
2755#endif
2756}
2757
066b2393
MG
2758/* Remove page from the per-cpu list, caller must protect the list */
2759static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
2760 bool cold, struct per_cpu_pages *pcp,
2761 struct list_head *list)
2762{
2763 struct page *page;
2764
2765 do {
2766 if (list_empty(list)) {
2767 pcp->count += rmqueue_bulk(zone, 0,
2768 pcp->batch, list,
2769 migratetype, cold);
2770 if (unlikely(list_empty(list)))
2771 return NULL;
2772 }
2773
2774 if (cold)
2775 page = list_last_entry(list, struct page, lru);
2776 else
2777 page = list_first_entry(list, struct page, lru);
2778
2779 list_del(&page->lru);
2780 pcp->count--;
2781 } while (check_new_pcp(page));
2782
2783 return page;
2784}
2785
2786/* Lock and remove page from the per-cpu list */
2787static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2788 struct zone *zone, unsigned int order,
2789 gfp_t gfp_flags, int migratetype)
2790{
2791 struct per_cpu_pages *pcp;
2792 struct list_head *list;
2793 bool cold = ((gfp_flags & __GFP_COLD) != 0);
2794 struct page *page;
d34b0733 2795 unsigned long flags;
066b2393 2796
d34b0733 2797 local_irq_save(flags);
066b2393
MG
2798 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2799 list = &pcp->lists[migratetype];
2800 page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list);
2801 if (page) {
2802 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2803 zone_statistics(preferred_zone, zone);
2804 }
d34b0733 2805 local_irq_restore(flags);
066b2393
MG
2806 return page;
2807}
2808
1da177e4 2809/*
75379191 2810 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
1da177e4 2811 */
0a15c3e9 2812static inline
066b2393 2813struct page *rmqueue(struct zone *preferred_zone,
7aeb09f9 2814 struct zone *zone, unsigned int order,
c603844b
MG
2815 gfp_t gfp_flags, unsigned int alloc_flags,
2816 int migratetype)
1da177e4
LT
2817{
2818 unsigned long flags;
689bcebf 2819 struct page *page;
1da177e4 2820
d34b0733 2821 if (likely(order == 0)) {
066b2393
MG
2822 page = rmqueue_pcplist(preferred_zone, zone, order,
2823 gfp_flags, migratetype);
2824 goto out;
2825 }
83b9355b 2826
066b2393
MG
2827 /*
2828 * We most definitely don't want callers attempting to
2829 * allocate greater than order-1 page units with __GFP_NOFAIL.
2830 */
2831 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2832 spin_lock_irqsave(&zone->lock, flags);
0aaa29a5 2833
066b2393
MG
2834 do {
2835 page = NULL;
2836 if (alloc_flags & ALLOC_HARDER) {
2837 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2838 if (page)
2839 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2840 }
a74609fa 2841 if (!page)
066b2393
MG
2842 page = __rmqueue(zone, order, migratetype);
2843 } while (page && check_new_pages(page, order));
2844 spin_unlock(&zone->lock);
2845 if (!page)
2846 goto failed;
2847 __mod_zone_freepage_state(zone, -(1 << order),
2848 get_pcppage_migratetype(page));
1da177e4 2849
16709d1d 2850 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
41b6167e 2851 zone_statistics(preferred_zone, zone);
a74609fa 2852 local_irq_restore(flags);
1da177e4 2853
066b2393
MG
2854out:
2855 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
1da177e4 2856 return page;
a74609fa
NP
2857
2858failed:
2859 local_irq_restore(flags);
a74609fa 2860 return NULL;
1da177e4
LT
2861}
2862
933e312e
AM
2863#ifdef CONFIG_FAIL_PAGE_ALLOC
2864
b2588c4b 2865static struct {
933e312e
AM
2866 struct fault_attr attr;
2867
621a5f7a 2868 bool ignore_gfp_highmem;
71baba4b 2869 bool ignore_gfp_reclaim;
54114994 2870 u32 min_order;
933e312e
AM
2871} fail_page_alloc = {
2872 .attr = FAULT_ATTR_INITIALIZER,
71baba4b 2873 .ignore_gfp_reclaim = true,
621a5f7a 2874 .ignore_gfp_highmem = true,
54114994 2875 .min_order = 1,
933e312e
AM
2876};
2877
2878static int __init setup_fail_page_alloc(char *str)
2879{
2880 return setup_fault_attr(&fail_page_alloc.attr, str);
2881}
2882__setup("fail_page_alloc=", setup_fail_page_alloc);
2883
deaf386e 2884static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
933e312e 2885{
54114994 2886 if (order < fail_page_alloc.min_order)
deaf386e 2887 return false;
933e312e 2888 if (gfp_mask & __GFP_NOFAIL)
deaf386e 2889 return false;
933e312e 2890 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
deaf386e 2891 return false;
71baba4b
MG
2892 if (fail_page_alloc.ignore_gfp_reclaim &&
2893 (gfp_mask & __GFP_DIRECT_RECLAIM))
deaf386e 2894 return false;
933e312e
AM
2895
2896 return should_fail(&fail_page_alloc.attr, 1 << order);
2897}
2898
2899#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2900
2901static int __init fail_page_alloc_debugfs(void)
2902{
f4ae40a6 2903 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
933e312e 2904 struct dentry *dir;
933e312e 2905
dd48c085
AM
2906 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2907 &fail_page_alloc.attr);
2908 if (IS_ERR(dir))
2909 return PTR_ERR(dir);
933e312e 2910
b2588c4b 2911 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
71baba4b 2912 &fail_page_alloc.ignore_gfp_reclaim))
b2588c4b
AM
2913 goto fail;
2914 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2915 &fail_page_alloc.ignore_gfp_highmem))
2916 goto fail;
2917 if (!debugfs_create_u32("min-order", mode, dir,
2918 &fail_page_alloc.min_order))
2919 goto fail;
2920
2921 return 0;
2922fail:
dd48c085 2923 debugfs_remove_recursive(dir);
933e312e 2924
b2588c4b 2925 return -ENOMEM;
933e312e
AM
2926}
2927
2928late_initcall(fail_page_alloc_debugfs);
2929
2930#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2931
2932#else /* CONFIG_FAIL_PAGE_ALLOC */
2933
deaf386e 2934static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
933e312e 2935{
deaf386e 2936 return false;
933e312e
AM
2937}
2938
2939#endif /* CONFIG_FAIL_PAGE_ALLOC */
2940
1da177e4 2941/*
97a16fc8
MG
2942 * Return true if free base pages are above 'mark'. For high-order checks it
2943 * will return true of the order-0 watermark is reached and there is at least
2944 * one free page of a suitable size. Checking now avoids taking the zone lock
2945 * to check in the allocation paths if no pages are free.
1da177e4 2946 */
86a294a8
MH
2947bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2948 int classzone_idx, unsigned int alloc_flags,
2949 long free_pages)
1da177e4 2950{
d23ad423 2951 long min = mark;
1da177e4 2952 int o;
c603844b 2953 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
1da177e4 2954
0aaa29a5 2955 /* free_pages may go negative - that's OK */
df0a6daa 2956 free_pages -= (1 << order) - 1;
0aaa29a5 2957
7fb1d9fc 2958 if (alloc_flags & ALLOC_HIGH)
1da177e4 2959 min -= min / 2;
0aaa29a5
MG
2960
2961 /*
2962 * If the caller does not have rights to ALLOC_HARDER then subtract
2963 * the high-atomic reserves. This will over-estimate the size of the
2964 * atomic reserve but it avoids a search.
2965 */
97a16fc8 2966 if (likely(!alloc_harder))
0aaa29a5
MG
2967 free_pages -= z->nr_reserved_highatomic;
2968 else
1da177e4 2969 min -= min / 4;
e2b19197 2970
d95ea5d1
BZ
2971#ifdef CONFIG_CMA
2972 /* If allocation can't use CMA areas don't use free CMA pages */
2973 if (!(alloc_flags & ALLOC_CMA))
97a16fc8 2974 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
d95ea5d1 2975#endif
026b0814 2976
97a16fc8
MG
2977 /*
2978 * Check watermarks for an order-0 allocation request. If these
2979 * are not met, then a high-order request also cannot go ahead
2980 * even if a suitable page happened to be free.
2981 */
2982 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
88f5acf8 2983 return false;
1da177e4 2984
97a16fc8
MG
2985 /* If this is an order-0 request then the watermark is fine */
2986 if (!order)
2987 return true;
2988
2989 /* For a high-order request, check at least one suitable page is free */
2990 for (o = order; o < MAX_ORDER; o++) {
2991 struct free_area *area = &z->free_area[o];
2992 int mt;
2993
2994 if (!area->nr_free)
2995 continue;
2996
2997 if (alloc_harder)
2998 return true;
1da177e4 2999
97a16fc8
MG
3000 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3001 if (!list_empty(&area->free_list[mt]))
3002 return true;
3003 }
3004
3005#ifdef CONFIG_CMA
3006 if ((alloc_flags & ALLOC_CMA) &&
3007 !list_empty(&area->free_list[MIGRATE_CMA])) {
3008 return true;
3009 }
3010#endif
1da177e4 3011 }
97a16fc8 3012 return false;
88f5acf8
MG
3013}
3014
7aeb09f9 3015bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
c603844b 3016 int classzone_idx, unsigned int alloc_flags)
88f5acf8
MG
3017{
3018 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3019 zone_page_state(z, NR_FREE_PAGES));
3020}
3021
48ee5f36
MG
3022static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3023 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3024{
3025 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3026 long cma_pages = 0;
3027
3028#ifdef CONFIG_CMA
3029 /* If allocation can't use CMA areas don't use free CMA pages */
3030 if (!(alloc_flags & ALLOC_CMA))
3031 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3032#endif
3033
3034 /*
3035 * Fast check for order-0 only. If this fails then the reserves
3036 * need to be calculated. There is a corner case where the check
3037 * passes but only the high-order atomic reserve are free. If
3038 * the caller is !atomic then it'll uselessly search the free
3039 * list. That corner case is then slower but it is harmless.
3040 */
3041 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3042 return true;
3043
3044 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3045 free_pages);
3046}
3047
7aeb09f9 3048bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
e2b19197 3049 unsigned long mark, int classzone_idx)
88f5acf8
MG
3050{
3051 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3052
3053 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3054 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3055
e2b19197 3056 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
88f5acf8 3057 free_pages);
1da177e4
LT
3058}
3059
9276b1bc 3060#ifdef CONFIG_NUMA
957f822a
DR
3061static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3062{
e02dc017 3063 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
5f7a75ac 3064 RECLAIM_DISTANCE;
957f822a 3065}
9276b1bc 3066#else /* CONFIG_NUMA */
957f822a
DR
3067static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3068{
3069 return true;
3070}
9276b1bc
PJ
3071#endif /* CONFIG_NUMA */
3072
7fb1d9fc 3073/*
0798e519 3074 * get_page_from_freelist goes through the zonelist trying to allocate
7fb1d9fc
RS
3075 * a page.
3076 */
3077static struct page *
a9263751
VB
3078get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3079 const struct alloc_context *ac)
753ee728 3080{
c33d6c06 3081 struct zoneref *z = ac->preferred_zoneref;
5117f45d 3082 struct zone *zone;
3b8c0be4
MG
3083 struct pglist_data *last_pgdat_dirty_limit = NULL;
3084
7fb1d9fc 3085 /*
9276b1bc 3086 * Scan zonelist, looking for a zone with enough free.
344736f2 3087 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
7fb1d9fc 3088 */
c33d6c06 3089 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
a9263751 3090 ac->nodemask) {
be06af00 3091 struct page *page;
e085dbc5
JW
3092 unsigned long mark;
3093
664eedde
MG
3094 if (cpusets_enabled() &&
3095 (alloc_flags & ALLOC_CPUSET) &&
002f2906 3096 !__cpuset_zone_allowed(zone, gfp_mask))
cd38b115 3097 continue;
a756cf59
JW
3098 /*
3099 * When allocating a page cache page for writing, we
281e3726
MG
3100 * want to get it from a node that is within its dirty
3101 * limit, such that no single node holds more than its
a756cf59 3102 * proportional share of globally allowed dirty pages.
281e3726 3103 * The dirty limits take into account the node's
a756cf59
JW
3104 * lowmem reserves and high watermark so that kswapd
3105 * should be able to balance it without having to
3106 * write pages from its LRU list.
3107 *
a756cf59 3108 * XXX: For now, allow allocations to potentially
281e3726 3109 * exceed the per-node dirty limit in the slowpath
c9ab0c4f 3110 * (spread_dirty_pages unset) before going into reclaim,
a756cf59 3111 * which is important when on a NUMA setup the allowed
281e3726 3112 * nodes are together not big enough to reach the
a756cf59 3113 * global limit. The proper fix for these situations
281e3726 3114 * will require awareness of nodes in the
a756cf59
JW
3115 * dirty-throttling and the flusher threads.
3116 */
3b8c0be4
MG
3117 if (ac->spread_dirty_pages) {
3118 if (last_pgdat_dirty_limit == zone->zone_pgdat)
3119 continue;
3120
3121 if (!node_dirty_ok(zone->zone_pgdat)) {
3122 last_pgdat_dirty_limit = zone->zone_pgdat;
3123 continue;
3124 }
3125 }
7fb1d9fc 3126
e085dbc5 3127 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
48ee5f36 3128 if (!zone_watermark_fast(zone, order, mark,
93ea9964 3129 ac_classzone_idx(ac), alloc_flags)) {
fa5e084e
MG
3130 int ret;
3131
5dab2911
MG
3132 /* Checked here to keep the fast path fast */
3133 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3134 if (alloc_flags & ALLOC_NO_WATERMARKS)
3135 goto try_this_zone;
3136
a5f5f91d 3137 if (node_reclaim_mode == 0 ||
c33d6c06 3138 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
cd38b115
MG
3139 continue;
3140
a5f5f91d 3141 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
fa5e084e 3142 switch (ret) {
a5f5f91d 3143 case NODE_RECLAIM_NOSCAN:
fa5e084e 3144 /* did not scan */
cd38b115 3145 continue;
a5f5f91d 3146 case NODE_RECLAIM_FULL:
fa5e084e 3147 /* scanned but unreclaimable */
cd38b115 3148 continue;
fa5e084e
MG
3149 default:
3150 /* did we reclaim enough */
fed2719e 3151 if (zone_watermark_ok(zone, order, mark,
93ea9964 3152 ac_classzone_idx(ac), alloc_flags))
fed2719e
MG
3153 goto try_this_zone;
3154
fed2719e 3155 continue;
0798e519 3156 }
7fb1d9fc
RS
3157 }
3158
fa5e084e 3159try_this_zone:
066b2393 3160 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
0aaa29a5 3161 gfp_mask, alloc_flags, ac->migratetype);
75379191 3162 if (page) {
479f854a 3163 prep_new_page(page, order, gfp_mask, alloc_flags);
0aaa29a5
MG
3164
3165 /*
3166 * If this is a high-order atomic allocation then check
3167 * if the pageblock should be reserved for the future
3168 */
3169 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3170 reserve_highatomic_pageblock(page, zone, order);
3171
75379191
VB
3172 return page;
3173 }
54a6eb5c 3174 }
9276b1bc 3175
4ffeaf35 3176 return NULL;
753ee728
MH
3177}
3178
29423e77
DR
3179/*
3180 * Large machines with many possible nodes should not always dump per-node
3181 * meminfo in irq context.
3182 */
3183static inline bool should_suppress_show_mem(void)
3184{
3185 bool ret = false;
3186
3187#if NODES_SHIFT > 8
3188 ret = in_interrupt();
3189#endif
3190 return ret;
3191}
3192
9af744d7 3193static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
a238ab5b 3194{
a238ab5b 3195 unsigned int filter = SHOW_MEM_FILTER_NODES;
aa187507 3196 static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
a238ab5b 3197
aa187507 3198 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs))
a238ab5b
DH
3199 return;
3200
3201 /*
3202 * This documents exceptions given to allocations in certain
3203 * contexts that are allowed to allocate outside current's set
3204 * of allowed nodes.
3205 */
3206 if (!(gfp_mask & __GFP_NOMEMALLOC))
3207 if (test_thread_flag(TIF_MEMDIE) ||
3208 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3209 filter &= ~SHOW_MEM_FILTER_NODES;
d0164adc 3210 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
a238ab5b
DH
3211 filter &= ~SHOW_MEM_FILTER_NODES;
3212
9af744d7 3213 show_mem(filter, nodemask);
aa187507
MH
3214}
3215
a8e99259 3216void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
aa187507
MH
3217{
3218 struct va_format vaf;
3219 va_list args;
3220 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3221 DEFAULT_RATELIMIT_BURST);
3222
0f7896f1 3223 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
aa187507
MH
3224 return;
3225
7877cdcc 3226 pr_warn("%s: ", current->comm);
3ee9a4f0 3227
7877cdcc
MH
3228 va_start(args, fmt);
3229 vaf.fmt = fmt;
3230 vaf.va = &args;
3231 pr_cont("%pV", &vaf);
3232 va_end(args);
3ee9a4f0 3233
685dbf6f
DR
3234 pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask);
3235 if (nodemask)
3236 pr_cont("%*pbl\n", nodemask_pr_args(nodemask));
3237 else
3238 pr_cont("(null)\n");
3239
a8e99259 3240 cpuset_print_current_mems_allowed();
3ee9a4f0 3241
a238ab5b 3242 dump_stack();
685dbf6f 3243 warn_alloc_show_mem(gfp_mask, nodemask);
a238ab5b
DH
3244}
3245
6c18ba7a
MH
3246static inline struct page *
3247__alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3248 unsigned int alloc_flags,
3249 const struct alloc_context *ac)
3250{
3251 struct page *page;
3252
3253 page = get_page_from_freelist(gfp_mask, order,
3254 alloc_flags|ALLOC_CPUSET, ac);
3255 /*
3256 * fallback to ignore cpuset restriction if our nodes
3257 * are depleted
3258 */
3259 if (!page)
3260 page = get_page_from_freelist(gfp_mask, order,
3261 alloc_flags, ac);
3262
3263 return page;
3264}
3265
11e33f6a
MG
3266static inline struct page *
3267__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
a9263751 3268 const struct alloc_context *ac, unsigned long *did_some_progress)
11e33f6a 3269{
6e0fc46d
DR
3270 struct oom_control oc = {
3271 .zonelist = ac->zonelist,
3272 .nodemask = ac->nodemask,
2a966b77 3273 .memcg = NULL,
6e0fc46d
DR
3274 .gfp_mask = gfp_mask,
3275 .order = order,
6e0fc46d 3276 };
11e33f6a
MG
3277 struct page *page;
3278
9879de73
JW
3279 *did_some_progress = 0;
3280
9879de73 3281 /*
dc56401f
JW
3282 * Acquire the oom lock. If that fails, somebody else is
3283 * making progress for us.
9879de73 3284 */
dc56401f 3285 if (!mutex_trylock(&oom_lock)) {
9879de73 3286 *did_some_progress = 1;
11e33f6a 3287 schedule_timeout_uninterruptible(1);
1da177e4
LT
3288 return NULL;
3289 }
6b1de916 3290
11e33f6a
MG
3291 /*
3292 * Go through the zonelist yet one more time, keep very high watermark
3293 * here, this is only to catch a parallel oom killing, we must fail if
e746bf73
TH
3294 * we're still under heavy pressure. But make sure that this reclaim
3295 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3296 * allocation which will never fail due to oom_lock already held.
11e33f6a 3297 */
e746bf73
TH
3298 page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3299 ~__GFP_DIRECT_RECLAIM, order,
3300 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
7fb1d9fc 3301 if (page)
11e33f6a
MG
3302 goto out;
3303
06ad276a
MH
3304 /* Coredumps can quickly deplete all memory reserves */
3305 if (current->flags & PF_DUMPCORE)
3306 goto out;
3307 /* The OOM killer will not help higher order allocs */
3308 if (order > PAGE_ALLOC_COSTLY_ORDER)
3309 goto out;
dcda9b04
MH
3310 /*
3311 * We have already exhausted all our reclaim opportunities without any
3312 * success so it is time to admit defeat. We will skip the OOM killer
3313 * because it is very likely that the caller has a more reasonable
3314 * fallback than shooting a random task.
3315 */
3316 if (gfp_mask & __GFP_RETRY_MAYFAIL)
3317 goto out;
06ad276a
MH
3318 /* The OOM killer does not needlessly kill tasks for lowmem */
3319 if (ac->high_zoneidx < ZONE_NORMAL)
3320 goto out;
3321 if (pm_suspended_storage())
3322 goto out;
3323 /*
3324 * XXX: GFP_NOFS allocations should rather fail than rely on
3325 * other request to make a forward progress.
3326 * We are in an unfortunate situation where out_of_memory cannot
3327 * do much for this context but let's try it to at least get
3328 * access to memory reserved if the current task is killed (see
3329 * out_of_memory). Once filesystems are ready to handle allocation
3330 * failures more gracefully we should just bail out here.
3331 */
3332
3333 /* The OOM killer may not free memory on a specific node */
3334 if (gfp_mask & __GFP_THISNODE)
3335 goto out;
3da88fb3 3336
11e33f6a 3337 /* Exhausted what can be done so it's blamo time */
5020e285 3338 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
c32b3cbe 3339 *did_some_progress = 1;
5020e285 3340
6c18ba7a
MH
3341 /*
3342 * Help non-failing allocations by giving them access to memory
3343 * reserves
3344 */
3345 if (gfp_mask & __GFP_NOFAIL)
3346 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
5020e285 3347 ALLOC_NO_WATERMARKS, ac);
5020e285 3348 }
11e33f6a 3349out:
dc56401f 3350 mutex_unlock(&oom_lock);
11e33f6a
MG
3351 return page;
3352}
3353
33c2d214
MH
3354/*
3355 * Maximum number of compaction retries wit a progress before OOM
3356 * killer is consider as the only way to move forward.
3357 */
3358#define MAX_COMPACT_RETRIES 16
3359
56de7263
MG
3360#ifdef CONFIG_COMPACTION
3361/* Try memory compaction for high-order allocations before reclaim */
3362static struct page *
3363__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
c603844b 3364 unsigned int alloc_flags, const struct alloc_context *ac,
a5508cd8 3365 enum compact_priority prio, enum compact_result *compact_result)
56de7263 3366{
98dd3b48 3367 struct page *page;
499118e9 3368 unsigned int noreclaim_flag;
53853e2d
VB
3369
3370 if (!order)
66199712 3371 return NULL;
66199712 3372
499118e9 3373 noreclaim_flag = memalloc_noreclaim_save();
c5d01d0d 3374 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
c3486f53 3375 prio);
499118e9 3376 memalloc_noreclaim_restore(noreclaim_flag);
56de7263 3377
c5d01d0d 3378 if (*compact_result <= COMPACT_INACTIVE)
98dd3b48 3379 return NULL;
53853e2d 3380
98dd3b48
VB
3381 /*
3382 * At least in one zone compaction wasn't deferred or skipped, so let's
3383 * count a compaction stall
3384 */
3385 count_vm_event(COMPACTSTALL);
8fb74b9f 3386
31a6c190 3387 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
53853e2d 3388
98dd3b48
VB
3389 if (page) {
3390 struct zone *zone = page_zone(page);
53853e2d 3391
98dd3b48
VB
3392 zone->compact_blockskip_flush = false;
3393 compaction_defer_reset(zone, order, true);
3394 count_vm_event(COMPACTSUCCESS);
3395 return page;
3396 }
56de7263 3397
98dd3b48
VB
3398 /*
3399 * It's bad if compaction run occurs and fails. The most likely reason
3400 * is that pages exist, but not enough to satisfy watermarks.
3401 */
3402 count_vm_event(COMPACTFAIL);
66199712 3403
98dd3b48 3404 cond_resched();
56de7263
MG
3405
3406 return NULL;
3407}
33c2d214 3408
3250845d
VB
3409static inline bool
3410should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3411 enum compact_result compact_result,
3412 enum compact_priority *compact_priority,
d9436498 3413 int *compaction_retries)
3250845d
VB
3414{
3415 int max_retries = MAX_COMPACT_RETRIES;
c2033b00 3416 int min_priority;
65190cff
MH
3417 bool ret = false;
3418 int retries = *compaction_retries;
3419 enum compact_priority priority = *compact_priority;
3250845d
VB
3420
3421 if (!order)
3422 return false;
3423
d9436498
VB
3424 if (compaction_made_progress(compact_result))
3425 (*compaction_retries)++;
3426
3250845d
VB
3427 /*
3428 * compaction considers all the zone as desperately out of memory
3429 * so it doesn't really make much sense to retry except when the
3430 * failure could be caused by insufficient priority
3431 */
d9436498
VB
3432 if (compaction_failed(compact_result))
3433 goto check_priority;
3250845d
VB
3434
3435 /*
3436 * make sure the compaction wasn't deferred or didn't bail out early
3437 * due to locks contention before we declare that we should give up.
3438 * But do not retry if the given zonelist is not suitable for
3439 * compaction.
3440 */
65190cff
MH
3441 if (compaction_withdrawn(compact_result)) {
3442 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3443 goto out;
3444 }
3250845d
VB
3445
3446 /*
dcda9b04 3447 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
3250845d
VB
3448 * costly ones because they are de facto nofail and invoke OOM
3449 * killer to move on while costly can fail and users are ready
3450 * to cope with that. 1/4 retries is rather arbitrary but we
3451 * would need much more detailed feedback from compaction to
3452 * make a better decision.
3453 */
3454 if (order > PAGE_ALLOC_COSTLY_ORDER)
3455 max_retries /= 4;
65190cff
MH
3456 if (*compaction_retries <= max_retries) {
3457 ret = true;
3458 goto out;
3459 }
3250845d 3460
d9436498
VB
3461 /*
3462 * Make sure there are attempts at the highest priority if we exhausted
3463 * all retries or failed at the lower priorities.
3464 */
3465check_priority:
c2033b00
VB
3466 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3467 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
65190cff 3468
c2033b00 3469 if (*compact_priority > min_priority) {
d9436498
VB
3470 (*compact_priority)--;
3471 *compaction_retries = 0;
65190cff 3472 ret = true;
d9436498 3473 }
65190cff
MH
3474out:
3475 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3476 return ret;
3250845d 3477}
56de7263
MG
3478#else
3479static inline struct page *
3480__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
c603844b 3481 unsigned int alloc_flags, const struct alloc_context *ac,
a5508cd8 3482 enum compact_priority prio, enum compact_result *compact_result)
56de7263 3483{
33c2d214 3484 *compact_result = COMPACT_SKIPPED;
56de7263
MG
3485 return NULL;
3486}
33c2d214
MH
3487
3488static inline bool
86a294a8
MH
3489should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3490 enum compact_result compact_result,
a5508cd8 3491 enum compact_priority *compact_priority,
d9436498 3492 int *compaction_retries)
33c2d214 3493{
31e49bfd
MH
3494 struct zone *zone;
3495 struct zoneref *z;
3496
3497 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3498 return false;
3499
3500 /*
3501 * There are setups with compaction disabled which would prefer to loop
3502 * inside the allocator rather than hit the oom killer prematurely.
3503 * Let's give them a good hope and keep retrying while the order-0
3504 * watermarks are OK.
3505 */
3506 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3507 ac->nodemask) {
3508 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3509 ac_classzone_idx(ac), alloc_flags))
3510 return true;
3511 }
33c2d214
MH
3512 return false;
3513}
3250845d 3514#endif /* CONFIG_COMPACTION */
56de7263 3515
bba90710
MS
3516/* Perform direct synchronous page reclaim */
3517static int
a9263751
VB
3518__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3519 const struct alloc_context *ac)
11e33f6a 3520{
11e33f6a 3521 struct reclaim_state reclaim_state;
bba90710 3522 int progress;
499118e9 3523 unsigned int noreclaim_flag;
11e33f6a
MG
3524
3525 cond_resched();
3526
3527 /* We now go into synchronous reclaim */
3528 cpuset_memory_pressure_bump();
499118e9 3529 noreclaim_flag = memalloc_noreclaim_save();
11e33f6a
MG
3530 lockdep_set_current_reclaim_state(gfp_mask);
3531 reclaim_state.reclaimed_slab = 0;
c06b1fca 3532 current->reclaim_state = &reclaim_state;
11e33f6a 3533
a9263751
VB
3534 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3535 ac->nodemask);
11e33f6a 3536
c06b1fca 3537 current->reclaim_state = NULL;
11e33f6a 3538 lockdep_clear_current_reclaim_state();
499118e9 3539 memalloc_noreclaim_restore(noreclaim_flag);
11e33f6a
MG
3540
3541 cond_resched();
3542
bba90710
MS
3543 return progress;
3544}
3545
3546/* The really slow allocator path where we enter direct reclaim */
3547static inline struct page *
3548__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
c603844b 3549 unsigned int alloc_flags, const struct alloc_context *ac,
a9263751 3550 unsigned long *did_some_progress)
bba90710
MS
3551{
3552 struct page *page = NULL;
3553 bool drained = false;
3554
a9263751 3555 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
9ee493ce
MG
3556 if (unlikely(!(*did_some_progress)))
3557 return NULL;
11e33f6a 3558
9ee493ce 3559retry:
31a6c190 3560 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
9ee493ce
MG
3561
3562 /*
3563 * If an allocation failed after direct reclaim, it could be because
0aaa29a5
MG
3564 * pages are pinned on the per-cpu lists or in high alloc reserves.
3565 * Shrink them them and try again
9ee493ce
MG
3566 */
3567 if (!page && !drained) {
29fac03b 3568 unreserve_highatomic_pageblock(ac, false);
93481ff0 3569 drain_all_pages(NULL);
9ee493ce
MG
3570 drained = true;
3571 goto retry;
3572 }
3573
11e33f6a
MG
3574 return page;
3575}
3576
a9263751 3577static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3a025760
JW
3578{
3579 struct zoneref *z;
3580 struct zone *zone;
e1a55637 3581 pg_data_t *last_pgdat = NULL;
3a025760 3582
a9263751 3583 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
e1a55637
MG
3584 ac->high_zoneidx, ac->nodemask) {
3585 if (last_pgdat != zone->zone_pgdat)
52e9f87a 3586 wakeup_kswapd(zone, order, ac->high_zoneidx);
e1a55637
MG
3587 last_pgdat = zone->zone_pgdat;
3588 }
3a025760
JW
3589}
3590
c603844b 3591static inline unsigned int
341ce06f
PZ
3592gfp_to_alloc_flags(gfp_t gfp_mask)
3593{
c603844b 3594 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1da177e4 3595
a56f57ff 3596 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
e6223a3b 3597 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
933e312e 3598
341ce06f
PZ
3599 /*
3600 * The caller may dip into page reserves a bit more if the caller
3601 * cannot run direct reclaim, or if the caller has realtime scheduling
3602 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
d0164adc 3603 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
341ce06f 3604 */
e6223a3b 3605 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
1da177e4 3606
d0164adc 3607 if (gfp_mask & __GFP_ATOMIC) {
5c3240d9 3608 /*
b104a35d
DR
3609 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3610 * if it can't schedule.
5c3240d9 3611 */
b104a35d 3612 if (!(gfp_mask & __GFP_NOMEMALLOC))
5c3240d9 3613 alloc_flags |= ALLOC_HARDER;
523b9458 3614 /*
b104a35d 3615 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
344736f2 3616 * comment for __cpuset_node_allowed().
523b9458 3617 */
341ce06f 3618 alloc_flags &= ~ALLOC_CPUSET;
c06b1fca 3619 } else if (unlikely(rt_task(current)) && !in_interrupt())
341ce06f
PZ
3620 alloc_flags |= ALLOC_HARDER;
3621
d95ea5d1 3622#ifdef CONFIG_CMA
43e7a34d 3623 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
d95ea5d1
BZ
3624 alloc_flags |= ALLOC_CMA;
3625#endif
341ce06f
PZ
3626 return alloc_flags;
3627}
3628
072bb0aa
MG
3629bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3630{
31a6c190
VB
3631 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3632 return false;
3633
3634 if (gfp_mask & __GFP_MEMALLOC)
3635 return true;
3636 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3637 return true;
3638 if (!in_interrupt() &&
3639 ((current->flags & PF_MEMALLOC) ||
3640 unlikely(test_thread_flag(TIF_MEMDIE))))
3641 return true;
3642
3643 return false;
072bb0aa
MG
3644}
3645
0a0337e0
MH
3646/*
3647 * Checks whether it makes sense to retry the reclaim to make a forward progress
3648 * for the given allocation request.
491d79ae
JW
3649 *
3650 * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3651 * without success, or when we couldn't even meet the watermark if we
3652 * reclaimed all remaining pages on the LRU lists.
0a0337e0
MH
3653 *
3654 * Returns true if a retry is viable or false to enter the oom path.
3655 */
3656static inline bool
3657should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3658 struct alloc_context *ac, int alloc_flags,
423b452e 3659 bool did_some_progress, int *no_progress_loops)
0a0337e0
MH
3660{
3661 struct zone *zone;
3662 struct zoneref *z;
3663
423b452e
VB
3664 /*
3665 * Costly allocations might have made a progress but this doesn't mean
3666 * their order will become available due to high fragmentation so
3667 * always increment the no progress counter for them
3668 */
3669 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3670 *no_progress_loops = 0;
3671 else
3672 (*no_progress_loops)++;
3673
0a0337e0
MH
3674 /*
3675 * Make sure we converge to OOM if we cannot make any progress
3676 * several times in the row.
3677 */
04c8716f
MK
3678 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
3679 /* Before OOM, exhaust highatomic_reserve */
29fac03b 3680 return unreserve_highatomic_pageblock(ac, true);
04c8716f 3681 }
0a0337e0 3682
bca67592
MG
3683 /*
3684 * Keep reclaiming pages while there is a chance this will lead
3685 * somewhere. If none of the target zones can satisfy our allocation
3686 * request even if all reclaimable pages are considered then we are
3687 * screwed and have to go OOM.
0a0337e0
MH
3688 */
3689 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3690 ac->nodemask) {
3691 unsigned long available;
ede37713 3692 unsigned long reclaimable;
d379f01d
MH
3693 unsigned long min_wmark = min_wmark_pages(zone);
3694 bool wmark;
0a0337e0 3695
5a1c84b4 3696 available = reclaimable = zone_reclaimable_pages(zone);
5a1c84b4 3697 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
0a0337e0
MH
3698
3699 /*
491d79ae
JW
3700 * Would the allocation succeed if we reclaimed all
3701 * reclaimable pages?
0a0337e0 3702 */
d379f01d
MH
3703 wmark = __zone_watermark_ok(zone, order, min_wmark,
3704 ac_classzone_idx(ac), alloc_flags, available);
3705 trace_reclaim_retry_zone(z, order, reclaimable,
3706 available, min_wmark, *no_progress_loops, wmark);
3707 if (wmark) {
ede37713
MH
3708 /*
3709 * If we didn't make any progress and have a lot of
3710 * dirty + writeback pages then we should wait for
3711 * an IO to complete to slow down the reclaim and
3712 * prevent from pre mature OOM
3713 */
3714 if (!did_some_progress) {
11fb9989 3715 unsigned long write_pending;
ede37713 3716
5a1c84b4
MG
3717 write_pending = zone_page_state_snapshot(zone,
3718 NR_ZONE_WRITE_PENDING);
ede37713 3719
11fb9989 3720 if (2 * write_pending > reclaimable) {
ede37713
MH
3721 congestion_wait(BLK_RW_ASYNC, HZ/10);
3722 return true;
3723 }
3724 }
5a1c84b4 3725
ede37713
MH
3726 /*
3727 * Memory allocation/reclaim might be called from a WQ
3728 * context and the current implementation of the WQ
3729 * concurrency control doesn't recognize that
3730 * a particular WQ is congested if the worker thread is
3731 * looping without ever sleeping. Therefore we have to
3732 * do a short sleep here rather than calling
3733 * cond_resched().
3734 */
3735 if (current->flags & PF_WQ_WORKER)
3736 schedule_timeout_uninterruptible(1);
3737 else
3738 cond_resched();
3739
0a0337e0
MH
3740 return true;
3741 }
3742 }
3743
3744 return false;
3745}
3746
902b6281
VB
3747static inline bool
3748check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
3749{
3750 /*
3751 * It's possible that cpuset's mems_allowed and the nodemask from
3752 * mempolicy don't intersect. This should be normally dealt with by
3753 * policy_nodemask(), but it's possible to race with cpuset update in
3754 * such a way the check therein was true, and then it became false
3755 * before we got our cpuset_mems_cookie here.
3756 * This assumes that for all allocations, ac->nodemask can come only
3757 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
3758 * when it does not intersect with the cpuset restrictions) or the
3759 * caller can deal with a violated nodemask.
3760 */
3761 if (cpusets_enabled() && ac->nodemask &&
3762 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
3763 ac->nodemask = NULL;
3764 return true;
3765 }
3766
3767 /*
3768 * When updating a task's mems_allowed or mempolicy nodemask, it is
3769 * possible to race with parallel threads in such a way that our
3770 * allocation can fail while the mask is being updated. If we are about
3771 * to fail, check if the cpuset changed during allocation and if so,
3772 * retry.
3773 */
3774 if (read_mems_allowed_retry(cpuset_mems_cookie))
3775 return true;
3776
3777 return false;
3778}
3779
11e33f6a
MG
3780static inline struct page *
3781__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
a9263751 3782 struct alloc_context *ac)
11e33f6a 3783{
d0164adc 3784 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
282722b0 3785 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
11e33f6a 3786 struct page *page = NULL;
c603844b 3787 unsigned int alloc_flags;
11e33f6a 3788 unsigned long did_some_progress;
5ce9bfef 3789 enum compact_priority compact_priority;
c5d01d0d 3790 enum compact_result compact_result;
5ce9bfef
VB
3791 int compaction_retries;
3792 int no_progress_loops;
63f53dea
MH
3793 unsigned long alloc_start = jiffies;
3794 unsigned int stall_timeout = 10 * HZ;
5ce9bfef 3795 unsigned int cpuset_mems_cookie;
1da177e4 3796
72807a74
MG
3797 /*
3798 * In the slowpath, we sanity check order to avoid ever trying to
3799 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3800 * be using allocators in order of preference for an area that is
3801 * too large.
3802 */
1fc28b70
MG
3803 if (order >= MAX_ORDER) {
3804 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
72807a74 3805 return NULL;
1fc28b70 3806 }
1da177e4 3807
d0164adc
MG
3808 /*
3809 * We also sanity check to catch abuse of atomic reserves being used by
3810 * callers that are not in atomic context.
3811 */
3812 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3813 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3814 gfp_mask &= ~__GFP_ATOMIC;
3815
5ce9bfef
VB
3816retry_cpuset:
3817 compaction_retries = 0;
3818 no_progress_loops = 0;
3819 compact_priority = DEF_COMPACT_PRIORITY;
3820 cpuset_mems_cookie = read_mems_allowed_begin();
9a67f648
MH
3821
3822 /*
3823 * The fast path uses conservative alloc_flags to succeed only until
3824 * kswapd needs to be woken up, and to avoid the cost of setting up
3825 * alloc_flags precisely. So we do that now.
3826 */
3827 alloc_flags = gfp_to_alloc_flags(gfp_mask);
3828
e47483bc
VB
3829 /*
3830 * We need to recalculate the starting point for the zonelist iterator
3831 * because we might have used different nodemask in the fast path, or
3832 * there was a cpuset modification and we are retrying - otherwise we
3833 * could end up iterating over non-eligible zones endlessly.
3834 */
3835 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3836 ac->high_zoneidx, ac->nodemask);
3837 if (!ac->preferred_zoneref->zone)
3838 goto nopage;
3839
23771235
VB
3840 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3841 wake_all_kswapds(order, ac);
3842
3843 /*
3844 * The adjusted alloc_flags might result in immediate success, so try
3845 * that first
3846 */
3847 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3848 if (page)
3849 goto got_pg;
3850
a8161d1e
VB
3851 /*
3852 * For costly allocations, try direct compaction first, as it's likely
282722b0
VB
3853 * that we have enough base pages and don't need to reclaim. For non-
3854 * movable high-order allocations, do that as well, as compaction will
3855 * try prevent permanent fragmentation by migrating from blocks of the
3856 * same migratetype.
3857 * Don't try this for allocations that are allowed to ignore
3858 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
a8161d1e 3859 */
282722b0
VB
3860 if (can_direct_reclaim &&
3861 (costly_order ||
3862 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
3863 && !gfp_pfmemalloc_allowed(gfp_mask)) {
a8161d1e
VB
3864 page = __alloc_pages_direct_compact(gfp_mask, order,
3865 alloc_flags, ac,
a5508cd8 3866 INIT_COMPACT_PRIORITY,
a8161d1e
VB
3867 &compact_result);
3868 if (page)
3869 goto got_pg;
3870
3eb2771b
VB
3871 /*
3872 * Checks for costly allocations with __GFP_NORETRY, which
3873 * includes THP page fault allocations
3874 */
282722b0 3875 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
a8161d1e
VB
3876 /*
3877 * If compaction is deferred for high-order allocations,
3878 * it is because sync compaction recently failed. If
3879 * this is the case and the caller requested a THP
3880 * allocation, we do not want to heavily disrupt the
3881 * system, so we fail the allocation instead of entering
3882 * direct reclaim.
3883 */
3884 if (compact_result == COMPACT_DEFERRED)
3885 goto nopage;
3886
a8161d1e 3887 /*
3eb2771b
VB
3888 * Looks like reclaim/compaction is worth trying, but
3889 * sync compaction could be very expensive, so keep
25160354 3890 * using async compaction.
a8161d1e 3891 */
a5508cd8 3892 compact_priority = INIT_COMPACT_PRIORITY;
a8161d1e
VB
3893 }
3894 }
23771235 3895
31a6c190 3896retry:
23771235 3897 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
31a6c190
VB
3898 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3899 wake_all_kswapds(order, ac);
3900
23771235
VB
3901 if (gfp_pfmemalloc_allowed(gfp_mask))
3902 alloc_flags = ALLOC_NO_WATERMARKS;
3903
e46e7b77
MG
3904 /*
3905 * Reset the zonelist iterators if memory policies can be ignored.
3906 * These allocations are high priority and system rather than user
3907 * orientated.
3908 */
23771235 3909 if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
e46e7b77
MG
3910 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3911 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3912 ac->high_zoneidx, ac->nodemask);
3913 }
3914
23771235 3915 /* Attempt with potentially adjusted zonelist and alloc_flags */
31a6c190 3916 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
7fb1d9fc
RS
3917 if (page)
3918 goto got_pg;
1da177e4 3919
d0164adc 3920 /* Caller is not willing to reclaim, we can't balance anything */
9a67f648 3921 if (!can_direct_reclaim)
1da177e4
LT
3922 goto nopage;
3923
9a67f648
MH
3924 /* Make sure we know about allocations which stall for too long */
3925 if (time_after(jiffies, alloc_start + stall_timeout)) {
82251963 3926 warn_alloc(gfp_mask & ~__GFP_NOWARN, ac->nodemask,
9a67f648
MH
3927 "page allocation stalls for %ums, order:%u",
3928 jiffies_to_msecs(jiffies-alloc_start), order);
3929 stall_timeout += 10 * HZ;
33d53103 3930 }
341ce06f 3931
9a67f648
MH
3932 /* Avoid recursion of direct reclaim */
3933 if (current->flags & PF_MEMALLOC)
6583bb64
DR
3934 goto nopage;
3935
a8161d1e
VB
3936 /* Try direct reclaim and then allocating */
3937 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3938 &did_some_progress);
3939 if (page)
3940 goto got_pg;
3941
3942 /* Try direct compaction and then allocating */
a9263751 3943 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
a5508cd8 3944 compact_priority, &compact_result);
56de7263
MG
3945 if (page)
3946 goto got_pg;
75f30861 3947
9083905a
JW
3948 /* Do not loop if specifically requested */
3949 if (gfp_mask & __GFP_NORETRY)
a8161d1e 3950 goto nopage;
9083905a 3951
0a0337e0
MH
3952 /*
3953 * Do not retry costly high order allocations unless they are
dcda9b04 3954 * __GFP_RETRY_MAYFAIL
0a0337e0 3955 */
dcda9b04 3956 if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
a8161d1e 3957 goto nopage;
0a0337e0 3958
0a0337e0 3959 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
423b452e 3960 did_some_progress > 0, &no_progress_loops))
0a0337e0
MH
3961 goto retry;
3962
33c2d214
MH
3963 /*
3964 * It doesn't make any sense to retry for the compaction if the order-0
3965 * reclaim is not able to make any progress because the current
3966 * implementation of the compaction depends on the sufficient amount
3967 * of free memory (see __compaction_suitable)
3968 */
3969 if (did_some_progress > 0 &&
86a294a8 3970 should_compact_retry(ac, order, alloc_flags,
a5508cd8 3971 compact_result, &compact_priority,
d9436498 3972 &compaction_retries))
33c2d214
MH
3973 goto retry;
3974
902b6281
VB
3975
3976 /* Deal with possible cpuset update races before we start OOM killing */
3977 if (check_retry_cpuset(cpuset_mems_cookie, ac))
e47483bc
VB
3978 goto retry_cpuset;
3979
9083905a
JW
3980 /* Reclaim has failed us, start killing things */
3981 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3982 if (page)
3983 goto got_pg;
3984
9a67f648 3985 /* Avoid allocations with no watermarks from looping endlessly */
c288983d
TH
3986 if (test_thread_flag(TIF_MEMDIE) &&
3987 (alloc_flags == ALLOC_NO_WATERMARKS ||
3988 (gfp_mask & __GFP_NOMEMALLOC)))
9a67f648
MH
3989 goto nopage;
3990
9083905a 3991 /* Retry as long as the OOM killer is making progress */
0a0337e0
MH
3992 if (did_some_progress) {
3993 no_progress_loops = 0;
9083905a 3994 goto retry;
0a0337e0 3995 }
9083905a 3996
1da177e4 3997nopage:
902b6281
VB
3998 /* Deal with possible cpuset update races before we fail */
3999 if (check_retry_cpuset(cpuset_mems_cookie, ac))
5ce9bfef
VB
4000 goto retry_cpuset;
4001
9a67f648
MH
4002 /*
4003 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4004 * we always retry
4005 */
4006 if (gfp_mask & __GFP_NOFAIL) {
4007 /*
4008 * All existing users of the __GFP_NOFAIL are blockable, so warn
4009 * of any new users that actually require GFP_NOWAIT
4010 */
4011 if (WARN_ON_ONCE(!can_direct_reclaim))
4012 goto fail;
4013
4014 /*
4015 * PF_MEMALLOC request from this context is rather bizarre
4016 * because we cannot reclaim anything and only can loop waiting
4017 * for somebody to do a work for us
4018 */
4019 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4020
4021 /*
4022 * non failing costly orders are a hard requirement which we
4023 * are not prepared for much so let's warn about these users
4024 * so that we can identify them and convert them to something
4025 * else.
4026 */
4027 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4028
6c18ba7a
MH
4029 /*
4030 * Help non-failing allocations by giving them access to memory
4031 * reserves but do not use ALLOC_NO_WATERMARKS because this
4032 * could deplete whole memory reserves which would just make
4033 * the situation worse
4034 */
4035 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4036 if (page)
4037 goto got_pg;
4038
9a67f648
MH
4039 cond_resched();
4040 goto retry;
4041 }
4042fail:
a8e99259 4043 warn_alloc(gfp_mask, ac->nodemask,
7877cdcc 4044 "page allocation failure: order:%u", order);
1da177e4 4045got_pg:
072bb0aa 4046 return page;
1da177e4 4047}
11e33f6a 4048
9cd75558 4049static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
04ec6264 4050 int preferred_nid, nodemask_t *nodemask,
9cd75558
MG
4051 struct alloc_context *ac, gfp_t *alloc_mask,
4052 unsigned int *alloc_flags)
11e33f6a 4053{
9cd75558 4054 ac->high_zoneidx = gfp_zone(gfp_mask);
04ec6264 4055 ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
9cd75558
MG
4056 ac->nodemask = nodemask;
4057 ac->migratetype = gfpflags_to_migratetype(gfp_mask);
11e33f6a 4058
682a3385 4059 if (cpusets_enabled()) {
9cd75558 4060 *alloc_mask |= __GFP_HARDWALL;
9cd75558
MG
4061 if (!ac->nodemask)
4062 ac->nodemask = &cpuset_current_mems_allowed;
51047820
VB
4063 else
4064 *alloc_flags |= ALLOC_CPUSET;
682a3385
MG
4065 }
4066
11e33f6a
MG
4067 lockdep_trace_alloc(gfp_mask);
4068
d0164adc 4069 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
11e33f6a
MG
4070
4071 if (should_fail_alloc_page(gfp_mask, order))
9cd75558 4072 return false;
11e33f6a 4073
9cd75558
MG
4074 if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4075 *alloc_flags |= ALLOC_CMA;
4076
4077 return true;
4078}
21bb9bd1 4079
9cd75558
MG
4080/* Determine whether to spread dirty pages and what the first usable zone */
4081static inline void finalise_ac(gfp_t gfp_mask,
4082 unsigned int order, struct alloc_context *ac)
4083{
c9ab0c4f 4084 /* Dirty zone balancing only done in the fast path */
9cd75558 4085 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
c9ab0c4f 4086
e46e7b77
MG
4087 /*
4088 * The preferred zone is used for statistics but crucially it is
4089 * also used as the starting point for the zonelist iterator. It
4090 * may get reset for allocations that ignore memory policies.
4091 */
9cd75558
MG
4092 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4093 ac->high_zoneidx, ac->nodemask);
4094}
4095
4096/*
4097 * This is the 'heart' of the zoned buddy allocator.
4098 */
4099struct page *
04ec6264
VB
4100__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4101 nodemask_t *nodemask)
9cd75558
MG
4102{
4103 struct page *page;
4104 unsigned int alloc_flags = ALLOC_WMARK_LOW;
4105 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
4106 struct alloc_context ac = { };
4107
4108 gfp_mask &= gfp_allowed_mask;
04ec6264 4109 if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
9cd75558
MG
4110 return NULL;
4111
4112 finalise_ac(gfp_mask, order, &ac);
5bb1b169 4113
5117f45d 4114 /* First allocation attempt */
a9263751 4115 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4fcb0971
MG
4116 if (likely(page))
4117 goto out;
11e33f6a 4118
4fcb0971 4119 /*
7dea19f9
MH
4120 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4121 * resp. GFP_NOIO which has to be inherited for all allocation requests
4122 * from a particular context which has been marked by
4123 * memalloc_no{fs,io}_{save,restore}.
4fcb0971 4124 */
7dea19f9 4125 alloc_mask = current_gfp_context(gfp_mask);
4fcb0971 4126 ac.spread_dirty_pages = false;
23f086f9 4127
4741526b
MG
4128 /*
4129 * Restore the original nodemask if it was potentially replaced with
4130 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4131 */
e47483bc 4132 if (unlikely(ac.nodemask != nodemask))
4741526b 4133 ac.nodemask = nodemask;
16096c25 4134
4fcb0971 4135 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
cc9a6c87 4136
4fcb0971 4137out:
c4159a75
VD
4138 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4139 unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4140 __free_pages(page, order);
4141 page = NULL;
4949148a
VD
4142 }
4143
4fcb0971
MG
4144 if (kmemcheck_enabled && page)
4145 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
4146
4147 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4148
11e33f6a 4149 return page;
1da177e4 4150}
d239171e 4151EXPORT_SYMBOL(__alloc_pages_nodemask);
1da177e4
LT
4152
4153/*
4154 * Common helper functions.
4155 */
920c7a5d 4156unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1da177e4 4157{
945a1113
AM
4158 struct page *page;
4159
4160 /*
4161 * __get_free_pages() returns a 32-bit address, which cannot represent
4162 * a highmem page
4163 */
4164 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
4165
1da177e4
LT
4166 page = alloc_pages(gfp_mask, order);
4167 if (!page)
4168 return 0;
4169 return (unsigned long) page_address(page);
4170}
1da177e4
LT
4171EXPORT_SYMBOL(__get_free_pages);
4172
920c7a5d 4173unsigned long get_zeroed_page(gfp_t gfp_mask)
1da177e4 4174{
945a1113 4175 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
1da177e4 4176}
1da177e4
LT
4177EXPORT_SYMBOL(get_zeroed_page);
4178
920c7a5d 4179void __free_pages(struct page *page, unsigned int order)
1da177e4 4180{
b5810039 4181 if (put_page_testzero(page)) {
1da177e4 4182 if (order == 0)
b745bc85 4183 free_hot_cold_page(page, false);
1da177e4
LT
4184 else
4185 __free_pages_ok(page, order);
4186 }
4187}
4188
4189EXPORT_SYMBOL(__free_pages);
4190
920c7a5d 4191void free_pages(unsigned long addr, unsigned int order)
1da177e4
LT
4192{
4193 if (addr != 0) {
725d704e 4194 VM_BUG_ON(!virt_addr_valid((void *)addr));
1da177e4
LT
4195 __free_pages(virt_to_page((void *)addr), order);
4196 }
4197}
4198
4199EXPORT_SYMBOL(free_pages);
4200
b63ae8ca
AD
4201/*
4202 * Page Fragment:
4203 * An arbitrary-length arbitrary-offset area of memory which resides
4204 * within a 0 or higher order page. Multiple fragments within that page
4205 * are individually refcounted, in the page's reference counter.
4206 *
4207 * The page_frag functions below provide a simple allocation framework for
4208 * page fragments. This is used by the network stack and network device
4209 * drivers to provide a backing region of memory for use as either an
4210 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4211 */
2976db80
AD
4212static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4213 gfp_t gfp_mask)
b63ae8ca
AD
4214{
4215 struct page *page = NULL;
4216 gfp_t gfp = gfp_mask;
4217
4218#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4219 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4220 __GFP_NOMEMALLOC;
4221 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4222 PAGE_FRAG_CACHE_MAX_ORDER);
4223 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4224#endif
4225 if (unlikely(!page))
4226 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4227
4228 nc->va = page ? page_address(page) : NULL;
4229
4230 return page;
4231}
4232
2976db80 4233void __page_frag_cache_drain(struct page *page, unsigned int count)
44fdffd7
AD
4234{
4235 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4236
4237 if (page_ref_sub_and_test(page, count)) {
2976db80
AD
4238 unsigned int order = compound_order(page);
4239
44fdffd7
AD
4240 if (order == 0)
4241 free_hot_cold_page(page, false);
4242 else
4243 __free_pages_ok(page, order);
4244 }
4245}
2976db80 4246EXPORT_SYMBOL(__page_frag_cache_drain);
44fdffd7 4247
8c2dd3e4
AD
4248void *page_frag_alloc(struct page_frag_cache *nc,
4249 unsigned int fragsz, gfp_t gfp_mask)
b63ae8ca
AD
4250{
4251 unsigned int size = PAGE_SIZE;
4252 struct page *page;
4253 int offset;
4254
4255 if (unlikely(!nc->va)) {
4256refill:
2976db80 4257 page = __page_frag_cache_refill(nc, gfp_mask);
b63ae8ca
AD
4258 if (!page)
4259 return NULL;
4260
4261#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4262 /* if size can vary use size else just use PAGE_SIZE */
4263 size = nc->size;
4264#endif
4265 /* Even if we own the page, we do not use atomic_set().
4266 * This would break get_page_unless_zero() users.
4267 */
fe896d18 4268 page_ref_add(page, size - 1);
b63ae8ca
AD
4269
4270 /* reset page count bias and offset to start of new frag */
2f064f34 4271 nc->pfmemalloc = page_is_pfmemalloc(page);
b63ae8ca
AD
4272 nc->pagecnt_bias = size;
4273 nc->offset = size;
4274 }
4275
4276 offset = nc->offset - fragsz;
4277 if (unlikely(offset < 0)) {
4278 page = virt_to_page(nc->va);
4279
fe896d18 4280 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
b63ae8ca
AD
4281 goto refill;
4282
4283#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4284 /* if size can vary use size else just use PAGE_SIZE */
4285 size = nc->size;
4286#endif
4287 /* OK, page count is 0, we can safely set it */
fe896d18 4288 set_page_count(page, size);
b63ae8ca
AD
4289
4290 /* reset page count bias and offset to start of new frag */
4291 nc->pagecnt_bias = size;
4292 offset = size - fragsz;
4293 }
4294
4295 nc->pagecnt_bias--;
4296 nc->offset = offset;
4297
4298 return nc->va + offset;
4299}
8c2dd3e4 4300EXPORT_SYMBOL(page_frag_alloc);
b63ae8ca
AD
4301
4302/*
4303 * Frees a page fragment allocated out of either a compound or order 0 page.
4304 */
8c2dd3e4 4305void page_frag_free(void *addr)
b63ae8ca
AD
4306{
4307 struct page *page = virt_to_head_page(addr);
4308
4309 if (unlikely(put_page_testzero(page)))
4310 __free_pages_ok(page, compound_order(page));
4311}
8c2dd3e4 4312EXPORT_SYMBOL(page_frag_free);
b63ae8ca 4313
d00181b9
KS
4314static void *make_alloc_exact(unsigned long addr, unsigned int order,
4315 size_t size)
ee85c2e1
AK
4316{
4317 if (addr) {
4318 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4319 unsigned long used = addr + PAGE_ALIGN(size);
4320
4321 split_page(virt_to_page((void *)addr), order);
4322 while (used < alloc_end) {
4323 free_page(used);
4324 used += PAGE_SIZE;
4325 }
4326 }
4327 return (void *)addr;
4328}
4329
2be0ffe2
TT
4330/**
4331 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4332 * @size: the number of bytes to allocate
4333 * @gfp_mask: GFP flags for the allocation
4334 *
4335 * This function is similar to alloc_pages(), except that it allocates the
4336 * minimum number of pages to satisfy the request. alloc_pages() can only
4337 * allocate memory in power-of-two pages.
4338 *
4339 * This function is also limited by MAX_ORDER.
4340 *
4341 * Memory allocated by this function must be released by free_pages_exact().
4342 */
4343void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4344{
4345 unsigned int order = get_order(size);
4346 unsigned long addr;
4347
4348 addr = __get_free_pages(gfp_mask, order);
ee85c2e1 4349 return make_alloc_exact(addr, order, size);
2be0ffe2
TT
4350}
4351EXPORT_SYMBOL(alloc_pages_exact);
4352
ee85c2e1
AK
4353/**
4354 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4355 * pages on a node.
b5e6ab58 4356 * @nid: the preferred node ID where memory should be allocated
ee85c2e1
AK
4357 * @size: the number of bytes to allocate
4358 * @gfp_mask: GFP flags for the allocation
4359 *
4360 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4361 * back.
ee85c2e1 4362 */
e1931811 4363void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
ee85c2e1 4364{
d00181b9 4365 unsigned int order = get_order(size);
ee85c2e1
AK
4366 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4367 if (!p)
4368 return NULL;
4369 return make_alloc_exact((unsigned long)page_address(p), order, size);
4370}
ee85c2e1 4371
2be0ffe2
TT
4372/**
4373 * free_pages_exact - release memory allocated via alloc_pages_exact()
4374 * @virt: the value returned by alloc_pages_exact.
4375 * @size: size of allocation, same value as passed to alloc_pages_exact().
4376 *
4377 * Release the memory allocated by a previous call to alloc_pages_exact.
4378 */
4379void free_pages_exact(void *virt, size_t size)
4380{
4381 unsigned long addr = (unsigned long)virt;
4382 unsigned long end = addr + PAGE_ALIGN(size);
4383
4384 while (addr < end) {
4385 free_page(addr);
4386 addr += PAGE_SIZE;
4387 }
4388}
4389EXPORT_SYMBOL(free_pages_exact);
4390
e0fb5815
ZY
4391/**
4392 * nr_free_zone_pages - count number of pages beyond high watermark
4393 * @offset: The zone index of the highest zone
4394 *
4395 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4396 * high watermark within all zones at or below a given zone index. For each
4397 * zone, the number of pages is calculated as:
0e056eb5
MCC
4398 *
4399 * nr_free_zone_pages = managed_pages - high_pages
e0fb5815 4400 */
ebec3862 4401static unsigned long nr_free_zone_pages(int offset)
1da177e4 4402{
dd1a239f 4403 struct zoneref *z;
54a6eb5c
MG
4404 struct zone *zone;
4405
e310fd43 4406 /* Just pick one node, since fallback list is circular */
ebec3862 4407 unsigned long sum = 0;
1da177e4 4408
0e88460d 4409 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1da177e4 4410
54a6eb5c 4411 for_each_zone_zonelist(zone, z, zonelist, offset) {
b40da049 4412 unsigned long size = zone->managed_pages;
41858966 4413 unsigned long high = high_wmark_pages(zone);
e310fd43
MB
4414 if (size > high)
4415 sum += size - high;
1da177e4
LT
4416 }
4417
4418 return sum;
4419}
4420
e0fb5815
ZY
4421/**
4422 * nr_free_buffer_pages - count number of pages beyond high watermark
4423 *
4424 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4425 * watermark within ZONE_DMA and ZONE_NORMAL.
1da177e4 4426 */
ebec3862 4427unsigned long nr_free_buffer_pages(void)
1da177e4 4428{
af4ca457 4429 return nr_free_zone_pages(gfp_zone(GFP_USER));
1da177e4 4430}
c2f1a551 4431EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1da177e4 4432
e0fb5815
ZY
4433/**
4434 * nr_free_pagecache_pages - count number of pages beyond high watermark
4435 *
4436 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4437 * high watermark within all zones.
1da177e4 4438 */
ebec3862 4439unsigned long nr_free_pagecache_pages(void)
1da177e4 4440{
2a1e274a 4441 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1da177e4 4442}
08e0f6a9
CL
4443
4444static inline void show_node(struct zone *zone)
1da177e4 4445{
e5adfffc 4446 if (IS_ENABLED(CONFIG_NUMA))
25ba77c1 4447 printk("Node %d ", zone_to_nid(zone));
1da177e4 4448}
1da177e4 4449
d02bd27b
IR
4450long si_mem_available(void)
4451{
4452 long available;
4453 unsigned long pagecache;
4454 unsigned long wmark_low = 0;
4455 unsigned long pages[NR_LRU_LISTS];
4456 struct zone *zone;
4457 int lru;
4458
4459 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
2f95ff90 4460 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
d02bd27b
IR
4461
4462 for_each_zone(zone)
4463 wmark_low += zone->watermark[WMARK_LOW];
4464
4465 /*
4466 * Estimate the amount of memory available for userspace allocations,
4467 * without causing swapping.
4468 */
4469 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4470
4471 /*
4472 * Not all the page cache can be freed, otherwise the system will
4473 * start swapping. Assume at least half of the page cache, or the
4474 * low watermark worth of cache, needs to stay.
4475 */
4476 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4477 pagecache -= min(pagecache / 2, wmark_low);
4478 available += pagecache;
4479
4480 /*
4481 * Part of the reclaimable slab consists of items that are in use,
4482 * and cannot be freed. Cap this estimate at the low watermark.
4483 */
d507e2eb
JW
4484 available += global_node_page_state(NR_SLAB_RECLAIMABLE) -
4485 min(global_node_page_state(NR_SLAB_RECLAIMABLE) / 2,
4486 wmark_low);
d02bd27b
IR
4487
4488 if (available < 0)
4489 available = 0;
4490 return available;
4491}
4492EXPORT_SYMBOL_GPL(si_mem_available);
4493
1da177e4
LT
4494void si_meminfo(struct sysinfo *val)
4495{
4496 val->totalram = totalram_pages;
11fb9989 4497 val->sharedram = global_node_page_state(NR_SHMEM);
d23ad423 4498 val->freeram = global_page_state(NR_FREE_PAGES);
1da177e4 4499 val->bufferram = nr_blockdev_pages();
1da177e4
LT
4500 val->totalhigh = totalhigh_pages;
4501 val->freehigh = nr_free_highpages();
1da177e4
LT
4502 val->mem_unit = PAGE_SIZE;
4503}
4504
4505EXPORT_SYMBOL(si_meminfo);
4506
4507#ifdef CONFIG_NUMA
4508void si_meminfo_node(struct sysinfo *val, int nid)
4509{
cdd91a77
JL
4510 int zone_type; /* needs to be signed */
4511 unsigned long managed_pages = 0;
fc2bd799
JK
4512 unsigned long managed_highpages = 0;
4513 unsigned long free_highpages = 0;
1da177e4
LT
4514 pg_data_t *pgdat = NODE_DATA(nid);
4515
cdd91a77
JL
4516 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4517 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4518 val->totalram = managed_pages;
11fb9989 4519 val->sharedram = node_page_state(pgdat, NR_SHMEM);
75ef7184 4520 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
98d2b0eb 4521#ifdef CONFIG_HIGHMEM
fc2bd799
JK
4522 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4523 struct zone *zone = &pgdat->node_zones[zone_type];
4524
4525 if (is_highmem(zone)) {
4526 managed_highpages += zone->managed_pages;
4527 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4528 }
4529 }
4530 val->totalhigh = managed_highpages;
4531 val->freehigh = free_highpages;
98d2b0eb 4532#else
fc2bd799
JK
4533 val->totalhigh = managed_highpages;
4534 val->freehigh = free_highpages;
98d2b0eb 4535#endif
1da177e4
LT
4536 val->mem_unit = PAGE_SIZE;
4537}
4538#endif
4539
ddd588b5 4540/*
7bf02ea2
DR
4541 * Determine whether the node should be displayed or not, depending on whether
4542 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
ddd588b5 4543 */
9af744d7 4544static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
ddd588b5 4545{
ddd588b5 4546 if (!(flags & SHOW_MEM_FILTER_NODES))
9af744d7 4547 return false;
ddd588b5 4548
9af744d7
MH
4549 /*
4550 * no node mask - aka implicit memory numa policy. Do not bother with
4551 * the synchronization - read_mems_allowed_begin - because we do not
4552 * have to be precise here.
4553 */
4554 if (!nodemask)
4555 nodemask = &cpuset_current_mems_allowed;
4556
4557 return !node_isset(nid, *nodemask);
ddd588b5
DR
4558}
4559
1da177e4
LT
4560#define K(x) ((x) << (PAGE_SHIFT-10))
4561
377e4f16
RV
4562static void show_migration_types(unsigned char type)
4563{
4564 static const char types[MIGRATE_TYPES] = {
4565 [MIGRATE_UNMOVABLE] = 'U',
377e4f16 4566 [MIGRATE_MOVABLE] = 'M',
475a2f90
VB
4567 [MIGRATE_RECLAIMABLE] = 'E',
4568 [MIGRATE_HIGHATOMIC] = 'H',
377e4f16
RV
4569#ifdef CONFIG_CMA
4570 [MIGRATE_CMA] = 'C',
4571#endif
194159fb 4572#ifdef CONFIG_MEMORY_ISOLATION
377e4f16 4573 [MIGRATE_ISOLATE] = 'I',
194159fb 4574#endif
377e4f16
RV
4575 };
4576 char tmp[MIGRATE_TYPES + 1];
4577 char *p = tmp;
4578 int i;
4579
4580 for (i = 0; i < MIGRATE_TYPES; i++) {
4581 if (type & (1 << i))
4582 *p++ = types[i];
4583 }
4584
4585 *p = '\0';
1f84a18f 4586 printk(KERN_CONT "(%s) ", tmp);
377e4f16
RV
4587}
4588
1da177e4
LT
4589/*
4590 * Show free area list (used inside shift_scroll-lock stuff)
4591 * We also calculate the percentage fragmentation. We do this by counting the
4592 * memory on each free list with the exception of the first item on the list.
d1bfcdb8
KK
4593 *
4594 * Bits in @filter:
4595 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4596 * cpuset.
1da177e4 4597 */
9af744d7 4598void show_free_areas(unsigned int filter, nodemask_t *nodemask)
1da177e4 4599{
d1bfcdb8 4600 unsigned long free_pcp = 0;
c7241913 4601 int cpu;
1da177e4 4602 struct zone *zone;
599d0c95 4603 pg_data_t *pgdat;
1da177e4 4604
ee99c71c 4605 for_each_populated_zone(zone) {
9af744d7 4606 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
ddd588b5 4607 continue;
d1bfcdb8 4608
761b0677
KK
4609 for_each_online_cpu(cpu)
4610 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
1da177e4
LT
4611 }
4612
a731286d
KM
4613 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4614 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
d1bfcdb8
KK
4615 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4616 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
d1ce749a 4617 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
d1bfcdb8 4618 " free:%lu free_pcp:%lu free_cma:%lu\n",
599d0c95
MG
4619 global_node_page_state(NR_ACTIVE_ANON),
4620 global_node_page_state(NR_INACTIVE_ANON),
4621 global_node_page_state(NR_ISOLATED_ANON),
4622 global_node_page_state(NR_ACTIVE_FILE),
4623 global_node_page_state(NR_INACTIVE_FILE),
4624 global_node_page_state(NR_ISOLATED_FILE),
4625 global_node_page_state(NR_UNEVICTABLE),
11fb9989
MG
4626 global_node_page_state(NR_FILE_DIRTY),
4627 global_node_page_state(NR_WRITEBACK),
4628 global_node_page_state(NR_UNSTABLE_NFS),
d507e2eb
JW
4629 global_node_page_state(NR_SLAB_RECLAIMABLE),
4630 global_node_page_state(NR_SLAB_UNRECLAIMABLE),
50658e2e 4631 global_node_page_state(NR_FILE_MAPPED),
11fb9989 4632 global_node_page_state(NR_SHMEM),
a25700a5 4633 global_page_state(NR_PAGETABLE),
d1ce749a 4634 global_page_state(NR_BOUNCE),
d1bfcdb8
KK
4635 global_page_state(NR_FREE_PAGES),
4636 free_pcp,
d1ce749a 4637 global_page_state(NR_FREE_CMA_PAGES));
1da177e4 4638
599d0c95 4639 for_each_online_pgdat(pgdat) {
9af744d7 4640 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
c02e50bb
MH
4641 continue;
4642
599d0c95
MG
4643 printk("Node %d"
4644 " active_anon:%lukB"
4645 " inactive_anon:%lukB"
4646 " active_file:%lukB"
4647 " inactive_file:%lukB"
4648 " unevictable:%lukB"
4649 " isolated(anon):%lukB"
4650 " isolated(file):%lukB"
50658e2e 4651 " mapped:%lukB"
11fb9989
MG
4652 " dirty:%lukB"
4653 " writeback:%lukB"
4654 " shmem:%lukB"
4655#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4656 " shmem_thp: %lukB"
4657 " shmem_pmdmapped: %lukB"
4658 " anon_thp: %lukB"
4659#endif
4660 " writeback_tmp:%lukB"
4661 " unstable:%lukB"
599d0c95
MG
4662 " all_unreclaimable? %s"
4663 "\n",
4664 pgdat->node_id,
4665 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4666 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4667 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4668 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4669 K(node_page_state(pgdat, NR_UNEVICTABLE)),
4670 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4671 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
50658e2e 4672 K(node_page_state(pgdat, NR_FILE_MAPPED)),
11fb9989
MG
4673 K(node_page_state(pgdat, NR_FILE_DIRTY)),
4674 K(node_page_state(pgdat, NR_WRITEBACK)),
1f06b81a 4675 K(node_page_state(pgdat, NR_SHMEM)),
11fb9989
MG
4676#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4677 K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4678 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4679 * HPAGE_PMD_NR),
4680 K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4681#endif
11fb9989
MG
4682 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4683 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
c73322d0
JW
4684 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
4685 "yes" : "no");
599d0c95
MG
4686 }
4687
ee99c71c 4688 for_each_populated_zone(zone) {
1da177e4
LT
4689 int i;
4690
9af744d7 4691 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
ddd588b5 4692 continue;
d1bfcdb8
KK
4693
4694 free_pcp = 0;
4695 for_each_online_cpu(cpu)
4696 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4697
1da177e4 4698 show_node(zone);
1f84a18f
JP
4699 printk(KERN_CONT
4700 "%s"
1da177e4
LT
4701 " free:%lukB"
4702 " min:%lukB"
4703 " low:%lukB"
4704 " high:%lukB"
71c799f4
MK
4705 " active_anon:%lukB"
4706 " inactive_anon:%lukB"
4707 " active_file:%lukB"
4708 " inactive_file:%lukB"
4709 " unevictable:%lukB"
5a1c84b4 4710 " writepending:%lukB"
1da177e4 4711 " present:%lukB"
9feedc9d 4712 " managed:%lukB"
4a0aa73f 4713 " mlocked:%lukB"
c6a7f572 4714 " kernel_stack:%lukB"
4a0aa73f 4715 " pagetables:%lukB"
4a0aa73f 4716 " bounce:%lukB"
d1bfcdb8
KK
4717 " free_pcp:%lukB"
4718 " local_pcp:%ukB"
d1ce749a 4719 " free_cma:%lukB"
1da177e4
LT
4720 "\n",
4721 zone->name,
88f5acf8 4722 K(zone_page_state(zone, NR_FREE_PAGES)),
41858966
MG
4723 K(min_wmark_pages(zone)),
4724 K(low_wmark_pages(zone)),
4725 K(high_wmark_pages(zone)),
71c799f4
MK
4726 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4727 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4728 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4729 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4730 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5a1c84b4 4731 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
1da177e4 4732 K(zone->present_pages),
9feedc9d 4733 K(zone->managed_pages),
4a0aa73f 4734 K(zone_page_state(zone, NR_MLOCK)),
d30dd8be 4735 zone_page_state(zone, NR_KERNEL_STACK_KB),
4a0aa73f 4736 K(zone_page_state(zone, NR_PAGETABLE)),
4a0aa73f 4737 K(zone_page_state(zone, NR_BOUNCE)),
d1bfcdb8
KK
4738 K(free_pcp),
4739 K(this_cpu_read(zone->pageset->pcp.count)),
33e077bd 4740 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
1da177e4
LT
4741 printk("lowmem_reserve[]:");
4742 for (i = 0; i < MAX_NR_ZONES; i++)
1f84a18f
JP
4743 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
4744 printk(KERN_CONT "\n");
1da177e4
LT
4745 }
4746
ee99c71c 4747 for_each_populated_zone(zone) {
d00181b9
KS
4748 unsigned int order;
4749 unsigned long nr[MAX_ORDER], flags, total = 0;
377e4f16 4750 unsigned char types[MAX_ORDER];
1da177e4 4751
9af744d7 4752 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
ddd588b5 4753 continue;
1da177e4 4754 show_node(zone);
1f84a18f 4755 printk(KERN_CONT "%s: ", zone->name);
1da177e4
LT
4756
4757 spin_lock_irqsave(&zone->lock, flags);
4758 for (order = 0; order < MAX_ORDER; order++) {
377e4f16
RV
4759 struct free_area *area = &zone->free_area[order];
4760 int type;
4761
4762 nr[order] = area->nr_free;
8f9de51a 4763 total += nr[order] << order;
377e4f16
RV
4764
4765 types[order] = 0;
4766 for (type = 0; type < MIGRATE_TYPES; type++) {
4767 if (!list_empty(&area->free_list[type]))
4768 types[order] |= 1 << type;
4769 }
1da177e4
LT
4770 }
4771 spin_unlock_irqrestore(&zone->lock, flags);
377e4f16 4772 for (order = 0; order < MAX_ORDER; order++) {
1f84a18f
JP
4773 printk(KERN_CONT "%lu*%lukB ",
4774 nr[order], K(1UL) << order);
377e4f16
RV
4775 if (nr[order])
4776 show_migration_types(types[order]);
4777 }
1f84a18f 4778 printk(KERN_CONT "= %lukB\n", K(total));
1da177e4
LT
4779 }
4780
949f7ec5
DR
4781 hugetlb_show_meminfo();
4782
11fb9989 4783 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
e6f3602d 4784
1da177e4
LT
4785 show_swap_cache_info();
4786}
4787
19770b32
MG
4788static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4789{
4790 zoneref->zone = zone;
4791 zoneref->zone_idx = zone_idx(zone);
4792}
4793
1da177e4
LT
4794/*
4795 * Builds allocation fallback zone lists.
1a93205b
CL
4796 *
4797 * Add all populated zones of a node to the zonelist.
1da177e4 4798 */
f0c0b2b8 4799static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
bc732f1d 4800 int nr_zones)
1da177e4 4801{
1a93205b 4802 struct zone *zone;
bc732f1d 4803 enum zone_type zone_type = MAX_NR_ZONES;
02a68a5e
CL
4804
4805 do {
2f6726e5 4806 zone_type--;
070f8032 4807 zone = pgdat->node_zones + zone_type;
6aa303de 4808 if (managed_zone(zone)) {
dd1a239f
MG
4809 zoneref_set_zone(zone,
4810 &zonelist->_zonerefs[nr_zones++]);
070f8032 4811 check_highest_zone(zone_type);
1da177e4 4812 }
2f6726e5 4813 } while (zone_type);
bc732f1d 4814
070f8032 4815 return nr_zones;
1da177e4
LT
4816}
4817
f0c0b2b8
KH
4818
4819/*
4820 * zonelist_order:
4821 * 0 = automatic detection of better ordering.
4822 * 1 = order by ([node] distance, -zonetype)
4823 * 2 = order by (-zonetype, [node] distance)
4824 *
4825 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4826 * the same zonelist. So only NUMA can configure this param.
4827 */
4828#define ZONELIST_ORDER_DEFAULT 0
4829#define ZONELIST_ORDER_NODE 1
4830#define ZONELIST_ORDER_ZONE 2
4831
4832/* zonelist order in the kernel.
4833 * set_zonelist_order() will set this to NODE or ZONE.
4834 */
4835static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4836static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4837
4838
1da177e4 4839#ifdef CONFIG_NUMA
f0c0b2b8
KH
4840/* The value user specified ....changed by config */
4841static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4842/* string for sysctl */
4843#define NUMA_ZONELIST_ORDER_LEN 16
4844char numa_zonelist_order[16] = "default";
4845
4846/*
4847 * interface for configure zonelist ordering.
4848 * command line option "numa_zonelist_order"
4849 * = "[dD]efault - default, automatic configuration.
4850 * = "[nN]ode - order by node locality, then by zone within node
4851 * = "[zZ]one - order by zone, then by locality within zone
4852 */
4853
4854static int __parse_numa_zonelist_order(char *s)
4855{
4856 if (*s == 'd' || *s == 'D') {
4857 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4858 } else if (*s == 'n' || *s == 'N') {
4859 user_zonelist_order = ZONELIST_ORDER_NODE;
4860 } else if (*s == 'z' || *s == 'Z') {
4861 user_zonelist_order = ZONELIST_ORDER_ZONE;
4862 } else {
1170532b 4863 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
f0c0b2b8
KH
4864 return -EINVAL;
4865 }
4866 return 0;
4867}
4868
4869static __init int setup_numa_zonelist_order(char *s)
4870{
ecb256f8
VL
4871 int ret;
4872
4873 if (!s)
4874 return 0;
4875
4876 ret = __parse_numa_zonelist_order(s);
4877 if (ret == 0)
4878 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4879
4880 return ret;
f0c0b2b8
KH
4881}
4882early_param("numa_zonelist_order", setup_numa_zonelist_order);
4883
4884/*
4885 * sysctl handler for numa_zonelist_order
4886 */
cccad5b9 4887int numa_zonelist_order_handler(struct ctl_table *table, int write,
8d65af78 4888 void __user *buffer, size_t *length,
f0c0b2b8
KH
4889 loff_t *ppos)
4890{
4891 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4892 int ret;
443c6f14 4893 static DEFINE_MUTEX(zl_order_mutex);
f0c0b2b8 4894
443c6f14 4895 mutex_lock(&zl_order_mutex);
dacbde09
CG
4896 if (write) {
4897 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4898 ret = -EINVAL;
4899 goto out;
4900 }
4901 strcpy(saved_string, (char *)table->data);
4902 }
8d65af78 4903 ret = proc_dostring(table, write, buffer, length, ppos);
f0c0b2b8 4904 if (ret)
443c6f14 4905 goto out;
f0c0b2b8
KH
4906 if (write) {
4907 int oldval = user_zonelist_order;
dacbde09
CG
4908
4909 ret = __parse_numa_zonelist_order((char *)table->data);
4910 if (ret) {
f0c0b2b8
KH
4911 /*
4912 * bogus value. restore saved string
4913 */
dacbde09 4914 strncpy((char *)table->data, saved_string,
f0c0b2b8
KH
4915 NUMA_ZONELIST_ORDER_LEN);
4916 user_zonelist_order = oldval;
4eaf3f64 4917 } else if (oldval != user_zonelist_order) {
167d0f25 4918 mem_hotplug_begin();
4eaf3f64 4919 mutex_lock(&zonelists_mutex);
9adb62a5 4920 build_all_zonelists(NULL, NULL);
4eaf3f64 4921 mutex_unlock(&zonelists_mutex);
167d0f25 4922 mem_hotplug_done();
4eaf3f64 4923 }
f0c0b2b8 4924 }
443c6f14
AK
4925out:
4926 mutex_unlock(&zl_order_mutex);
4927 return ret;
f0c0b2b8
KH
4928}
4929
4930
62bc62a8 4931#define MAX_NODE_LOAD (nr_online_nodes)
f0c0b2b8
KH
4932static int node_load[MAX_NUMNODES];
4933
1da177e4 4934/**
4dc3b16b 4935 * find_next_best_node - find the next node that should appear in a given node's fallback list
1da177e4
LT
4936 * @node: node whose fallback list we're appending
4937 * @used_node_mask: nodemask_t of already used nodes
4938 *
4939 * We use a number of factors to determine which is the next node that should
4940 * appear on a given node's fallback list. The node should not have appeared
4941 * already in @node's fallback list, and it should be the next closest node
4942 * according to the distance array (which contains arbitrary distance values
4943 * from each node to each node in the system), and should also prefer nodes
4944 * with no CPUs, since presumably they'll have very little allocation pressure
4945 * on them otherwise.
4946 * It returns -1 if no node is found.
4947 */
f0c0b2b8 4948static int find_next_best_node(int node, nodemask_t *used_node_mask)
1da177e4 4949{
4cf808eb 4950 int n, val;
1da177e4 4951 int min_val = INT_MAX;
00ef2d2f 4952 int best_node = NUMA_NO_NODE;
a70f7302 4953 const struct cpumask *tmp = cpumask_of_node(0);
1da177e4 4954
4cf808eb
LT
4955 /* Use the local node if we haven't already */
4956 if (!node_isset(node, *used_node_mask)) {
4957 node_set(node, *used_node_mask);
4958 return node;
4959 }
1da177e4 4960
4b0ef1fe 4961 for_each_node_state(n, N_MEMORY) {
1da177e4
LT
4962
4963 /* Don't want a node to appear more than once */
4964 if (node_isset(n, *used_node_mask))
4965 continue;
4966
1da177e4
LT
4967 /* Use the distance array to find the distance */
4968 val = node_distance(node, n);
4969
4cf808eb
LT
4970 /* Penalize nodes under us ("prefer the next node") */
4971 val += (n < node);
4972
1da177e4 4973 /* Give preference to headless and unused nodes */
a70f7302
RR
4974 tmp = cpumask_of_node(n);
4975 if (!cpumask_empty(tmp))
1da177e4
LT
4976 val += PENALTY_FOR_NODE_WITH_CPUS;
4977
4978 /* Slight preference for less loaded node */
4979 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4980 val += node_load[n];
4981
4982 if (val < min_val) {
4983 min_val = val;
4984 best_node = n;
4985 }
4986 }
4987
4988 if (best_node >= 0)
4989 node_set(best_node, *used_node_mask);
4990
4991 return best_node;
4992}
4993
f0c0b2b8
KH
4994
4995/*
4996 * Build zonelists ordered by node and zones within node.
4997 * This results in maximum locality--normal zone overflows into local
4998 * DMA zone, if any--but risks exhausting DMA zone.
4999 */
5000static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
1da177e4 5001{
f0c0b2b8 5002 int j;
1da177e4 5003 struct zonelist *zonelist;
f0c0b2b8 5004
c9634cf0 5005 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
dd1a239f 5006 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
54a6eb5c 5007 ;
bc732f1d 5008 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
dd1a239f
MG
5009 zonelist->_zonerefs[j].zone = NULL;
5010 zonelist->_zonerefs[j].zone_idx = 0;
f0c0b2b8
KH
5011}
5012
523b9458
CL
5013/*
5014 * Build gfp_thisnode zonelists
5015 */
5016static void build_thisnode_zonelists(pg_data_t *pgdat)
5017{
523b9458
CL
5018 int j;
5019 struct zonelist *zonelist;
5020
c9634cf0 5021 zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK];
bc732f1d 5022 j = build_zonelists_node(pgdat, zonelist, 0);
dd1a239f
MG
5023 zonelist->_zonerefs[j].zone = NULL;
5024 zonelist->_zonerefs[j].zone_idx = 0;
523b9458
CL
5025}
5026
f0c0b2b8
KH
5027/*
5028 * Build zonelists ordered by zone and nodes within zones.
5029 * This results in conserving DMA zone[s] until all Normal memory is
5030 * exhausted, but results in overflowing to remote node while memory
5031 * may still exist in local DMA zone.
5032 */
5033static int node_order[MAX_NUMNODES];
5034
5035static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
5036{
f0c0b2b8
KH
5037 int pos, j, node;
5038 int zone_type; /* needs to be signed */
5039 struct zone *z;
5040 struct zonelist *zonelist;
5041
c9634cf0 5042 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
54a6eb5c
MG
5043 pos = 0;
5044 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
5045 for (j = 0; j < nr_nodes; j++) {
5046 node = node_order[j];
5047 z = &NODE_DATA(node)->node_zones[zone_type];
6aa303de 5048 if (managed_zone(z)) {
dd1a239f
MG
5049 zoneref_set_zone(z,
5050 &zonelist->_zonerefs[pos++]);
54a6eb5c 5051 check_highest_zone(zone_type);
f0c0b2b8
KH
5052 }
5053 }
f0c0b2b8 5054 }
dd1a239f
MG
5055 zonelist->_zonerefs[pos].zone = NULL;
5056 zonelist->_zonerefs[pos].zone_idx = 0;
f0c0b2b8
KH
5057}
5058
3193913c
MG
5059#if defined(CONFIG_64BIT)
5060/*
5061 * Devices that require DMA32/DMA are relatively rare and do not justify a
5062 * penalty to every machine in case the specialised case applies. Default
5063 * to Node-ordering on 64-bit NUMA machines
5064 */
5065static int default_zonelist_order(void)
5066{
5067 return ZONELIST_ORDER_NODE;
5068}
5069#else
5070/*
5071 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
5072 * by the kernel. If processes running on node 0 deplete the low memory zone
5073 * then reclaim will occur more frequency increasing stalls and potentially
5074 * be easier to OOM if a large percentage of the zone is under writeback or
5075 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
5076 * Hence, default to zone ordering on 32-bit.
5077 */
f0c0b2b8
KH
5078static int default_zonelist_order(void)
5079{
f0c0b2b8
KH
5080 return ZONELIST_ORDER_ZONE;
5081}
3193913c 5082#endif /* CONFIG_64BIT */
f0c0b2b8
KH
5083
5084static void set_zonelist_order(void)
5085{
5086 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
5087 current_zonelist_order = default_zonelist_order();
5088 else
5089 current_zonelist_order = user_zonelist_order;
5090}
5091
5092static void build_zonelists(pg_data_t *pgdat)
5093{
c00eb15a 5094 int i, node, load;
1da177e4 5095 nodemask_t used_mask;
f0c0b2b8
KH
5096 int local_node, prev_node;
5097 struct zonelist *zonelist;
d00181b9 5098 unsigned int order = current_zonelist_order;
1da177e4
LT
5099
5100 /* initialize zonelists */
523b9458 5101 for (i = 0; i < MAX_ZONELISTS; i++) {
1da177e4 5102 zonelist = pgdat->node_zonelists + i;
dd1a239f
MG
5103 zonelist->_zonerefs[0].zone = NULL;
5104 zonelist->_zonerefs[0].zone_idx = 0;
1da177e4
LT
5105 }
5106
5107 /* NUMA-aware ordering of nodes */
5108 local_node = pgdat->node_id;
62bc62a8 5109 load = nr_online_nodes;
1da177e4
LT
5110 prev_node = local_node;
5111 nodes_clear(used_mask);
f0c0b2b8 5112
f0c0b2b8 5113 memset(node_order, 0, sizeof(node_order));
c00eb15a 5114 i = 0;
f0c0b2b8 5115
1da177e4
LT
5116 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5117 /*
5118 * We don't want to pressure a particular node.
5119 * So adding penalty to the first node in same
5120 * distance group to make it round-robin.
5121 */
957f822a
DR
5122 if (node_distance(local_node, node) !=
5123 node_distance(local_node, prev_node))
f0c0b2b8
KH
5124 node_load[node] = load;
5125
1da177e4
LT
5126 prev_node = node;
5127 load--;
f0c0b2b8
KH
5128 if (order == ZONELIST_ORDER_NODE)
5129 build_zonelists_in_node_order(pgdat, node);
5130 else
c00eb15a 5131 node_order[i++] = node; /* remember order */
f0c0b2b8 5132 }
1da177e4 5133
f0c0b2b8
KH
5134 if (order == ZONELIST_ORDER_ZONE) {
5135 /* calculate node order -- i.e., DMA last! */
c00eb15a 5136 build_zonelists_in_zone_order(pgdat, i);
1da177e4 5137 }
523b9458
CL
5138
5139 build_thisnode_zonelists(pgdat);
1da177e4
LT
5140}
5141
7aac7898
LS
5142#ifdef CONFIG_HAVE_MEMORYLESS_NODES
5143/*
5144 * Return node id of node used for "local" allocations.
5145 * I.e., first node id of first zone in arg node's generic zonelist.
5146 * Used for initializing percpu 'numa_mem', which is used primarily
5147 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5148 */
5149int local_memory_node(int node)
5150{
c33d6c06 5151 struct zoneref *z;
7aac7898 5152
c33d6c06 5153 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
7aac7898 5154 gfp_zone(GFP_KERNEL),
c33d6c06
MG
5155 NULL);
5156 return z->zone->node;
7aac7898
LS
5157}
5158#endif
f0c0b2b8 5159
6423aa81
JK
5160static void setup_min_unmapped_ratio(void);
5161static void setup_min_slab_ratio(void);
1da177e4
LT
5162#else /* CONFIG_NUMA */
5163
f0c0b2b8
KH
5164static void set_zonelist_order(void)
5165{
5166 current_zonelist_order = ZONELIST_ORDER_ZONE;
5167}
5168
5169static void build_zonelists(pg_data_t *pgdat)
1da177e4 5170{
19655d34 5171 int node, local_node;
54a6eb5c
MG
5172 enum zone_type j;
5173 struct zonelist *zonelist;
1da177e4
LT
5174
5175 local_node = pgdat->node_id;
1da177e4 5176
c9634cf0 5177 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
bc732f1d 5178 j = build_zonelists_node(pgdat, zonelist, 0);
1da177e4 5179
54a6eb5c
MG
5180 /*
5181 * Now we build the zonelist so that it contains the zones
5182 * of all the other nodes.
5183 * We don't want to pressure a particular node, so when
5184 * building the zones for node N, we make sure that the
5185 * zones coming right after the local ones are those from
5186 * node N+1 (modulo N)
5187 */
5188 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5189 if (!node_online(node))
5190 continue;
bc732f1d 5191 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
1da177e4 5192 }
54a6eb5c
MG
5193 for (node = 0; node < local_node; node++) {
5194 if (!node_online(node))
5195 continue;
bc732f1d 5196 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
54a6eb5c
MG
5197 }
5198
dd1a239f
MG
5199 zonelist->_zonerefs[j].zone = NULL;
5200 zonelist->_zonerefs[j].zone_idx = 0;
1da177e4
LT
5201}
5202
5203#endif /* CONFIG_NUMA */
5204
99dcc3e5
CL
5205/*
5206 * Boot pageset table. One per cpu which is going to be used for all
5207 * zones and all nodes. The parameters will be set in such a way
5208 * that an item put on a list will immediately be handed over to
5209 * the buddy list. This is safe since pageset manipulation is done
5210 * with interrupts disabled.
5211 *
5212 * The boot_pagesets must be kept even after bootup is complete for
5213 * unused processors and/or zones. They do play a role for bootstrapping
5214 * hotplugged processors.
5215 *
5216 * zoneinfo_show() and maybe other functions do
5217 * not check if the processor is online before following the pageset pointer.
5218 * Other parts of the kernel may not check if the zone is available.
5219 */
5220static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5221static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
385386cf 5222static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
1f522509 5223static void setup_zone_pageset(struct zone *zone);
99dcc3e5 5224
4eaf3f64
HL
5225/*
5226 * Global mutex to protect against size modification of zonelists
5227 * as well as to serialize pageset setup for the new populated zone.
5228 */
5229DEFINE_MUTEX(zonelists_mutex);
5230
9b1a4d38 5231/* return values int ....just for stop_machine() */
4ed7e022 5232static int __build_all_zonelists(void *data)
1da177e4 5233{
6811378e 5234 int nid;
99dcc3e5 5235 int cpu;
9adb62a5 5236 pg_data_t *self = data;
9276b1bc 5237
7f9cfb31
BL
5238#ifdef CONFIG_NUMA
5239 memset(node_load, 0, sizeof(node_load));
5240#endif
9adb62a5
JL
5241
5242 if (self && !node_online(self->node_id)) {
5243 build_zonelists(self);
9adb62a5
JL
5244 }
5245
9276b1bc 5246 for_each_online_node(nid) {
7ea1530a
CL
5247 pg_data_t *pgdat = NODE_DATA(nid);
5248
5249 build_zonelists(pgdat);
9276b1bc 5250 }
99dcc3e5
CL
5251
5252 /*
5253 * Initialize the boot_pagesets that are going to be used
5254 * for bootstrapping processors. The real pagesets for
5255 * each zone will be allocated later when the per cpu
5256 * allocator is available.
5257 *
5258 * boot_pagesets are used also for bootstrapping offline
5259 * cpus if the system is already booted because the pagesets
5260 * are needed to initialize allocators on a specific cpu too.
5261 * F.e. the percpu allocator needs the page allocator which
5262 * needs the percpu allocator in order to allocate its pagesets
5263 * (a chicken-egg dilemma).
5264 */
7aac7898 5265 for_each_possible_cpu(cpu) {
99dcc3e5
CL
5266 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5267
7aac7898
LS
5268#ifdef CONFIG_HAVE_MEMORYLESS_NODES
5269 /*
5270 * We now know the "local memory node" for each node--
5271 * i.e., the node of the first zone in the generic zonelist.
5272 * Set up numa_mem percpu variable for on-line cpus. During
5273 * boot, only the boot cpu should be on-line; we'll init the
5274 * secondary cpus' numa_mem as they come on-line. During
5275 * node/memory hotplug, we'll fixup all on-line cpus.
5276 */
5277 if (cpu_online(cpu))
5278 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5279#endif
5280 }
5281
6811378e
YG
5282 return 0;
5283}
5284
061f67bc
RV
5285static noinline void __init
5286build_all_zonelists_init(void)
5287{
5288 __build_all_zonelists(NULL);
5289 mminit_verify_zonelist();
5290 cpuset_init_current_mems_allowed();
5291}
5292
4eaf3f64
HL
5293/*
5294 * Called with zonelists_mutex held always
5295 * unless system_state == SYSTEM_BOOTING.
061f67bc
RV
5296 *
5297 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5298 * [we're only called with non-NULL zone through __meminit paths] and
5299 * (2) call of __init annotated helper build_all_zonelists_init
5300 * [protected by SYSTEM_BOOTING].
4eaf3f64 5301 */
9adb62a5 5302void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
6811378e 5303{
f0c0b2b8
KH
5304 set_zonelist_order();
5305
6811378e 5306 if (system_state == SYSTEM_BOOTING) {
061f67bc 5307 build_all_zonelists_init();
6811378e 5308 } else {
e9959f0f 5309#ifdef CONFIG_MEMORY_HOTPLUG
9adb62a5
JL
5310 if (zone)
5311 setup_zone_pageset(zone);
e9959f0f 5312#endif
dd1895e2
CS
5313 /* we have to stop all cpus to guarantee there is no user
5314 of zonelist */
3f906ba2 5315 stop_machine_cpuslocked(__build_all_zonelists, pgdat, NULL);
6811378e
YG
5316 /* cpuset refresh routine should be here */
5317 }
bd1e22b8 5318 vm_total_pages = nr_free_pagecache_pages();
9ef9acb0
MG
5319 /*
5320 * Disable grouping by mobility if the number of pages in the
5321 * system is too low to allow the mechanism to work. It would be
5322 * more accurate, but expensive to check per-zone. This check is
5323 * made on memory-hotadd so a system can start with mobility
5324 * disabled and enable it later
5325 */
d9c23400 5326 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
9ef9acb0
MG
5327 page_group_by_mobility_disabled = 1;
5328 else
5329 page_group_by_mobility_disabled = 0;
5330
756a025f
JP
5331 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5332 nr_online_nodes,
5333 zonelist_order_name[current_zonelist_order],
5334 page_group_by_mobility_disabled ? "off" : "on",
5335 vm_total_pages);
f0c0b2b8 5336#ifdef CONFIG_NUMA
f88dfff5 5337 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
f0c0b2b8 5338#endif
1da177e4
LT
5339}
5340
1da177e4
LT
5341/*
5342 * Initially all pages are reserved - free ones are freed
5343 * up by free_all_bootmem() once the early boot process is
5344 * done. Non-atomic initialization, single-pass.
5345 */
c09b4240 5346void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
a2f3aa02 5347 unsigned long start_pfn, enum memmap_context context)
1da177e4 5348{
4b94ffdc 5349 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
29751f69 5350 unsigned long end_pfn = start_pfn + size;
4b94ffdc 5351 pg_data_t *pgdat = NODE_DATA(nid);
29751f69 5352 unsigned long pfn;
3a80a7fa 5353 unsigned long nr_initialised = 0;
342332e6
TI
5354#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5355 struct memblock_region *r = NULL, *tmp;
5356#endif
1da177e4 5357
22b31eec
HD
5358 if (highest_memmap_pfn < end_pfn - 1)
5359 highest_memmap_pfn = end_pfn - 1;
5360
4b94ffdc
DW
5361 /*
5362 * Honor reservation requested by the driver for this ZONE_DEVICE
5363 * memory
5364 */
5365 if (altmap && start_pfn == altmap->base_pfn)
5366 start_pfn += altmap->reserve;
5367
cbe8dd4a 5368 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
a2f3aa02 5369 /*
b72d0ffb
AM
5370 * There can be holes in boot-time mem_map[]s handed to this
5371 * function. They do not exist on hotplugged memory.
a2f3aa02 5372 */
b72d0ffb
AM
5373 if (context != MEMMAP_EARLY)
5374 goto not_early;
5375
b92df1de
PB
5376 if (!early_pfn_valid(pfn)) {
5377#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5378 /*
5379 * Skip to the pfn preceding the next valid one (or
5380 * end_pfn), such that we hit a valid pfn (or end_pfn)
5381 * on our next iteration of the loop.
5382 */
5383 pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1;
5384#endif
b72d0ffb 5385 continue;
b92df1de 5386 }
b72d0ffb
AM
5387 if (!early_pfn_in_nid(pfn, nid))
5388 continue;
5389 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5390 break;
342332e6
TI
5391
5392#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
b72d0ffb
AM
5393 /*
5394 * Check given memblock attribute by firmware which can affect
5395 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5396 * mirrored, it's an overlapped memmap init. skip it.
5397 */
5398 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5399 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5400 for_each_memblock(memory, tmp)
5401 if (pfn < memblock_region_memory_end_pfn(tmp))
5402 break;
5403 r = tmp;
5404 }
5405 if (pfn >= memblock_region_memory_base_pfn(r) &&
5406 memblock_is_mirror(r)) {
5407 /* already initialized as NORMAL */
5408 pfn = memblock_region_memory_end_pfn(r);
5409 continue;
342332e6 5410 }
a2f3aa02 5411 }
b72d0ffb 5412#endif
ac5d2539 5413
b72d0ffb 5414not_early:
ac5d2539
MG
5415 /*
5416 * Mark the block movable so that blocks are reserved for
5417 * movable at startup. This will force kernel allocations
5418 * to reserve their blocks rather than leaking throughout
5419 * the address space during boot when many long-lived
974a786e 5420 * kernel allocations are made.
ac5d2539
MG
5421 *
5422 * bitmap is created for zone's valid pfn range. but memmap
5423 * can be created for invalid pages (for alignment)
5424 * check here not to call set_pageblock_migratetype() against
5425 * pfn out of zone.
5426 */
5427 if (!(pfn & (pageblock_nr_pages - 1))) {
5428 struct page *page = pfn_to_page(pfn);
5429
5430 __init_single_page(page, pfn, zone, nid);
5431 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5432 } else {
5433 __init_single_pfn(pfn, zone, nid);
5434 }
1da177e4
LT
5435 }
5436}
5437
1e548deb 5438static void __meminit zone_init_free_lists(struct zone *zone)
1da177e4 5439{
7aeb09f9 5440 unsigned int order, t;
b2a0ac88
MG
5441 for_each_migratetype_order(order, t) {
5442 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1da177e4
LT
5443 zone->free_area[order].nr_free = 0;
5444 }
5445}
5446
5447#ifndef __HAVE_ARCH_MEMMAP_INIT
5448#define memmap_init(size, nid, zone, start_pfn) \
a2f3aa02 5449 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
1da177e4
LT
5450#endif
5451
7cd2b0a3 5452static int zone_batchsize(struct zone *zone)
e7c8d5c9 5453{
3a6be87f 5454#ifdef CONFIG_MMU
e7c8d5c9
CL
5455 int batch;
5456
5457 /*
5458 * The per-cpu-pages pools are set to around 1000th of the
ba56e91c 5459 * size of the zone. But no more than 1/2 of a meg.
e7c8d5c9
CL
5460 *
5461 * OK, so we don't know how big the cache is. So guess.
5462 */
b40da049 5463 batch = zone->managed_pages / 1024;
ba56e91c
SR
5464 if (batch * PAGE_SIZE > 512 * 1024)
5465 batch = (512 * 1024) / PAGE_SIZE;
e7c8d5c9
CL
5466 batch /= 4; /* We effectively *= 4 below */
5467 if (batch < 1)
5468 batch = 1;
5469
5470 /*
0ceaacc9
NP
5471 * Clamp the batch to a 2^n - 1 value. Having a power
5472 * of 2 value was found to be more likely to have
5473 * suboptimal cache aliasing properties in some cases.
e7c8d5c9 5474 *
0ceaacc9
NP
5475 * For example if 2 tasks are alternately allocating
5476 * batches of pages, one task can end up with a lot
5477 * of pages of one half of the possible page colors
5478 * and the other with pages of the other colors.
e7c8d5c9 5479 */
9155203a 5480 batch = rounddown_pow_of_two(batch + batch/2) - 1;
ba56e91c 5481
e7c8d5c9 5482 return batch;
3a6be87f
DH
5483
5484#else
5485 /* The deferral and batching of frees should be suppressed under NOMMU
5486 * conditions.
5487 *
5488 * The problem is that NOMMU needs to be able to allocate large chunks
5489 * of contiguous memory as there's no hardware page translation to
5490 * assemble apparent contiguous memory from discontiguous pages.
5491 *
5492 * Queueing large contiguous runs of pages for batching, however,
5493 * causes the pages to actually be freed in smaller chunks. As there
5494 * can be a significant delay between the individual batches being
5495 * recycled, this leads to the once large chunks of space being
5496 * fragmented and becoming unavailable for high-order allocations.
5497 */
5498 return 0;
5499#endif
e7c8d5c9
CL
5500}
5501
8d7a8fa9
CS
5502/*
5503 * pcp->high and pcp->batch values are related and dependent on one another:
5504 * ->batch must never be higher then ->high.
5505 * The following function updates them in a safe manner without read side
5506 * locking.
5507 *
5508 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5509 * those fields changing asynchronously (acording the the above rule).
5510 *
5511 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5512 * outside of boot time (or some other assurance that no concurrent updaters
5513 * exist).
5514 */
5515static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5516 unsigned long batch)
5517{
5518 /* start with a fail safe value for batch */
5519 pcp->batch = 1;
5520 smp_wmb();
5521
5522 /* Update high, then batch, in order */
5523 pcp->high = high;
5524 smp_wmb();
5525
5526 pcp->batch = batch;
5527}
5528
3664033c 5529/* a companion to pageset_set_high() */
4008bab7
CS
5530static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5531{
8d7a8fa9 5532 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4008bab7
CS
5533}
5534
88c90dbc 5535static void pageset_init(struct per_cpu_pageset *p)
2caaad41
CL
5536{
5537 struct per_cpu_pages *pcp;
5f8dcc21 5538 int migratetype;
2caaad41 5539
1c6fe946
MD
5540 memset(p, 0, sizeof(*p));
5541
3dfa5721 5542 pcp = &p->pcp;
2caaad41 5543 pcp->count = 0;
5f8dcc21
MG
5544 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5545 INIT_LIST_HEAD(&pcp->lists[migratetype]);
2caaad41
CL
5546}
5547
88c90dbc
CS
5548static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5549{
5550 pageset_init(p);
5551 pageset_set_batch(p, batch);
5552}
5553
8ad4b1fb 5554/*
3664033c 5555 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
8ad4b1fb
RS
5556 * to the value high for the pageset p.
5557 */
3664033c 5558static void pageset_set_high(struct per_cpu_pageset *p,
8ad4b1fb
RS
5559 unsigned long high)
5560{
8d7a8fa9
CS
5561 unsigned long batch = max(1UL, high / 4);
5562 if ((high / 4) > (PAGE_SHIFT * 8))
5563 batch = PAGE_SHIFT * 8;
8ad4b1fb 5564
8d7a8fa9 5565 pageset_update(&p->pcp, high, batch);
8ad4b1fb
RS
5566}
5567
7cd2b0a3
DR
5568static void pageset_set_high_and_batch(struct zone *zone,
5569 struct per_cpu_pageset *pcp)
56cef2b8 5570{
56cef2b8 5571 if (percpu_pagelist_fraction)
3664033c 5572 pageset_set_high(pcp,
56cef2b8
CS
5573 (zone->managed_pages /
5574 percpu_pagelist_fraction));
5575 else
5576 pageset_set_batch(pcp, zone_batchsize(zone));
5577}
5578
169f6c19
CS
5579static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5580{
5581 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5582
5583 pageset_init(pcp);
5584 pageset_set_high_and_batch(zone, pcp);
5585}
5586
4ed7e022 5587static void __meminit setup_zone_pageset(struct zone *zone)
319774e2
WF
5588{
5589 int cpu;
319774e2 5590 zone->pageset = alloc_percpu(struct per_cpu_pageset);
56cef2b8
CS
5591 for_each_possible_cpu(cpu)
5592 zone_pageset_init(zone, cpu);
319774e2
WF
5593}
5594
2caaad41 5595/*
99dcc3e5
CL
5596 * Allocate per cpu pagesets and initialize them.
5597 * Before this call only boot pagesets were available.
e7c8d5c9 5598 */
99dcc3e5 5599void __init setup_per_cpu_pageset(void)
e7c8d5c9 5600{
b4911ea2 5601 struct pglist_data *pgdat;
99dcc3e5 5602 struct zone *zone;
e7c8d5c9 5603
319774e2
WF
5604 for_each_populated_zone(zone)
5605 setup_zone_pageset(zone);
b4911ea2
MG
5606
5607 for_each_online_pgdat(pgdat)
5608 pgdat->per_cpu_nodestats =
5609 alloc_percpu(struct per_cpu_nodestat);
e7c8d5c9
CL
5610}
5611
c09b4240 5612static __meminit void zone_pcp_init(struct zone *zone)
ed8ece2e 5613{
99dcc3e5
CL
5614 /*
5615 * per cpu subsystem is not up at this point. The following code
5616 * relies on the ability of the linker to provide the
5617 * offset of a (static) per cpu variable into the per cpu area.
5618 */
5619 zone->pageset = &boot_pageset;
ed8ece2e 5620
b38a8725 5621 if (populated_zone(zone))
99dcc3e5
CL
5622 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5623 zone->name, zone->present_pages,
5624 zone_batchsize(zone));
ed8ece2e
DH
5625}
5626
dc0bbf3b 5627void __meminit init_currently_empty_zone(struct zone *zone,
718127cc 5628 unsigned long zone_start_pfn,
b171e409 5629 unsigned long size)
ed8ece2e
DH
5630{
5631 struct pglist_data *pgdat = zone->zone_pgdat;
9dcb8b68 5632
ed8ece2e
DH
5633 pgdat->nr_zones = zone_idx(zone) + 1;
5634
ed8ece2e
DH
5635 zone->zone_start_pfn = zone_start_pfn;
5636
708614e6
MG
5637 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5638 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5639 pgdat->node_id,
5640 (unsigned long)zone_idx(zone),
5641 zone_start_pfn, (zone_start_pfn + size));
5642
1e548deb 5643 zone_init_free_lists(zone);
9dcb8b68 5644 zone->initialized = 1;
ed8ece2e
DH
5645}
5646
0ee332c1 5647#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
c713216d 5648#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
8a942fde 5649
c713216d
MG
5650/*
5651 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
c713216d 5652 */
8a942fde
MG
5653int __meminit __early_pfn_to_nid(unsigned long pfn,
5654 struct mminit_pfnnid_cache *state)
c713216d 5655{
c13291a5 5656 unsigned long start_pfn, end_pfn;
e76b63f8 5657 int nid;
7c243c71 5658
8a942fde
MG
5659 if (state->last_start <= pfn && pfn < state->last_end)
5660 return state->last_nid;
c713216d 5661
e76b63f8
YL
5662 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5663 if (nid != -1) {
8a942fde
MG
5664 state->last_start = start_pfn;
5665 state->last_end = end_pfn;
5666 state->last_nid = nid;
e76b63f8
YL
5667 }
5668
5669 return nid;
c713216d
MG
5670}
5671#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5672
c713216d 5673/**
6782832e 5674 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
88ca3b94 5675 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
6782832e 5676 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
c713216d 5677 *
7d018176
ZZ
5678 * If an architecture guarantees that all ranges registered contain no holes
5679 * and may be freed, this this function may be used instead of calling
5680 * memblock_free_early_nid() manually.
c713216d 5681 */
c13291a5 5682void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
cc289894 5683{
c13291a5
TH
5684 unsigned long start_pfn, end_pfn;
5685 int i, this_nid;
edbe7d23 5686
c13291a5
TH
5687 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5688 start_pfn = min(start_pfn, max_low_pfn);
5689 end_pfn = min(end_pfn, max_low_pfn);
edbe7d23 5690
c13291a5 5691 if (start_pfn < end_pfn)
6782832e
SS
5692 memblock_free_early_nid(PFN_PHYS(start_pfn),
5693 (end_pfn - start_pfn) << PAGE_SHIFT,
5694 this_nid);
edbe7d23 5695 }
edbe7d23 5696}
edbe7d23 5697
c713216d
MG
5698/**
5699 * sparse_memory_present_with_active_regions - Call memory_present for each active range
88ca3b94 5700 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
c713216d 5701 *
7d018176
ZZ
5702 * If an architecture guarantees that all ranges registered contain no holes and may
5703 * be freed, this function may be used instead of calling memory_present() manually.
c713216d
MG
5704 */
5705void __init sparse_memory_present_with_active_regions(int nid)
5706{
c13291a5
TH
5707 unsigned long start_pfn, end_pfn;
5708 int i, this_nid;
c713216d 5709
c13291a5
TH
5710 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5711 memory_present(this_nid, start_pfn, end_pfn);
c713216d
MG
5712}
5713
5714/**
5715 * get_pfn_range_for_nid - Return the start and end page frames for a node
88ca3b94
RD
5716 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5717 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5718 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
c713216d
MG
5719 *
5720 * It returns the start and end page frame of a node based on information
7d018176 5721 * provided by memblock_set_node(). If called for a node
c713216d 5722 * with no available memory, a warning is printed and the start and end
88ca3b94 5723 * PFNs will be 0.
c713216d 5724 */
a3142c8e 5725void __meminit get_pfn_range_for_nid(unsigned int nid,
c713216d
MG
5726 unsigned long *start_pfn, unsigned long *end_pfn)
5727{
c13291a5 5728 unsigned long this_start_pfn, this_end_pfn;
c713216d 5729 int i;
c13291a5 5730
c713216d
MG
5731 *start_pfn = -1UL;
5732 *end_pfn = 0;
5733
c13291a5
TH
5734 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5735 *start_pfn = min(*start_pfn, this_start_pfn);
5736 *end_pfn = max(*end_pfn, this_end_pfn);
c713216d
MG
5737 }
5738
633c0666 5739 if (*start_pfn == -1UL)
c713216d 5740 *start_pfn = 0;
c713216d
MG
5741}
5742
2a1e274a
MG
5743/*
5744 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5745 * assumption is made that zones within a node are ordered in monotonic
5746 * increasing memory addresses so that the "highest" populated zone is used
5747 */
b69a7288 5748static void __init find_usable_zone_for_movable(void)
2a1e274a
MG
5749{
5750 int zone_index;
5751 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5752 if (zone_index == ZONE_MOVABLE)
5753 continue;
5754
5755 if (arch_zone_highest_possible_pfn[zone_index] >
5756 arch_zone_lowest_possible_pfn[zone_index])
5757 break;
5758 }
5759
5760 VM_BUG_ON(zone_index == -1);
5761 movable_zone = zone_index;
5762}
5763
5764/*
5765 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
25985edc 5766 * because it is sized independent of architecture. Unlike the other zones,
2a1e274a
MG
5767 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5768 * in each node depending on the size of each node and how evenly kernelcore
5769 * is distributed. This helper function adjusts the zone ranges
5770 * provided by the architecture for a given node by using the end of the
5771 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5772 * zones within a node are in order of monotonic increases memory addresses
5773 */
b69a7288 5774static void __meminit adjust_zone_range_for_zone_movable(int nid,
2a1e274a
MG
5775 unsigned long zone_type,
5776 unsigned long node_start_pfn,
5777 unsigned long node_end_pfn,
5778 unsigned long *zone_start_pfn,
5779 unsigned long *zone_end_pfn)
5780{
5781 /* Only adjust if ZONE_MOVABLE is on this node */
5782 if (zone_movable_pfn[nid]) {
5783 /* Size ZONE_MOVABLE */
5784 if (zone_type == ZONE_MOVABLE) {
5785 *zone_start_pfn = zone_movable_pfn[nid];
5786 *zone_end_pfn = min(node_end_pfn,
5787 arch_zone_highest_possible_pfn[movable_zone]);
5788
e506b996
XQ
5789 /* Adjust for ZONE_MOVABLE starting within this range */
5790 } else if (!mirrored_kernelcore &&
5791 *zone_start_pfn < zone_movable_pfn[nid] &&
5792 *zone_end_pfn > zone_movable_pfn[nid]) {
5793 *zone_end_pfn = zone_movable_pfn[nid];
5794
2a1e274a
MG
5795 /* Check if this whole range is within ZONE_MOVABLE */
5796 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5797 *zone_start_pfn = *zone_end_pfn;
5798 }
5799}
5800
c713216d
MG
5801/*
5802 * Return the number of pages a zone spans in a node, including holes
5803 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5804 */
6ea6e688 5805static unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d 5806 unsigned long zone_type,
7960aedd
ZY
5807 unsigned long node_start_pfn,
5808 unsigned long node_end_pfn,
d91749c1
TI
5809 unsigned long *zone_start_pfn,
5810 unsigned long *zone_end_pfn,
c713216d
MG
5811 unsigned long *ignored)
5812{
b5685e92 5813 /* When hotadd a new node from cpu_up(), the node should be empty */
f9126ab9
XQ
5814 if (!node_start_pfn && !node_end_pfn)
5815 return 0;
5816
7960aedd 5817 /* Get the start and end of the zone */
d91749c1
TI
5818 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5819 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2a1e274a
MG
5820 adjust_zone_range_for_zone_movable(nid, zone_type,
5821 node_start_pfn, node_end_pfn,
d91749c1 5822 zone_start_pfn, zone_end_pfn);
c713216d
MG
5823
5824 /* Check that this node has pages within the zone's required range */
d91749c1 5825 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
c713216d
MG
5826 return 0;
5827
5828 /* Move the zone boundaries inside the node if necessary */
d91749c1
TI
5829 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5830 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
c713216d
MG
5831
5832 /* Return the spanned pages */
d91749c1 5833 return *zone_end_pfn - *zone_start_pfn;
c713216d
MG
5834}
5835
5836/*
5837 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
88ca3b94 5838 * then all holes in the requested range will be accounted for.
c713216d 5839 */
32996250 5840unsigned long __meminit __absent_pages_in_range(int nid,
c713216d
MG
5841 unsigned long range_start_pfn,
5842 unsigned long range_end_pfn)
5843{
96e907d1
TH
5844 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5845 unsigned long start_pfn, end_pfn;
5846 int i;
c713216d 5847
96e907d1
TH
5848 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5849 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5850 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5851 nr_absent -= end_pfn - start_pfn;
c713216d 5852 }
96e907d1 5853 return nr_absent;
c713216d
MG
5854}
5855
5856/**
5857 * absent_pages_in_range - Return number of page frames in holes within a range
5858 * @start_pfn: The start PFN to start searching for holes
5859 * @end_pfn: The end PFN to stop searching for holes
5860 *
88ca3b94 5861 * It returns the number of pages frames in memory holes within a range.
c713216d
MG
5862 */
5863unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5864 unsigned long end_pfn)
5865{
5866 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5867}
5868
5869/* Return the number of page frames in holes in a zone on a node */
6ea6e688 5870static unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d 5871 unsigned long zone_type,
7960aedd
ZY
5872 unsigned long node_start_pfn,
5873 unsigned long node_end_pfn,
c713216d
MG
5874 unsigned long *ignored)
5875{
96e907d1
TH
5876 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5877 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
9c7cd687 5878 unsigned long zone_start_pfn, zone_end_pfn;
342332e6 5879 unsigned long nr_absent;
9c7cd687 5880
b5685e92 5881 /* When hotadd a new node from cpu_up(), the node should be empty */
f9126ab9
XQ
5882 if (!node_start_pfn && !node_end_pfn)
5883 return 0;
5884
96e907d1
TH
5885 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5886 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
9c7cd687 5887
2a1e274a
MG
5888 adjust_zone_range_for_zone_movable(nid, zone_type,
5889 node_start_pfn, node_end_pfn,
5890 &zone_start_pfn, &zone_end_pfn);
342332e6
TI
5891 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5892
5893 /*
5894 * ZONE_MOVABLE handling.
5895 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5896 * and vice versa.
5897 */
e506b996
XQ
5898 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
5899 unsigned long start_pfn, end_pfn;
5900 struct memblock_region *r;
5901
5902 for_each_memblock(memory, r) {
5903 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5904 zone_start_pfn, zone_end_pfn);
5905 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5906 zone_start_pfn, zone_end_pfn);
5907
5908 if (zone_type == ZONE_MOVABLE &&
5909 memblock_is_mirror(r))
5910 nr_absent += end_pfn - start_pfn;
5911
5912 if (zone_type == ZONE_NORMAL &&
5913 !memblock_is_mirror(r))
5914 nr_absent += end_pfn - start_pfn;
342332e6
TI
5915 }
5916 }
5917
5918 return nr_absent;
c713216d 5919}
0e0b864e 5920
0ee332c1 5921#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6ea6e688 5922static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d 5923 unsigned long zone_type,
7960aedd
ZY
5924 unsigned long node_start_pfn,
5925 unsigned long node_end_pfn,
d91749c1
TI
5926 unsigned long *zone_start_pfn,
5927 unsigned long *zone_end_pfn,
c713216d
MG
5928 unsigned long *zones_size)
5929{
d91749c1
TI
5930 unsigned int zone;
5931
5932 *zone_start_pfn = node_start_pfn;
5933 for (zone = 0; zone < zone_type; zone++)
5934 *zone_start_pfn += zones_size[zone];
5935
5936 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5937
c713216d
MG
5938 return zones_size[zone_type];
5939}
5940
6ea6e688 5941static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d 5942 unsigned long zone_type,
7960aedd
ZY
5943 unsigned long node_start_pfn,
5944 unsigned long node_end_pfn,
c713216d
MG
5945 unsigned long *zholes_size)
5946{
5947 if (!zholes_size)
5948 return 0;
5949
5950 return zholes_size[zone_type];
5951}
20e6926d 5952
0ee332c1 5953#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 5954
a3142c8e 5955static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
7960aedd
ZY
5956 unsigned long node_start_pfn,
5957 unsigned long node_end_pfn,
5958 unsigned long *zones_size,
5959 unsigned long *zholes_size)
c713216d 5960{
febd5949 5961 unsigned long realtotalpages = 0, totalpages = 0;
c713216d
MG
5962 enum zone_type i;
5963
febd5949
GZ
5964 for (i = 0; i < MAX_NR_ZONES; i++) {
5965 struct zone *zone = pgdat->node_zones + i;
d91749c1 5966 unsigned long zone_start_pfn, zone_end_pfn;
febd5949 5967 unsigned long size, real_size;
c713216d 5968
febd5949
GZ
5969 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5970 node_start_pfn,
5971 node_end_pfn,
d91749c1
TI
5972 &zone_start_pfn,
5973 &zone_end_pfn,
febd5949
GZ
5974 zones_size);
5975 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
7960aedd
ZY
5976 node_start_pfn, node_end_pfn,
5977 zholes_size);
d91749c1
TI
5978 if (size)
5979 zone->zone_start_pfn = zone_start_pfn;
5980 else
5981 zone->zone_start_pfn = 0;
febd5949
GZ
5982 zone->spanned_pages = size;
5983 zone->present_pages = real_size;
5984
5985 totalpages += size;
5986 realtotalpages += real_size;
5987 }
5988
5989 pgdat->node_spanned_pages = totalpages;
c713216d
MG
5990 pgdat->node_present_pages = realtotalpages;
5991 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5992 realtotalpages);
5993}
5994
835c134e
MG
5995#ifndef CONFIG_SPARSEMEM
5996/*
5997 * Calculate the size of the zone->blockflags rounded to an unsigned long
d9c23400
MG
5998 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5999 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
835c134e
MG
6000 * round what is now in bits to nearest long in bits, then return it in
6001 * bytes.
6002 */
7c45512d 6003static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
835c134e
MG
6004{
6005 unsigned long usemapsize;
6006
7c45512d 6007 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
d9c23400
MG
6008 usemapsize = roundup(zonesize, pageblock_nr_pages);
6009 usemapsize = usemapsize >> pageblock_order;
835c134e
MG
6010 usemapsize *= NR_PAGEBLOCK_BITS;
6011 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6012
6013 return usemapsize / 8;
6014}
6015
6016static void __init setup_usemap(struct pglist_data *pgdat,
7c45512d
LT
6017 struct zone *zone,
6018 unsigned long zone_start_pfn,
6019 unsigned long zonesize)
835c134e 6020{
7c45512d 6021 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
835c134e 6022 zone->pageblock_flags = NULL;
58a01a45 6023 if (usemapsize)
6782832e
SS
6024 zone->pageblock_flags =
6025 memblock_virt_alloc_node_nopanic(usemapsize,
6026 pgdat->node_id);
835c134e
MG
6027}
6028#else
7c45512d
LT
6029static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6030 unsigned long zone_start_pfn, unsigned long zonesize) {}
835c134e
MG
6031#endif /* CONFIG_SPARSEMEM */
6032
d9c23400 6033#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
ba72cb8c 6034
d9c23400 6035/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
15ca220e 6036void __paginginit set_pageblock_order(void)
d9c23400 6037{
955c1cd7
AM
6038 unsigned int order;
6039
d9c23400
MG
6040 /* Check that pageblock_nr_pages has not already been setup */
6041 if (pageblock_order)
6042 return;
6043
955c1cd7
AM
6044 if (HPAGE_SHIFT > PAGE_SHIFT)
6045 order = HUGETLB_PAGE_ORDER;
6046 else
6047 order = MAX_ORDER - 1;
6048
d9c23400
MG
6049 /*
6050 * Assume the largest contiguous order of interest is a huge page.
955c1cd7
AM
6051 * This value may be variable depending on boot parameters on IA64 and
6052 * powerpc.
d9c23400
MG
6053 */
6054 pageblock_order = order;
6055}
6056#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6057
ba72cb8c
MG
6058/*
6059 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
955c1cd7
AM
6060 * is unused as pageblock_order is set at compile-time. See
6061 * include/linux/pageblock-flags.h for the values of pageblock_order based on
6062 * the kernel config
ba72cb8c 6063 */
15ca220e 6064void __paginginit set_pageblock_order(void)
ba72cb8c 6065{
ba72cb8c 6066}
d9c23400
MG
6067
6068#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6069
01cefaef
JL
6070static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
6071 unsigned long present_pages)
6072{
6073 unsigned long pages = spanned_pages;
6074
6075 /*
6076 * Provide a more accurate estimation if there are holes within
6077 * the zone and SPARSEMEM is in use. If there are holes within the
6078 * zone, each populated memory region may cost us one or two extra
6079 * memmap pages due to alignment because memmap pages for each
89d790ab 6080 * populated regions may not be naturally aligned on page boundary.
01cefaef
JL
6081 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6082 */
6083 if (spanned_pages > present_pages + (present_pages >> 4) &&
6084 IS_ENABLED(CONFIG_SPARSEMEM))
6085 pages = present_pages;
6086
6087 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6088}
6089
1da177e4
LT
6090/*
6091 * Set up the zone data structures:
6092 * - mark all pages reserved
6093 * - mark all memory queues empty
6094 * - clear the memory bitmaps
6527af5d
MK
6095 *
6096 * NOTE: pgdat should get zeroed by caller.
1da177e4 6097 */
7f3eb55b 6098static void __paginginit free_area_init_core(struct pglist_data *pgdat)
1da177e4 6099{
2f1b6248 6100 enum zone_type j;
ed8ece2e 6101 int nid = pgdat->node_id;
1da177e4 6102
208d54e5 6103 pgdat_resize_init(pgdat);
8177a420
AA
6104#ifdef CONFIG_NUMA_BALANCING
6105 spin_lock_init(&pgdat->numabalancing_migrate_lock);
6106 pgdat->numabalancing_migrate_nr_pages = 0;
6107 pgdat->numabalancing_migrate_next_window = jiffies;
a3d0a918
KS
6108#endif
6109#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6110 spin_lock_init(&pgdat->split_queue_lock);
6111 INIT_LIST_HEAD(&pgdat->split_queue);
6112 pgdat->split_queue_len = 0;
8177a420 6113#endif
1da177e4 6114 init_waitqueue_head(&pgdat->kswapd_wait);
5515061d 6115 init_waitqueue_head(&pgdat->pfmemalloc_wait);
698b1b30
VB
6116#ifdef CONFIG_COMPACTION
6117 init_waitqueue_head(&pgdat->kcompactd_wait);
6118#endif
eefa864b 6119 pgdat_page_ext_init(pgdat);
a52633d8 6120 spin_lock_init(&pgdat->lru_lock);
a9dd0a83 6121 lruvec_init(node_lruvec(pgdat));
5f63b720 6122
385386cf
JW
6123 pgdat->per_cpu_nodestats = &boot_nodestats;
6124
1da177e4
LT
6125 for (j = 0; j < MAX_NR_ZONES; j++) {
6126 struct zone *zone = pgdat->node_zones + j;
9feedc9d 6127 unsigned long size, realsize, freesize, memmap_pages;
d91749c1 6128 unsigned long zone_start_pfn = zone->zone_start_pfn;
1da177e4 6129
febd5949
GZ
6130 size = zone->spanned_pages;
6131 realsize = freesize = zone->present_pages;
1da177e4 6132
0e0b864e 6133 /*
9feedc9d 6134 * Adjust freesize so that it accounts for how much memory
0e0b864e
MG
6135 * is used by this zone for memmap. This affects the watermark
6136 * and per-cpu initialisations
6137 */
01cefaef 6138 memmap_pages = calc_memmap_size(size, realsize);
ba914f48
ZH
6139 if (!is_highmem_idx(j)) {
6140 if (freesize >= memmap_pages) {
6141 freesize -= memmap_pages;
6142 if (memmap_pages)
6143 printk(KERN_DEBUG
6144 " %s zone: %lu pages used for memmap\n",
6145 zone_names[j], memmap_pages);
6146 } else
1170532b 6147 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
ba914f48
ZH
6148 zone_names[j], memmap_pages, freesize);
6149 }
0e0b864e 6150
6267276f 6151 /* Account for reserved pages */
9feedc9d
JL
6152 if (j == 0 && freesize > dma_reserve) {
6153 freesize -= dma_reserve;
d903ef9f 6154 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
6267276f 6155 zone_names[0], dma_reserve);
0e0b864e
MG
6156 }
6157
98d2b0eb 6158 if (!is_highmem_idx(j))
9feedc9d 6159 nr_kernel_pages += freesize;
01cefaef
JL
6160 /* Charge for highmem memmap if there are enough kernel pages */
6161 else if (nr_kernel_pages > memmap_pages * 2)
6162 nr_kernel_pages -= memmap_pages;
9feedc9d 6163 nr_all_pages += freesize;
1da177e4 6164
9feedc9d
JL
6165 /*
6166 * Set an approximate value for lowmem here, it will be adjusted
6167 * when the bootmem allocator frees pages into the buddy system.
6168 * And all highmem pages will be managed by the buddy system.
6169 */
6170 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
9614634f 6171#ifdef CONFIG_NUMA
d5f541ed 6172 zone->node = nid;
9614634f 6173#endif
1da177e4 6174 zone->name = zone_names[j];
a52633d8 6175 zone->zone_pgdat = pgdat;
1da177e4 6176 spin_lock_init(&zone->lock);
bdc8cb98 6177 zone_seqlock_init(zone);
ed8ece2e 6178 zone_pcp_init(zone);
81c0a2bb 6179
1da177e4
LT
6180 if (!size)
6181 continue;
6182
955c1cd7 6183 set_pageblock_order();
7c45512d 6184 setup_usemap(pgdat, zone, zone_start_pfn, size);
dc0bbf3b 6185 init_currently_empty_zone(zone, zone_start_pfn, size);
76cdd58e 6186 memmap_init(size, nid, j, zone_start_pfn);
1da177e4
LT
6187 }
6188}
6189
bd721ea7 6190static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
1da177e4 6191{
b0aeba74 6192 unsigned long __maybe_unused start = 0;
a1c34a3b
LA
6193 unsigned long __maybe_unused offset = 0;
6194
1da177e4
LT
6195 /* Skip empty nodes */
6196 if (!pgdat->node_spanned_pages)
6197 return;
6198
d41dee36 6199#ifdef CONFIG_FLAT_NODE_MEM_MAP
b0aeba74
TL
6200 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6201 offset = pgdat->node_start_pfn - start;
1da177e4
LT
6202 /* ia64 gets its own node_mem_map, before this, without bootmem */
6203 if (!pgdat->node_mem_map) {
b0aeba74 6204 unsigned long size, end;
d41dee36
AW
6205 struct page *map;
6206
e984bb43
BP
6207 /*
6208 * The zone's endpoints aren't required to be MAX_ORDER
6209 * aligned but the node_mem_map endpoints must be in order
6210 * for the buddy allocator to function correctly.
6211 */
108bcc96 6212 end = pgdat_end_pfn(pgdat);
e984bb43
BP
6213 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6214 size = (end - start) * sizeof(struct page);
6f167ec7
DH
6215 map = alloc_remap(pgdat->node_id, size);
6216 if (!map)
6782832e
SS
6217 map = memblock_virt_alloc_node_nopanic(size,
6218 pgdat->node_id);
a1c34a3b 6219 pgdat->node_mem_map = map + offset;
1da177e4 6220 }
12d810c1 6221#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
6222 /*
6223 * With no DISCONTIG, the global mem_map is just set as node 0's
6224 */
c713216d 6225 if (pgdat == NODE_DATA(0)) {
1da177e4 6226 mem_map = NODE_DATA(0)->node_mem_map;
a1c34a3b 6227#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
c713216d 6228 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
a1c34a3b 6229 mem_map -= offset;
0ee332c1 6230#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 6231 }
1da177e4 6232#endif
d41dee36 6233#endif /* CONFIG_FLAT_NODE_MEM_MAP */
1da177e4
LT
6234}
6235
9109fb7b
JW
6236void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
6237 unsigned long node_start_pfn, unsigned long *zholes_size)
1da177e4 6238{
9109fb7b 6239 pg_data_t *pgdat = NODE_DATA(nid);
7960aedd
ZY
6240 unsigned long start_pfn = 0;
6241 unsigned long end_pfn = 0;
9109fb7b 6242
88fdf75d 6243 /* pg_data_t should be reset to zero when it's allocated */
38087d9b 6244 WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
88fdf75d 6245
1da177e4
LT
6246 pgdat->node_id = nid;
6247 pgdat->node_start_pfn = node_start_pfn;
75ef7184 6248 pgdat->per_cpu_nodestats = NULL;
7960aedd
ZY
6249#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6250 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
8d29e18a 6251 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
4ada0c5a
ZL
6252 (u64)start_pfn << PAGE_SHIFT,
6253 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
d91749c1
TI
6254#else
6255 start_pfn = node_start_pfn;
7960aedd
ZY
6256#endif
6257 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6258 zones_size, zholes_size);
1da177e4
LT
6259
6260 alloc_node_mem_map(pgdat);
e8c27ac9
YL
6261#ifdef CONFIG_FLAT_NODE_MEM_MAP
6262 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6263 nid, (unsigned long)pgdat,
6264 (unsigned long)pgdat->node_mem_map);
6265#endif
1da177e4 6266
864b9a39 6267 reset_deferred_meminit(pgdat);
7f3eb55b 6268 free_area_init_core(pgdat);
1da177e4
LT
6269}
6270
0ee332c1 6271#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
418508c1
MS
6272
6273#if MAX_NUMNODES > 1
6274/*
6275 * Figure out the number of possible node ids.
6276 */
f9872caf 6277void __init setup_nr_node_ids(void)
418508c1 6278{
904a9553 6279 unsigned int highest;
418508c1 6280
904a9553 6281 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
418508c1
MS
6282 nr_node_ids = highest + 1;
6283}
418508c1
MS
6284#endif
6285
1e01979c
TH
6286/**
6287 * node_map_pfn_alignment - determine the maximum internode alignment
6288 *
6289 * This function should be called after node map is populated and sorted.
6290 * It calculates the maximum power of two alignment which can distinguish
6291 * all the nodes.
6292 *
6293 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6294 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6295 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6296 * shifted, 1GiB is enough and this function will indicate so.
6297 *
6298 * This is used to test whether pfn -> nid mapping of the chosen memory
6299 * model has fine enough granularity to avoid incorrect mapping for the
6300 * populated node map.
6301 *
6302 * Returns the determined alignment in pfn's. 0 if there is no alignment
6303 * requirement (single node).
6304 */
6305unsigned long __init node_map_pfn_alignment(void)
6306{
6307 unsigned long accl_mask = 0, last_end = 0;
c13291a5 6308 unsigned long start, end, mask;
1e01979c 6309 int last_nid = -1;
c13291a5 6310 int i, nid;
1e01979c 6311
c13291a5 6312 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
1e01979c
TH
6313 if (!start || last_nid < 0 || last_nid == nid) {
6314 last_nid = nid;
6315 last_end = end;
6316 continue;
6317 }
6318
6319 /*
6320 * Start with a mask granular enough to pin-point to the
6321 * start pfn and tick off bits one-by-one until it becomes
6322 * too coarse to separate the current node from the last.
6323 */
6324 mask = ~((1 << __ffs(start)) - 1);
6325 while (mask && last_end <= (start & (mask << 1)))
6326 mask <<= 1;
6327
6328 /* accumulate all internode masks */
6329 accl_mask |= mask;
6330 }
6331
6332 /* convert mask to number of pages */
6333 return ~accl_mask + 1;
6334}
6335
a6af2bc3 6336/* Find the lowest pfn for a node */
b69a7288 6337static unsigned long __init find_min_pfn_for_node(int nid)
c713216d 6338{
a6af2bc3 6339 unsigned long min_pfn = ULONG_MAX;
c13291a5
TH
6340 unsigned long start_pfn;
6341 int i;
1abbfb41 6342
c13291a5
TH
6343 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6344 min_pfn = min(min_pfn, start_pfn);
c713216d 6345
a6af2bc3 6346 if (min_pfn == ULONG_MAX) {
1170532b 6347 pr_warn("Could not find start_pfn for node %d\n", nid);
a6af2bc3
MG
6348 return 0;
6349 }
6350
6351 return min_pfn;
c713216d
MG
6352}
6353
6354/**
6355 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6356 *
6357 * It returns the minimum PFN based on information provided via
7d018176 6358 * memblock_set_node().
c713216d
MG
6359 */
6360unsigned long __init find_min_pfn_with_active_regions(void)
6361{
6362 return find_min_pfn_for_node(MAX_NUMNODES);
6363}
6364
37b07e41
LS
6365/*
6366 * early_calculate_totalpages()
6367 * Sum pages in active regions for movable zone.
4b0ef1fe 6368 * Populate N_MEMORY for calculating usable_nodes.
37b07e41 6369 */
484f51f8 6370static unsigned long __init early_calculate_totalpages(void)
7e63efef 6371{
7e63efef 6372 unsigned long totalpages = 0;
c13291a5
TH
6373 unsigned long start_pfn, end_pfn;
6374 int i, nid;
6375
6376 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6377 unsigned long pages = end_pfn - start_pfn;
7e63efef 6378
37b07e41
LS
6379 totalpages += pages;
6380 if (pages)
4b0ef1fe 6381 node_set_state(nid, N_MEMORY);
37b07e41 6382 }
b8af2941 6383 return totalpages;
7e63efef
MG
6384}
6385
2a1e274a
MG
6386/*
6387 * Find the PFN the Movable zone begins in each node. Kernel memory
6388 * is spread evenly between nodes as long as the nodes have enough
6389 * memory. When they don't, some nodes will have more kernelcore than
6390 * others
6391 */
b224ef85 6392static void __init find_zone_movable_pfns_for_nodes(void)
2a1e274a
MG
6393{
6394 int i, nid;
6395 unsigned long usable_startpfn;
6396 unsigned long kernelcore_node, kernelcore_remaining;
66918dcd 6397 /* save the state before borrow the nodemask */
4b0ef1fe 6398 nodemask_t saved_node_state = node_states[N_MEMORY];
37b07e41 6399 unsigned long totalpages = early_calculate_totalpages();
4b0ef1fe 6400 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
136199f0 6401 struct memblock_region *r;
b2f3eebe
TC
6402
6403 /* Need to find movable_zone earlier when movable_node is specified. */
6404 find_usable_zone_for_movable();
6405
6406 /*
6407 * If movable_node is specified, ignore kernelcore and movablecore
6408 * options.
6409 */
6410 if (movable_node_is_enabled()) {
136199f0
EM
6411 for_each_memblock(memory, r) {
6412 if (!memblock_is_hotpluggable(r))
b2f3eebe
TC
6413 continue;
6414
136199f0 6415 nid = r->nid;
b2f3eebe 6416
136199f0 6417 usable_startpfn = PFN_DOWN(r->base);
b2f3eebe
TC
6418 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6419 min(usable_startpfn, zone_movable_pfn[nid]) :
6420 usable_startpfn;
6421 }
6422
6423 goto out2;
6424 }
2a1e274a 6425
342332e6
TI
6426 /*
6427 * If kernelcore=mirror is specified, ignore movablecore option
6428 */
6429 if (mirrored_kernelcore) {
6430 bool mem_below_4gb_not_mirrored = false;
6431
6432 for_each_memblock(memory, r) {
6433 if (memblock_is_mirror(r))
6434 continue;
6435
6436 nid = r->nid;
6437
6438 usable_startpfn = memblock_region_memory_base_pfn(r);
6439
6440 if (usable_startpfn < 0x100000) {
6441 mem_below_4gb_not_mirrored = true;
6442 continue;
6443 }
6444
6445 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6446 min(usable_startpfn, zone_movable_pfn[nid]) :
6447 usable_startpfn;
6448 }
6449
6450 if (mem_below_4gb_not_mirrored)
6451 pr_warn("This configuration results in unmirrored kernel memory.");
6452
6453 goto out2;
6454 }
6455
7e63efef 6456 /*
b2f3eebe 6457 * If movablecore=nn[KMG] was specified, calculate what size of
7e63efef
MG
6458 * kernelcore that corresponds so that memory usable for
6459 * any allocation type is evenly spread. If both kernelcore
6460 * and movablecore are specified, then the value of kernelcore
6461 * will be used for required_kernelcore if it's greater than
6462 * what movablecore would have allowed.
6463 */
6464 if (required_movablecore) {
7e63efef
MG
6465 unsigned long corepages;
6466
6467 /*
6468 * Round-up so that ZONE_MOVABLE is at least as large as what
6469 * was requested by the user
6470 */
6471 required_movablecore =
6472 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
9fd745d4 6473 required_movablecore = min(totalpages, required_movablecore);
7e63efef
MG
6474 corepages = totalpages - required_movablecore;
6475
6476 required_kernelcore = max(required_kernelcore, corepages);
6477 }
6478
bde304bd
XQ
6479 /*
6480 * If kernelcore was not specified or kernelcore size is larger
6481 * than totalpages, there is no ZONE_MOVABLE.
6482 */
6483 if (!required_kernelcore || required_kernelcore >= totalpages)
66918dcd 6484 goto out;
2a1e274a
MG
6485
6486 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
2a1e274a
MG
6487 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6488
6489restart:
6490 /* Spread kernelcore memory as evenly as possible throughout nodes */
6491 kernelcore_node = required_kernelcore / usable_nodes;
4b0ef1fe 6492 for_each_node_state(nid, N_MEMORY) {
c13291a5
TH
6493 unsigned long start_pfn, end_pfn;
6494
2a1e274a
MG
6495 /*
6496 * Recalculate kernelcore_node if the division per node
6497 * now exceeds what is necessary to satisfy the requested
6498 * amount of memory for the kernel
6499 */
6500 if (required_kernelcore < kernelcore_node)
6501 kernelcore_node = required_kernelcore / usable_nodes;
6502
6503 /*
6504 * As the map is walked, we track how much memory is usable
6505 * by the kernel using kernelcore_remaining. When it is
6506 * 0, the rest of the node is usable by ZONE_MOVABLE
6507 */
6508 kernelcore_remaining = kernelcore_node;
6509
6510 /* Go through each range of PFNs within this node */
c13291a5 6511 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
2a1e274a
MG
6512 unsigned long size_pages;
6513
c13291a5 6514 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
2a1e274a
MG
6515 if (start_pfn >= end_pfn)
6516 continue;
6517
6518 /* Account for what is only usable for kernelcore */
6519 if (start_pfn < usable_startpfn) {
6520 unsigned long kernel_pages;
6521 kernel_pages = min(end_pfn, usable_startpfn)
6522 - start_pfn;
6523
6524 kernelcore_remaining -= min(kernel_pages,
6525 kernelcore_remaining);
6526 required_kernelcore -= min(kernel_pages,
6527 required_kernelcore);
6528
6529 /* Continue if range is now fully accounted */
6530 if (end_pfn <= usable_startpfn) {
6531
6532 /*
6533 * Push zone_movable_pfn to the end so
6534 * that if we have to rebalance
6535 * kernelcore across nodes, we will
6536 * not double account here
6537 */
6538 zone_movable_pfn[nid] = end_pfn;
6539 continue;
6540 }
6541 start_pfn = usable_startpfn;
6542 }
6543
6544 /*
6545 * The usable PFN range for ZONE_MOVABLE is from
6546 * start_pfn->end_pfn. Calculate size_pages as the
6547 * number of pages used as kernelcore
6548 */
6549 size_pages = end_pfn - start_pfn;
6550 if (size_pages > kernelcore_remaining)
6551 size_pages = kernelcore_remaining;
6552 zone_movable_pfn[nid] = start_pfn + size_pages;
6553
6554 /*
6555 * Some kernelcore has been met, update counts and
6556 * break if the kernelcore for this node has been
b8af2941 6557 * satisfied
2a1e274a
MG
6558 */
6559 required_kernelcore -= min(required_kernelcore,
6560 size_pages);
6561 kernelcore_remaining -= size_pages;
6562 if (!kernelcore_remaining)
6563 break;
6564 }
6565 }
6566
6567 /*
6568 * If there is still required_kernelcore, we do another pass with one
6569 * less node in the count. This will push zone_movable_pfn[nid] further
6570 * along on the nodes that still have memory until kernelcore is
b8af2941 6571 * satisfied
2a1e274a
MG
6572 */
6573 usable_nodes--;
6574 if (usable_nodes && required_kernelcore > usable_nodes)
6575 goto restart;
6576
b2f3eebe 6577out2:
2a1e274a
MG
6578 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6579 for (nid = 0; nid < MAX_NUMNODES; nid++)
6580 zone_movable_pfn[nid] =
6581 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
66918dcd 6582
20e6926d 6583out:
66918dcd 6584 /* restore the node_state */
4b0ef1fe 6585 node_states[N_MEMORY] = saved_node_state;
2a1e274a
MG
6586}
6587
4b0ef1fe
LJ
6588/* Any regular or high memory on that node ? */
6589static void check_for_memory(pg_data_t *pgdat, int nid)
37b07e41 6590{
37b07e41
LS
6591 enum zone_type zone_type;
6592
4b0ef1fe
LJ
6593 if (N_MEMORY == N_NORMAL_MEMORY)
6594 return;
6595
6596 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
37b07e41 6597 struct zone *zone = &pgdat->node_zones[zone_type];
b38a8725 6598 if (populated_zone(zone)) {
4b0ef1fe
LJ
6599 node_set_state(nid, N_HIGH_MEMORY);
6600 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6601 zone_type <= ZONE_NORMAL)
6602 node_set_state(nid, N_NORMAL_MEMORY);
d0048b0e
BL
6603 break;
6604 }
37b07e41 6605 }
37b07e41
LS
6606}
6607
c713216d
MG
6608/**
6609 * free_area_init_nodes - Initialise all pg_data_t and zone data
88ca3b94 6610 * @max_zone_pfn: an array of max PFNs for each zone
c713216d
MG
6611 *
6612 * This will call free_area_init_node() for each active node in the system.
7d018176 6613 * Using the page ranges provided by memblock_set_node(), the size of each
c713216d
MG
6614 * zone in each node and their holes is calculated. If the maximum PFN
6615 * between two adjacent zones match, it is assumed that the zone is empty.
6616 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6617 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6618 * starts where the previous one ended. For example, ZONE_DMA32 starts
6619 * at arch_max_dma_pfn.
6620 */
6621void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6622{
c13291a5
TH
6623 unsigned long start_pfn, end_pfn;
6624 int i, nid;
a6af2bc3 6625
c713216d
MG
6626 /* Record where the zone boundaries are */
6627 memset(arch_zone_lowest_possible_pfn, 0,
6628 sizeof(arch_zone_lowest_possible_pfn));
6629 memset(arch_zone_highest_possible_pfn, 0,
6630 sizeof(arch_zone_highest_possible_pfn));
90cae1fe
OH
6631
6632 start_pfn = find_min_pfn_with_active_regions();
6633
6634 for (i = 0; i < MAX_NR_ZONES; i++) {
2a1e274a
MG
6635 if (i == ZONE_MOVABLE)
6636 continue;
90cae1fe
OH
6637
6638 end_pfn = max(max_zone_pfn[i], start_pfn);
6639 arch_zone_lowest_possible_pfn[i] = start_pfn;
6640 arch_zone_highest_possible_pfn[i] = end_pfn;
6641
6642 start_pfn = end_pfn;
c713216d 6643 }
2a1e274a
MG
6644
6645 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6646 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
b224ef85 6647 find_zone_movable_pfns_for_nodes();
c713216d 6648
c713216d 6649 /* Print out the zone ranges */
f88dfff5 6650 pr_info("Zone ranges:\n");
2a1e274a
MG
6651 for (i = 0; i < MAX_NR_ZONES; i++) {
6652 if (i == ZONE_MOVABLE)
6653 continue;
f88dfff5 6654 pr_info(" %-8s ", zone_names[i]);
72f0ba02
DR
6655 if (arch_zone_lowest_possible_pfn[i] ==
6656 arch_zone_highest_possible_pfn[i])
f88dfff5 6657 pr_cont("empty\n");
72f0ba02 6658 else
8d29e18a
JG
6659 pr_cont("[mem %#018Lx-%#018Lx]\n",
6660 (u64)arch_zone_lowest_possible_pfn[i]
6661 << PAGE_SHIFT,
6662 ((u64)arch_zone_highest_possible_pfn[i]
a62e2f4f 6663 << PAGE_SHIFT) - 1);
2a1e274a
MG
6664 }
6665
6666 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
f88dfff5 6667 pr_info("Movable zone start for each node\n");
2a1e274a
MG
6668 for (i = 0; i < MAX_NUMNODES; i++) {
6669 if (zone_movable_pfn[i])
8d29e18a
JG
6670 pr_info(" Node %d: %#018Lx\n", i,
6671 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
2a1e274a 6672 }
c713216d 6673
f2d52fe5 6674 /* Print out the early node map */
f88dfff5 6675 pr_info("Early memory node ranges\n");
c13291a5 6676 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
8d29e18a
JG
6677 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6678 (u64)start_pfn << PAGE_SHIFT,
6679 ((u64)end_pfn << PAGE_SHIFT) - 1);
c713216d
MG
6680
6681 /* Initialise every node */
708614e6 6682 mminit_verify_pageflags_layout();
8ef82866 6683 setup_nr_node_ids();
c713216d
MG
6684 for_each_online_node(nid) {
6685 pg_data_t *pgdat = NODE_DATA(nid);
9109fb7b 6686 free_area_init_node(nid, NULL,
c713216d 6687 find_min_pfn_for_node(nid), NULL);
37b07e41
LS
6688
6689 /* Any memory on that node */
6690 if (pgdat->node_present_pages)
4b0ef1fe
LJ
6691 node_set_state(nid, N_MEMORY);
6692 check_for_memory(pgdat, nid);
c713216d
MG
6693 }
6694}
2a1e274a 6695
7e63efef 6696static int __init cmdline_parse_core(char *p, unsigned long *core)
2a1e274a
MG
6697{
6698 unsigned long long coremem;
6699 if (!p)
6700 return -EINVAL;
6701
6702 coremem = memparse(p, &p);
7e63efef 6703 *core = coremem >> PAGE_SHIFT;
2a1e274a 6704
7e63efef 6705 /* Paranoid check that UL is enough for the coremem value */
2a1e274a
MG
6706 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6707
6708 return 0;
6709}
ed7ed365 6710
7e63efef
MG
6711/*
6712 * kernelcore=size sets the amount of memory for use for allocations that
6713 * cannot be reclaimed or migrated.
6714 */
6715static int __init cmdline_parse_kernelcore(char *p)
6716{
342332e6
TI
6717 /* parse kernelcore=mirror */
6718 if (parse_option_str(p, "mirror")) {
6719 mirrored_kernelcore = true;
6720 return 0;
6721 }
6722
7e63efef
MG
6723 return cmdline_parse_core(p, &required_kernelcore);
6724}
6725
6726/*
6727 * movablecore=size sets the amount of memory for use for allocations that
6728 * can be reclaimed or migrated.
6729 */
6730static int __init cmdline_parse_movablecore(char *p)
6731{
6732 return cmdline_parse_core(p, &required_movablecore);
6733}
6734
ed7ed365 6735early_param("kernelcore", cmdline_parse_kernelcore);
7e63efef 6736early_param("movablecore", cmdline_parse_movablecore);
ed7ed365 6737
0ee332c1 6738#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 6739
c3d5f5f0
JL
6740void adjust_managed_page_count(struct page *page, long count)
6741{
6742 spin_lock(&managed_page_count_lock);
6743 page_zone(page)->managed_pages += count;
6744 totalram_pages += count;
3dcc0571
JL
6745#ifdef CONFIG_HIGHMEM
6746 if (PageHighMem(page))
6747 totalhigh_pages += count;
6748#endif
c3d5f5f0
JL
6749 spin_unlock(&managed_page_count_lock);
6750}
3dcc0571 6751EXPORT_SYMBOL(adjust_managed_page_count);
c3d5f5f0 6752
11199692 6753unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
69afade7 6754{
11199692
JL
6755 void *pos;
6756 unsigned long pages = 0;
69afade7 6757
11199692
JL
6758 start = (void *)PAGE_ALIGN((unsigned long)start);
6759 end = (void *)((unsigned long)end & PAGE_MASK);
6760 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
dbe67df4 6761 if ((unsigned int)poison <= 0xFF)
11199692
JL
6762 memset(pos, poison, PAGE_SIZE);
6763 free_reserved_page(virt_to_page(pos));
69afade7
JL
6764 }
6765
6766 if (pages && s)
adb1fe9a
JP
6767 pr_info("Freeing %s memory: %ldK\n",
6768 s, pages << (PAGE_SHIFT - 10));
69afade7
JL
6769
6770 return pages;
6771}
11199692 6772EXPORT_SYMBOL(free_reserved_area);
69afade7 6773
cfa11e08
JL
6774#ifdef CONFIG_HIGHMEM
6775void free_highmem_page(struct page *page)
6776{
6777 __free_reserved_page(page);
6778 totalram_pages++;
7b4b2a0d 6779 page_zone(page)->managed_pages++;
cfa11e08
JL
6780 totalhigh_pages++;
6781}
6782#endif
6783
7ee3d4e8
JL
6784
6785void __init mem_init_print_info(const char *str)
6786{
6787 unsigned long physpages, codesize, datasize, rosize, bss_size;
6788 unsigned long init_code_size, init_data_size;
6789
6790 physpages = get_num_physpages();
6791 codesize = _etext - _stext;
6792 datasize = _edata - _sdata;
6793 rosize = __end_rodata - __start_rodata;
6794 bss_size = __bss_stop - __bss_start;
6795 init_data_size = __init_end - __init_begin;
6796 init_code_size = _einittext - _sinittext;
6797
6798 /*
6799 * Detect special cases and adjust section sizes accordingly:
6800 * 1) .init.* may be embedded into .data sections
6801 * 2) .init.text.* may be out of [__init_begin, __init_end],
6802 * please refer to arch/tile/kernel/vmlinux.lds.S.
6803 * 3) .rodata.* may be embedded into .text or .data sections.
6804 */
6805#define adj_init_size(start, end, size, pos, adj) \
b8af2941
PK
6806 do { \
6807 if (start <= pos && pos < end && size > adj) \
6808 size -= adj; \
6809 } while (0)
7ee3d4e8
JL
6810
6811 adj_init_size(__init_begin, __init_end, init_data_size,
6812 _sinittext, init_code_size);
6813 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6814 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6815 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6816 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6817
6818#undef adj_init_size
6819
756a025f 6820 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7ee3d4e8 6821#ifdef CONFIG_HIGHMEM
756a025f 6822 ", %luK highmem"
7ee3d4e8 6823#endif
756a025f
JP
6824 "%s%s)\n",
6825 nr_free_pages() << (PAGE_SHIFT - 10),
6826 physpages << (PAGE_SHIFT - 10),
6827 codesize >> 10, datasize >> 10, rosize >> 10,
6828 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6829 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6830 totalcma_pages << (PAGE_SHIFT - 10),
7ee3d4e8 6831#ifdef CONFIG_HIGHMEM
756a025f 6832 totalhigh_pages << (PAGE_SHIFT - 10),
7ee3d4e8 6833#endif
756a025f 6834 str ? ", " : "", str ? str : "");
7ee3d4e8
JL
6835}
6836
0e0b864e 6837/**
88ca3b94
RD
6838 * set_dma_reserve - set the specified number of pages reserved in the first zone
6839 * @new_dma_reserve: The number of pages to mark reserved
0e0b864e 6840 *
013110a7 6841 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
0e0b864e
MG
6842 * In the DMA zone, a significant percentage may be consumed by kernel image
6843 * and other unfreeable allocations which can skew the watermarks badly. This
88ca3b94
RD
6844 * function may optionally be used to account for unfreeable pages in the
6845 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6846 * smaller per-cpu batchsize.
0e0b864e
MG
6847 */
6848void __init set_dma_reserve(unsigned long new_dma_reserve)
6849{
6850 dma_reserve = new_dma_reserve;
6851}
6852
1da177e4
LT
6853void __init free_area_init(unsigned long *zones_size)
6854{
9109fb7b 6855 free_area_init_node(0, zones_size,
1da177e4
LT
6856 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6857}
1da177e4 6858
005fd4bb 6859static int page_alloc_cpu_dead(unsigned int cpu)
1da177e4 6860{
1da177e4 6861
005fd4bb
SAS
6862 lru_add_drain_cpu(cpu);
6863 drain_pages(cpu);
9f8f2172 6864
005fd4bb
SAS
6865 /*
6866 * Spill the event counters of the dead processor
6867 * into the current processors event counters.
6868 * This artificially elevates the count of the current
6869 * processor.
6870 */
6871 vm_events_fold_cpu(cpu);
9f8f2172 6872
005fd4bb
SAS
6873 /*
6874 * Zero the differential counters of the dead processor
6875 * so that the vm statistics are consistent.
6876 *
6877 * This is only okay since the processor is dead and cannot
6878 * race with what we are doing.
6879 */
6880 cpu_vm_stats_fold(cpu);
6881 return 0;
1da177e4 6882}
1da177e4
LT
6883
6884void __init page_alloc_init(void)
6885{
005fd4bb
SAS
6886 int ret;
6887
6888 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
6889 "mm/page_alloc:dead", NULL,
6890 page_alloc_cpu_dead);
6891 WARN_ON(ret < 0);
1da177e4
LT
6892}
6893
cb45b0e9 6894/*
34b10060 6895 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
cb45b0e9
HA
6896 * or min_free_kbytes changes.
6897 */
6898static void calculate_totalreserve_pages(void)
6899{
6900 struct pglist_data *pgdat;
6901 unsigned long reserve_pages = 0;
2f6726e5 6902 enum zone_type i, j;
cb45b0e9
HA
6903
6904 for_each_online_pgdat(pgdat) {
281e3726
MG
6905
6906 pgdat->totalreserve_pages = 0;
6907
cb45b0e9
HA
6908 for (i = 0; i < MAX_NR_ZONES; i++) {
6909 struct zone *zone = pgdat->node_zones + i;
3484b2de 6910 long max = 0;
cb45b0e9
HA
6911
6912 /* Find valid and maximum lowmem_reserve in the zone */
6913 for (j = i; j < MAX_NR_ZONES; j++) {
6914 if (zone->lowmem_reserve[j] > max)
6915 max = zone->lowmem_reserve[j];
6916 }
6917
41858966
MG
6918 /* we treat the high watermark as reserved pages. */
6919 max += high_wmark_pages(zone);
cb45b0e9 6920
b40da049
JL
6921 if (max > zone->managed_pages)
6922 max = zone->managed_pages;
a8d01437 6923
281e3726 6924 pgdat->totalreserve_pages += max;
a8d01437 6925
cb45b0e9
HA
6926 reserve_pages += max;
6927 }
6928 }
6929 totalreserve_pages = reserve_pages;
6930}
6931
1da177e4
LT
6932/*
6933 * setup_per_zone_lowmem_reserve - called whenever
34b10060 6934 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
1da177e4
LT
6935 * has a correct pages reserved value, so an adequate number of
6936 * pages are left in the zone after a successful __alloc_pages().
6937 */
6938static void setup_per_zone_lowmem_reserve(void)
6939{
6940 struct pglist_data *pgdat;
2f6726e5 6941 enum zone_type j, idx;
1da177e4 6942
ec936fc5 6943 for_each_online_pgdat(pgdat) {
1da177e4
LT
6944 for (j = 0; j < MAX_NR_ZONES; j++) {
6945 struct zone *zone = pgdat->node_zones + j;
b40da049 6946 unsigned long managed_pages = zone->managed_pages;
1da177e4
LT
6947
6948 zone->lowmem_reserve[j] = 0;
6949
2f6726e5
CL
6950 idx = j;
6951 while (idx) {
1da177e4
LT
6952 struct zone *lower_zone;
6953
2f6726e5
CL
6954 idx--;
6955
1da177e4
LT
6956 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6957 sysctl_lowmem_reserve_ratio[idx] = 1;
6958
6959 lower_zone = pgdat->node_zones + idx;
b40da049 6960 lower_zone->lowmem_reserve[j] = managed_pages /
1da177e4 6961 sysctl_lowmem_reserve_ratio[idx];
b40da049 6962 managed_pages += lower_zone->managed_pages;
1da177e4
LT
6963 }
6964 }
6965 }
cb45b0e9
HA
6966
6967 /* update totalreserve_pages */
6968 calculate_totalreserve_pages();
1da177e4
LT
6969}
6970
cfd3da1e 6971static void __setup_per_zone_wmarks(void)
1da177e4
LT
6972{
6973 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6974 unsigned long lowmem_pages = 0;
6975 struct zone *zone;
6976 unsigned long flags;
6977
6978 /* Calculate total number of !ZONE_HIGHMEM pages */
6979 for_each_zone(zone) {
6980 if (!is_highmem(zone))
b40da049 6981 lowmem_pages += zone->managed_pages;
1da177e4
LT
6982 }
6983
6984 for_each_zone(zone) {
ac924c60
AM
6985 u64 tmp;
6986
1125b4e3 6987 spin_lock_irqsave(&zone->lock, flags);
b40da049 6988 tmp = (u64)pages_min * zone->managed_pages;
ac924c60 6989 do_div(tmp, lowmem_pages);
1da177e4
LT
6990 if (is_highmem(zone)) {
6991 /*
669ed175
NP
6992 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6993 * need highmem pages, so cap pages_min to a small
6994 * value here.
6995 *
41858966 6996 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
42ff2703 6997 * deltas control asynch page reclaim, and so should
669ed175 6998 * not be capped for highmem.
1da177e4 6999 */
90ae8d67 7000 unsigned long min_pages;
1da177e4 7001
b40da049 7002 min_pages = zone->managed_pages / 1024;
90ae8d67 7003 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
41858966 7004 zone->watermark[WMARK_MIN] = min_pages;
1da177e4 7005 } else {
669ed175
NP
7006 /*
7007 * If it's a lowmem zone, reserve a number of pages
1da177e4
LT
7008 * proportionate to the zone's size.
7009 */
41858966 7010 zone->watermark[WMARK_MIN] = tmp;
1da177e4
LT
7011 }
7012
795ae7a0
JW
7013 /*
7014 * Set the kswapd watermarks distance according to the
7015 * scale factor in proportion to available memory, but
7016 * ensure a minimum size on small systems.
7017 */
7018 tmp = max_t(u64, tmp >> 2,
7019 mult_frac(zone->managed_pages,
7020 watermark_scale_factor, 10000));
7021
7022 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
7023 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
49f223a9 7024
1125b4e3 7025 spin_unlock_irqrestore(&zone->lock, flags);
1da177e4 7026 }
cb45b0e9
HA
7027
7028 /* update totalreserve_pages */
7029 calculate_totalreserve_pages();
1da177e4
LT
7030}
7031
cfd3da1e
MG
7032/**
7033 * setup_per_zone_wmarks - called when min_free_kbytes changes
7034 * or when memory is hot-{added|removed}
7035 *
7036 * Ensures that the watermark[min,low,high] values for each zone are set
7037 * correctly with respect to min_free_kbytes.
7038 */
7039void setup_per_zone_wmarks(void)
7040{
7041 mutex_lock(&zonelists_mutex);
7042 __setup_per_zone_wmarks();
7043 mutex_unlock(&zonelists_mutex);
7044}
7045
1da177e4
LT
7046/*
7047 * Initialise min_free_kbytes.
7048 *
7049 * For small machines we want it small (128k min). For large machines
7050 * we want it large (64MB max). But it is not linear, because network
7051 * bandwidth does not increase linearly with machine size. We use
7052 *
b8af2941 7053 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
1da177e4
LT
7054 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
7055 *
7056 * which yields
7057 *
7058 * 16MB: 512k
7059 * 32MB: 724k
7060 * 64MB: 1024k
7061 * 128MB: 1448k
7062 * 256MB: 2048k
7063 * 512MB: 2896k
7064 * 1024MB: 4096k
7065 * 2048MB: 5792k
7066 * 4096MB: 8192k
7067 * 8192MB: 11584k
7068 * 16384MB: 16384k
7069 */
1b79acc9 7070int __meminit init_per_zone_wmark_min(void)
1da177e4
LT
7071{
7072 unsigned long lowmem_kbytes;
5f12733e 7073 int new_min_free_kbytes;
1da177e4
LT
7074
7075 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5f12733e
MH
7076 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7077
7078 if (new_min_free_kbytes > user_min_free_kbytes) {
7079 min_free_kbytes = new_min_free_kbytes;
7080 if (min_free_kbytes < 128)
7081 min_free_kbytes = 128;
7082 if (min_free_kbytes > 65536)
7083 min_free_kbytes = 65536;
7084 } else {
7085 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7086 new_min_free_kbytes, user_min_free_kbytes);
7087 }
bc75d33f 7088 setup_per_zone_wmarks();
a6cccdc3 7089 refresh_zone_stat_thresholds();
1da177e4 7090 setup_per_zone_lowmem_reserve();
6423aa81
JK
7091
7092#ifdef CONFIG_NUMA
7093 setup_min_unmapped_ratio();
7094 setup_min_slab_ratio();
7095#endif
7096
1da177e4
LT
7097 return 0;
7098}
bc22af74 7099core_initcall(init_per_zone_wmark_min)
1da177e4
LT
7100
7101/*
b8af2941 7102 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
1da177e4
LT
7103 * that we can call two helper functions whenever min_free_kbytes
7104 * changes.
7105 */
cccad5b9 7106int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8d65af78 7107 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 7108{
da8c757b
HP
7109 int rc;
7110
7111 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7112 if (rc)
7113 return rc;
7114
5f12733e
MH
7115 if (write) {
7116 user_min_free_kbytes = min_free_kbytes;
bc75d33f 7117 setup_per_zone_wmarks();
5f12733e 7118 }
1da177e4
LT
7119 return 0;
7120}
7121
795ae7a0
JW
7122int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7123 void __user *buffer, size_t *length, loff_t *ppos)
7124{
7125 int rc;
7126
7127 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7128 if (rc)
7129 return rc;
7130
7131 if (write)
7132 setup_per_zone_wmarks();
7133
7134 return 0;
7135}
7136
9614634f 7137#ifdef CONFIG_NUMA
6423aa81 7138static void setup_min_unmapped_ratio(void)
9614634f 7139{
6423aa81 7140 pg_data_t *pgdat;
9614634f 7141 struct zone *zone;
9614634f 7142
a5f5f91d 7143 for_each_online_pgdat(pgdat)
81cbcbc2 7144 pgdat->min_unmapped_pages = 0;
a5f5f91d 7145
9614634f 7146 for_each_zone(zone)
a5f5f91d 7147 zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
9614634f 7148 sysctl_min_unmapped_ratio) / 100;
9614634f 7149}
0ff38490 7150
6423aa81
JK
7151
7152int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8d65af78 7153 void __user *buffer, size_t *length, loff_t *ppos)
0ff38490 7154{
0ff38490
CL
7155 int rc;
7156
8d65af78 7157 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
0ff38490
CL
7158 if (rc)
7159 return rc;
7160
6423aa81
JK
7161 setup_min_unmapped_ratio();
7162
7163 return 0;
7164}
7165
7166static void setup_min_slab_ratio(void)
7167{
7168 pg_data_t *pgdat;
7169 struct zone *zone;
7170
a5f5f91d
MG
7171 for_each_online_pgdat(pgdat)
7172 pgdat->min_slab_pages = 0;
7173
0ff38490 7174 for_each_zone(zone)
a5f5f91d 7175 zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
0ff38490 7176 sysctl_min_slab_ratio) / 100;
6423aa81
JK
7177}
7178
7179int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7180 void __user *buffer, size_t *length, loff_t *ppos)
7181{
7182 int rc;
7183
7184 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7185 if (rc)
7186 return rc;
7187
7188 setup_min_slab_ratio();
7189
0ff38490
CL
7190 return 0;
7191}
9614634f
CL
7192#endif
7193
1da177e4
LT
7194/*
7195 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7196 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7197 * whenever sysctl_lowmem_reserve_ratio changes.
7198 *
7199 * The reserve ratio obviously has absolutely no relation with the
41858966 7200 * minimum watermarks. The lowmem reserve ratio can only make sense
1da177e4
LT
7201 * if in function of the boot time zone sizes.
7202 */
cccad5b9 7203int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8d65af78 7204 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 7205{
8d65af78 7206 proc_dointvec_minmax(table, write, buffer, length, ppos);
1da177e4
LT
7207 setup_per_zone_lowmem_reserve();
7208 return 0;
7209}
7210
8ad4b1fb
RS
7211/*
7212 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
b8af2941
PK
7213 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7214 * pagelist can have before it gets flushed back to buddy allocator.
8ad4b1fb 7215 */
cccad5b9 7216int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
8d65af78 7217 void __user *buffer, size_t *length, loff_t *ppos)
8ad4b1fb
RS
7218{
7219 struct zone *zone;
7cd2b0a3 7220 int old_percpu_pagelist_fraction;
8ad4b1fb
RS
7221 int ret;
7222
7cd2b0a3
DR
7223 mutex_lock(&pcp_batch_high_lock);
7224 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7225
8d65af78 7226 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7cd2b0a3
DR
7227 if (!write || ret < 0)
7228 goto out;
7229
7230 /* Sanity checking to avoid pcp imbalance */
7231 if (percpu_pagelist_fraction &&
7232 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7233 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7234 ret = -EINVAL;
7235 goto out;
7236 }
7237
7238 /* No change? */
7239 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7240 goto out;
c8e251fa 7241
364df0eb 7242 for_each_populated_zone(zone) {
7cd2b0a3
DR
7243 unsigned int cpu;
7244
22a7f12b 7245 for_each_possible_cpu(cpu)
7cd2b0a3
DR
7246 pageset_set_high_and_batch(zone,
7247 per_cpu_ptr(zone->pageset, cpu));
8ad4b1fb 7248 }
7cd2b0a3 7249out:
c8e251fa 7250 mutex_unlock(&pcp_batch_high_lock);
7cd2b0a3 7251 return ret;
8ad4b1fb
RS
7252}
7253
a9919c79 7254#ifdef CONFIG_NUMA
f034b5d4 7255int hashdist = HASHDIST_DEFAULT;
1da177e4 7256
1da177e4
LT
7257static int __init set_hashdist(char *str)
7258{
7259 if (!str)
7260 return 0;
7261 hashdist = simple_strtoul(str, &str, 0);
7262 return 1;
7263}
7264__setup("hashdist=", set_hashdist);
7265#endif
7266
f6f34b43
SD
7267#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7268/*
7269 * Returns the number of pages that arch has reserved but
7270 * is not known to alloc_large_system_hash().
7271 */
7272static unsigned long __init arch_reserved_kernel_pages(void)
7273{
7274 return 0;
7275}
7276#endif
7277
9017217b
PT
7278/*
7279 * Adaptive scale is meant to reduce sizes of hash tables on large memory
7280 * machines. As memory size is increased the scale is also increased but at
7281 * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory
7282 * quadruples the scale is increased by one, which means the size of hash table
7283 * only doubles, instead of quadrupling as well.
7284 * Because 32-bit systems cannot have large physical memory, where this scaling
7285 * makes sense, it is disabled on such platforms.
7286 */
7287#if __BITS_PER_LONG > 32
7288#define ADAPT_SCALE_BASE (64ul << 30)
7289#define ADAPT_SCALE_SHIFT 2
7290#define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
7291#endif
7292
1da177e4
LT
7293/*
7294 * allocate a large system hash table from bootmem
7295 * - it is assumed that the hash table must contain an exact power-of-2
7296 * quantity of entries
7297 * - limit is the number of hash buckets, not the total allocation size
7298 */
7299void *__init alloc_large_system_hash(const char *tablename,
7300 unsigned long bucketsize,
7301 unsigned long numentries,
7302 int scale,
7303 int flags,
7304 unsigned int *_hash_shift,
7305 unsigned int *_hash_mask,
31fe62b9
TB
7306 unsigned long low_limit,
7307 unsigned long high_limit)
1da177e4 7308{
31fe62b9 7309 unsigned long long max = high_limit;
1da177e4
LT
7310 unsigned long log2qty, size;
7311 void *table = NULL;
3749a8f0 7312 gfp_t gfp_flags;
1da177e4
LT
7313
7314 /* allow the kernel cmdline to have a say */
7315 if (!numentries) {
7316 /* round applicable memory size up to nearest megabyte */
04903664 7317 numentries = nr_kernel_pages;
f6f34b43 7318 numentries -= arch_reserved_kernel_pages();
a7e83318
JZ
7319
7320 /* It isn't necessary when PAGE_SIZE >= 1MB */
7321 if (PAGE_SHIFT < 20)
7322 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
1da177e4 7323
9017217b
PT
7324#if __BITS_PER_LONG > 32
7325 if (!high_limit) {
7326 unsigned long adapt;
7327
7328 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
7329 adapt <<= ADAPT_SCALE_SHIFT)
7330 scale++;
7331 }
7332#endif
7333
1da177e4
LT
7334 /* limit to 1 bucket per 2^scale bytes of low memory */
7335 if (scale > PAGE_SHIFT)
7336 numentries >>= (scale - PAGE_SHIFT);
7337 else
7338 numentries <<= (PAGE_SHIFT - scale);
9ab37b8f
PM
7339
7340 /* Make sure we've got at least a 0-order allocation.. */
2c85f51d
JB
7341 if (unlikely(flags & HASH_SMALL)) {
7342 /* Makes no sense without HASH_EARLY */
7343 WARN_ON(!(flags & HASH_EARLY));
7344 if (!(numentries >> *_hash_shift)) {
7345 numentries = 1UL << *_hash_shift;
7346 BUG_ON(!numentries);
7347 }
7348 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
9ab37b8f 7349 numentries = PAGE_SIZE / bucketsize;
1da177e4 7350 }
6e692ed3 7351 numentries = roundup_pow_of_two(numentries);
1da177e4
LT
7352
7353 /* limit allocation size to 1/16 total memory by default */
7354 if (max == 0) {
7355 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7356 do_div(max, bucketsize);
7357 }
074b8517 7358 max = min(max, 0x80000000ULL);
1da177e4 7359
31fe62b9
TB
7360 if (numentries < low_limit)
7361 numentries = low_limit;
1da177e4
LT
7362 if (numentries > max)
7363 numentries = max;
7364
f0d1b0b3 7365 log2qty = ilog2(numentries);
1da177e4 7366
3749a8f0
PT
7367 /*
7368 * memblock allocator returns zeroed memory already, so HASH_ZERO is
7369 * currently not used when HASH_EARLY is specified.
7370 */
7371 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
1da177e4
LT
7372 do {
7373 size = bucketsize << log2qty;
7374 if (flags & HASH_EARLY)
6782832e 7375 table = memblock_virt_alloc_nopanic(size, 0);
1da177e4 7376 else if (hashdist)
3749a8f0 7377 table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
1da177e4 7378 else {
1037b83b
ED
7379 /*
7380 * If bucketsize is not a power-of-two, we may free
a1dd268c
MG
7381 * some pages at the end of hash table which
7382 * alloc_pages_exact() automatically does
1037b83b 7383 */
264ef8a9 7384 if (get_order(size) < MAX_ORDER) {
3749a8f0
PT
7385 table = alloc_pages_exact(size, gfp_flags);
7386 kmemleak_alloc(table, size, 1, gfp_flags);
264ef8a9 7387 }
1da177e4
LT
7388 }
7389 } while (!table && size > PAGE_SIZE && --log2qty);
7390
7391 if (!table)
7392 panic("Failed to allocate %s hash table\n", tablename);
7393
1170532b
JP
7394 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7395 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
1da177e4
LT
7396
7397 if (_hash_shift)
7398 *_hash_shift = log2qty;
7399 if (_hash_mask)
7400 *_hash_mask = (1 << log2qty) - 1;
7401
7402 return table;
7403}
a117e66e 7404
a5d76b54 7405/*
80934513
MK
7406 * This function checks whether pageblock includes unmovable pages or not.
7407 * If @count is not zero, it is okay to include less @count unmovable pages
7408 *
b8af2941 7409 * PageLRU check without isolation or lru_lock could race so that
0efadf48
YX
7410 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7411 * check without lock_page also may miss some movable non-lru pages at
7412 * race condition. So you can't expect this function should be exact.
a5d76b54 7413 */
b023f468
WC
7414bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7415 bool skip_hwpoisoned_pages)
49ac8255
KH
7416{
7417 unsigned long pfn, iter, found;
47118af0
MN
7418 int mt;
7419
49ac8255
KH
7420 /*
7421 * For avoiding noise data, lru_add_drain_all() should be called
80934513 7422 * If ZONE_MOVABLE, the zone never contains unmovable pages
49ac8255
KH
7423 */
7424 if (zone_idx(zone) == ZONE_MOVABLE)
80934513 7425 return false;
47118af0
MN
7426 mt = get_pageblock_migratetype(page);
7427 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
80934513 7428 return false;
49ac8255
KH
7429
7430 pfn = page_to_pfn(page);
7431 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7432 unsigned long check = pfn + iter;
7433
29723fcc 7434 if (!pfn_valid_within(check))
49ac8255 7435 continue;
29723fcc 7436
49ac8255 7437 page = pfn_to_page(check);
c8721bbb
NH
7438
7439 /*
7440 * Hugepages are not in LRU lists, but they're movable.
7441 * We need not scan over tail pages bacause we don't
7442 * handle each tail page individually in migration.
7443 */
7444 if (PageHuge(page)) {
7445 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7446 continue;
7447 }
7448
97d255c8
MK
7449 /*
7450 * We can't use page_count without pin a page
7451 * because another CPU can free compound page.
7452 * This check already skips compound tails of THP
0139aa7b 7453 * because their page->_refcount is zero at all time.
97d255c8 7454 */
fe896d18 7455 if (!page_ref_count(page)) {
49ac8255
KH
7456 if (PageBuddy(page))
7457 iter += (1 << page_order(page)) - 1;
7458 continue;
7459 }
97d255c8 7460
b023f468
WC
7461 /*
7462 * The HWPoisoned page may be not in buddy system, and
7463 * page_count() is not 0.
7464 */
7465 if (skip_hwpoisoned_pages && PageHWPoison(page))
7466 continue;
7467
0efadf48
YX
7468 if (__PageMovable(page))
7469 continue;
7470
49ac8255
KH
7471 if (!PageLRU(page))
7472 found++;
7473 /*
6b4f7799
JW
7474 * If there are RECLAIMABLE pages, we need to check
7475 * it. But now, memory offline itself doesn't call
7476 * shrink_node_slabs() and it still to be fixed.
49ac8255
KH
7477 */
7478 /*
7479 * If the page is not RAM, page_count()should be 0.
7480 * we don't need more check. This is an _used_ not-movable page.
7481 *
7482 * The problematic thing here is PG_reserved pages. PG_reserved
7483 * is set to both of a memory hole page and a _used_ kernel
7484 * page at boot.
7485 */
7486 if (found > count)
80934513 7487 return true;
49ac8255 7488 }
80934513 7489 return false;
49ac8255
KH
7490}
7491
7492bool is_pageblock_removable_nolock(struct page *page)
7493{
656a0706
MH
7494 struct zone *zone;
7495 unsigned long pfn;
687875fb
MH
7496
7497 /*
7498 * We have to be careful here because we are iterating over memory
7499 * sections which are not zone aware so we might end up outside of
7500 * the zone but still within the section.
656a0706
MH
7501 * We have to take care about the node as well. If the node is offline
7502 * its NODE_DATA will be NULL - see page_zone.
687875fb 7503 */
656a0706
MH
7504 if (!node_online(page_to_nid(page)))
7505 return false;
7506
7507 zone = page_zone(page);
7508 pfn = page_to_pfn(page);
108bcc96 7509 if (!zone_spans_pfn(zone, pfn))
687875fb
MH
7510 return false;
7511
b023f468 7512 return !has_unmovable_pages(zone, page, 0, true);
a5d76b54 7513}
0c0e6195 7514
080fe206 7515#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
041d3a8c
MN
7516
7517static unsigned long pfn_max_align_down(unsigned long pfn)
7518{
7519 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7520 pageblock_nr_pages) - 1);
7521}
7522
7523static unsigned long pfn_max_align_up(unsigned long pfn)
7524{
7525 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7526 pageblock_nr_pages));
7527}
7528
041d3a8c 7529/* [start, end) must belong to a single zone. */
bb13ffeb
MG
7530static int __alloc_contig_migrate_range(struct compact_control *cc,
7531 unsigned long start, unsigned long end)
041d3a8c
MN
7532{
7533 /* This function is based on compact_zone() from compaction.c. */
beb51eaa 7534 unsigned long nr_reclaimed;
041d3a8c
MN
7535 unsigned long pfn = start;
7536 unsigned int tries = 0;
7537 int ret = 0;
7538
be49a6e1 7539 migrate_prep();
041d3a8c 7540
bb13ffeb 7541 while (pfn < end || !list_empty(&cc->migratepages)) {
041d3a8c
MN
7542 if (fatal_signal_pending(current)) {
7543 ret = -EINTR;
7544 break;
7545 }
7546
bb13ffeb
MG
7547 if (list_empty(&cc->migratepages)) {
7548 cc->nr_migratepages = 0;
edc2ca61 7549 pfn = isolate_migratepages_range(cc, pfn, end);
041d3a8c
MN
7550 if (!pfn) {
7551 ret = -EINTR;
7552 break;
7553 }
7554 tries = 0;
7555 } else if (++tries == 5) {
7556 ret = ret < 0 ? ret : -EBUSY;
7557 break;
7558 }
7559
beb51eaa
MK
7560 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7561 &cc->migratepages);
7562 cc->nr_migratepages -= nr_reclaimed;
02c6de8d 7563
9c620e2b 7564 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
e0b9daeb 7565 NULL, 0, cc->mode, MR_CMA);
041d3a8c 7566 }
2a6f5124
SP
7567 if (ret < 0) {
7568 putback_movable_pages(&cc->migratepages);
7569 return ret;
7570 }
7571 return 0;
041d3a8c
MN
7572}
7573
7574/**
7575 * alloc_contig_range() -- tries to allocate given range of pages
7576 * @start: start PFN to allocate
7577 * @end: one-past-the-last PFN to allocate
0815f3d8
MN
7578 * @migratetype: migratetype of the underlaying pageblocks (either
7579 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7580 * in range must have the same migratetype and it must
7581 * be either of the two.
ca96b625 7582 * @gfp_mask: GFP mask to use during compaction
041d3a8c
MN
7583 *
7584 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7585 * aligned, however it's the caller's responsibility to guarantee that
7586 * we are the only thread that changes migrate type of pageblocks the
7587 * pages fall in.
7588 *
7589 * The PFN range must belong to a single zone.
7590 *
7591 * Returns zero on success or negative error code. On success all
7592 * pages which PFN is in [start, end) are allocated for the caller and
7593 * need to be freed with free_contig_range().
7594 */
0815f3d8 7595int alloc_contig_range(unsigned long start, unsigned long end,
ca96b625 7596 unsigned migratetype, gfp_t gfp_mask)
041d3a8c 7597{
041d3a8c 7598 unsigned long outer_start, outer_end;
d00181b9
KS
7599 unsigned int order;
7600 int ret = 0;
041d3a8c 7601
bb13ffeb
MG
7602 struct compact_control cc = {
7603 .nr_migratepages = 0,
7604 .order = -1,
7605 .zone = page_zone(pfn_to_page(start)),
e0b9daeb 7606 .mode = MIGRATE_SYNC,
bb13ffeb 7607 .ignore_skip_hint = true,
7dea19f9 7608 .gfp_mask = current_gfp_context(gfp_mask),
bb13ffeb
MG
7609 };
7610 INIT_LIST_HEAD(&cc.migratepages);
7611
041d3a8c
MN
7612 /*
7613 * What we do here is we mark all pageblocks in range as
7614 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7615 * have different sizes, and due to the way page allocator
7616 * work, we align the range to biggest of the two pages so
7617 * that page allocator won't try to merge buddies from
7618 * different pageblocks and change MIGRATE_ISOLATE to some
7619 * other migration type.
7620 *
7621 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7622 * migrate the pages from an unaligned range (ie. pages that
7623 * we are interested in). This will put all the pages in
7624 * range back to page allocator as MIGRATE_ISOLATE.
7625 *
7626 * When this is done, we take the pages in range from page
7627 * allocator removing them from the buddy system. This way
7628 * page allocator will never consider using them.
7629 *
7630 * This lets us mark the pageblocks back as
7631 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7632 * aligned range but not in the unaligned, original range are
7633 * put back to page allocator so that buddy can use them.
7634 */
7635
7636 ret = start_isolate_page_range(pfn_max_align_down(start),
b023f468
WC
7637 pfn_max_align_up(end), migratetype,
7638 false);
041d3a8c 7639 if (ret)
86a595f9 7640 return ret;
041d3a8c 7641
8ef5849f
JK
7642 /*
7643 * In case of -EBUSY, we'd like to know which page causes problem.
7644 * So, just fall through. We will check it in test_pages_isolated().
7645 */
bb13ffeb 7646 ret = __alloc_contig_migrate_range(&cc, start, end);
8ef5849f 7647 if (ret && ret != -EBUSY)
041d3a8c
MN
7648 goto done;
7649
7650 /*
7651 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7652 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7653 * more, all pages in [start, end) are free in page allocator.
7654 * What we are going to do is to allocate all pages from
7655 * [start, end) (that is remove them from page allocator).
7656 *
7657 * The only problem is that pages at the beginning and at the
7658 * end of interesting range may be not aligned with pages that
7659 * page allocator holds, ie. they can be part of higher order
7660 * pages. Because of this, we reserve the bigger range and
7661 * once this is done free the pages we are not interested in.
7662 *
7663 * We don't have to hold zone->lock here because the pages are
7664 * isolated thus they won't get removed from buddy.
7665 */
7666
7667 lru_add_drain_all();
510f5507 7668 drain_all_pages(cc.zone);
041d3a8c
MN
7669
7670 order = 0;
7671 outer_start = start;
7672 while (!PageBuddy(pfn_to_page(outer_start))) {
7673 if (++order >= MAX_ORDER) {
8ef5849f
JK
7674 outer_start = start;
7675 break;
041d3a8c
MN
7676 }
7677 outer_start &= ~0UL << order;
7678 }
7679
8ef5849f
JK
7680 if (outer_start != start) {
7681 order = page_order(pfn_to_page(outer_start));
7682
7683 /*
7684 * outer_start page could be small order buddy page and
7685 * it doesn't include start page. Adjust outer_start
7686 * in this case to report failed page properly
7687 * on tracepoint in test_pages_isolated()
7688 */
7689 if (outer_start + (1UL << order) <= start)
7690 outer_start = start;
7691 }
7692
041d3a8c 7693 /* Make sure the range is really isolated. */
b023f468 7694 if (test_pages_isolated(outer_start, end, false)) {
75dddef3 7695 pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
dae803e1 7696 __func__, outer_start, end);
041d3a8c
MN
7697 ret = -EBUSY;
7698 goto done;
7699 }
7700
49f223a9 7701 /* Grab isolated pages from freelists. */
bb13ffeb 7702 outer_end = isolate_freepages_range(&cc, outer_start, end);
041d3a8c
MN
7703 if (!outer_end) {
7704 ret = -EBUSY;
7705 goto done;
7706 }
7707
7708 /* Free head and tail (if any) */
7709 if (start != outer_start)
7710 free_contig_range(outer_start, start - outer_start);
7711 if (end != outer_end)
7712 free_contig_range(end, outer_end - end);
7713
7714done:
7715 undo_isolate_page_range(pfn_max_align_down(start),
0815f3d8 7716 pfn_max_align_up(end), migratetype);
041d3a8c
MN
7717 return ret;
7718}
7719
7720void free_contig_range(unsigned long pfn, unsigned nr_pages)
7721{
bcc2b02f
MS
7722 unsigned int count = 0;
7723
7724 for (; nr_pages--; pfn++) {
7725 struct page *page = pfn_to_page(pfn);
7726
7727 count += page_count(page) != 1;
7728 __free_page(page);
7729 }
7730 WARN(count != 0, "%d pages are still in use!\n", count);
041d3a8c
MN
7731}
7732#endif
7733
4ed7e022 7734#ifdef CONFIG_MEMORY_HOTPLUG
0a647f38
CS
7735/*
7736 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7737 * page high values need to be recalulated.
7738 */
4ed7e022
JL
7739void __meminit zone_pcp_update(struct zone *zone)
7740{
0a647f38 7741 unsigned cpu;
c8e251fa 7742 mutex_lock(&pcp_batch_high_lock);
0a647f38 7743 for_each_possible_cpu(cpu)
169f6c19
CS
7744 pageset_set_high_and_batch(zone,
7745 per_cpu_ptr(zone->pageset, cpu));
c8e251fa 7746 mutex_unlock(&pcp_batch_high_lock);
4ed7e022
JL
7747}
7748#endif
7749
340175b7
JL
7750void zone_pcp_reset(struct zone *zone)
7751{
7752 unsigned long flags;
5a883813
MK
7753 int cpu;
7754 struct per_cpu_pageset *pset;
340175b7
JL
7755
7756 /* avoid races with drain_pages() */
7757 local_irq_save(flags);
7758 if (zone->pageset != &boot_pageset) {
5a883813
MK
7759 for_each_online_cpu(cpu) {
7760 pset = per_cpu_ptr(zone->pageset, cpu);
7761 drain_zonestat(zone, pset);
7762 }
340175b7
JL
7763 free_percpu(zone->pageset);
7764 zone->pageset = &boot_pageset;
7765 }
7766 local_irq_restore(flags);
7767}
7768
6dcd73d7 7769#ifdef CONFIG_MEMORY_HOTREMOVE
0c0e6195 7770/*
b9eb6319
JK
7771 * All pages in the range must be in a single zone and isolated
7772 * before calling this.
0c0e6195
KH
7773 */
7774void
7775__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7776{
7777 struct page *page;
7778 struct zone *zone;
7aeb09f9 7779 unsigned int order, i;
0c0e6195
KH
7780 unsigned long pfn;
7781 unsigned long flags;
7782 /* find the first valid pfn */
7783 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7784 if (pfn_valid(pfn))
7785 break;
7786 if (pfn == end_pfn)
7787 return;
2d070eab 7788 offline_mem_sections(pfn, end_pfn);
0c0e6195
KH
7789 zone = page_zone(pfn_to_page(pfn));
7790 spin_lock_irqsave(&zone->lock, flags);
7791 pfn = start_pfn;
7792 while (pfn < end_pfn) {
7793 if (!pfn_valid(pfn)) {
7794 pfn++;
7795 continue;
7796 }
7797 page = pfn_to_page(pfn);
b023f468
WC
7798 /*
7799 * The HWPoisoned page may be not in buddy system, and
7800 * page_count() is not 0.
7801 */
7802 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7803 pfn++;
7804 SetPageReserved(page);
7805 continue;
7806 }
7807
0c0e6195
KH
7808 BUG_ON(page_count(page));
7809 BUG_ON(!PageBuddy(page));
7810 order = page_order(page);
7811#ifdef CONFIG_DEBUG_VM
1170532b
JP
7812 pr_info("remove from free list %lx %d %lx\n",
7813 pfn, 1 << order, end_pfn);
0c0e6195
KH
7814#endif
7815 list_del(&page->lru);
7816 rmv_page_order(page);
7817 zone->free_area[order].nr_free--;
0c0e6195
KH
7818 for (i = 0; i < (1 << order); i++)
7819 SetPageReserved((page+i));
7820 pfn += (1 << order);
7821 }
7822 spin_unlock_irqrestore(&zone->lock, flags);
7823}
7824#endif
8d22ba1b 7825
8d22ba1b
WF
7826bool is_free_buddy_page(struct page *page)
7827{
7828 struct zone *zone = page_zone(page);
7829 unsigned long pfn = page_to_pfn(page);
7830 unsigned long flags;
7aeb09f9 7831 unsigned int order;
8d22ba1b
WF
7832
7833 spin_lock_irqsave(&zone->lock, flags);
7834 for (order = 0; order < MAX_ORDER; order++) {
7835 struct page *page_head = page - (pfn & ((1 << order) - 1));
7836
7837 if (PageBuddy(page_head) && page_order(page_head) >= order)
7838 break;
7839 }
7840 spin_unlock_irqrestore(&zone->lock, flags);
7841
7842 return order < MAX_ORDER;
7843}