]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - mm/page_alloc.c
mm: page_alloc: generalize the dirty balance reserve
[mirror_ubuntu-zesty-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>
6811378e 46#include <linux/stop_machine.h>
c713216d
MG
47#include <linux/sort.h>
48#include <linux/pfn.h>
3fcfab16 49#include <linux/backing-dev.h>
933e312e 50#include <linux/fault-inject.h>
a5d76b54 51#include <linux/page-isolation.h>
eefa864b 52#include <linux/page_ext.h>
3ac7fe5a 53#include <linux/debugobjects.h>
dbb1f81c 54#include <linux/kmemleak.h>
56de7263 55#include <linux/compaction.h>
0d3d062a 56#include <trace/events/kmem.h>
268bb0ce 57#include <linux/prefetch.h>
6e543d57 58#include <linux/mm_inline.h>
041d3a8c 59#include <linux/migrate.h>
e30825f1 60#include <linux/page_ext.h>
949f7ec5 61#include <linux/hugetlb.h>
8bd75c77 62#include <linux/sched/rt.h>
48c96a36 63#include <linux/page_owner.h>
0e1cc95b 64#include <linux/kthread.h>
1da177e4 65
7ee3d4e8 66#include <asm/sections.h>
1da177e4 67#include <asm/tlbflush.h>
ac924c60 68#include <asm/div64.h>
1da177e4
LT
69#include "internal.h"
70
c8e251fa
CS
71/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
72static DEFINE_MUTEX(pcp_batch_high_lock);
7cd2b0a3 73#define MIN_PERCPU_PAGELIST_FRACTION (8)
c8e251fa 74
72812019
LS
75#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
76DEFINE_PER_CPU(int, numa_node);
77EXPORT_PER_CPU_SYMBOL(numa_node);
78#endif
79
7aac7898
LS
80#ifdef CONFIG_HAVE_MEMORYLESS_NODES
81/*
82 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
83 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
84 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
85 * defined in <linux/topology.h>.
86 */
87DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
88EXPORT_PER_CPU_SYMBOL(_numa_mem_);
ad2c8144 89int _node_numa_mem_[MAX_NUMNODES];
7aac7898
LS
90#endif
91
1da177e4 92/*
13808910 93 * Array of node states.
1da177e4 94 */
13808910
CL
95nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
96 [N_POSSIBLE] = NODE_MASK_ALL,
97 [N_ONLINE] = { { [0] = 1UL } },
98#ifndef CONFIG_NUMA
99 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100#ifdef CONFIG_HIGHMEM
101 [N_HIGH_MEMORY] = { { [0] = 1UL } },
20b2f52b
LJ
102#endif
103#ifdef CONFIG_MOVABLE_NODE
104 [N_MEMORY] = { { [0] = 1UL } },
13808910
CL
105#endif
106 [N_CPU] = { { [0] = 1UL } },
107#endif /* NUMA */
108};
109EXPORT_SYMBOL(node_states);
110
c3d5f5f0
JL
111/* Protect totalram_pages and zone->managed_pages */
112static DEFINE_SPINLOCK(managed_page_count_lock);
113
6c231b7b 114unsigned long totalram_pages __read_mostly;
cb45b0e9 115unsigned long totalreserve_pages __read_mostly;
e48322ab 116unsigned long totalcma_pages __read_mostly;
ab8fabd4 117
1b76b02f 118int percpu_pagelist_fraction;
dcce284a 119gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
1da177e4 120
bb14c2c7
VB
121/*
122 * A cached value of the page's pageblock's migratetype, used when the page is
123 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
124 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
125 * Also the migratetype set in the page does not necessarily match the pcplist
126 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
127 * other index - this ensures that it will be put on the correct CMA freelist.
128 */
129static inline int get_pcppage_migratetype(struct page *page)
130{
131 return page->index;
132}
133
134static inline void set_pcppage_migratetype(struct page *page, int migratetype)
135{
136 page->index = migratetype;
137}
138
452aa699
RW
139#ifdef CONFIG_PM_SLEEP
140/*
141 * The following functions are used by the suspend/hibernate code to temporarily
142 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
143 * while devices are suspended. To avoid races with the suspend/hibernate code,
144 * they should always be called with pm_mutex held (gfp_allowed_mask also should
145 * only be modified with pm_mutex held, unless the suspend/hibernate code is
146 * guaranteed not to run in parallel with that modification).
147 */
c9e664f1
RW
148
149static gfp_t saved_gfp_mask;
150
151void pm_restore_gfp_mask(void)
452aa699
RW
152{
153 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
154 if (saved_gfp_mask) {
155 gfp_allowed_mask = saved_gfp_mask;
156 saved_gfp_mask = 0;
157 }
452aa699
RW
158}
159
c9e664f1 160void pm_restrict_gfp_mask(void)
452aa699 161{
452aa699 162 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
163 WARN_ON(saved_gfp_mask);
164 saved_gfp_mask = gfp_allowed_mask;
d0164adc 165 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
452aa699 166}
f90ac398
MG
167
168bool pm_suspended_storage(void)
169{
d0164adc 170 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
f90ac398
MG
171 return false;
172 return true;
173}
452aa699
RW
174#endif /* CONFIG_PM_SLEEP */
175
d9c23400 176#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
d00181b9 177unsigned int pageblock_order __read_mostly;
d9c23400
MG
178#endif
179
d98c7a09 180static void __free_pages_ok(struct page *page, unsigned int order);
a226f6c8 181
1da177e4
LT
182/*
183 * results with 256, 32 in the lowmem_reserve sysctl:
184 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
185 * 1G machine -> (16M dma, 784M normal, 224M high)
186 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
187 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
84109e15 188 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
a2f1b424
AK
189 *
190 * TBD: should special case ZONE_DMA32 machines here - in those we normally
191 * don't need any ZONE_NORMAL reservation
1da177e4 192 */
2f1b6248 193int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
4b51d669 194#ifdef CONFIG_ZONE_DMA
2f1b6248 195 256,
4b51d669 196#endif
fb0e7942 197#ifdef CONFIG_ZONE_DMA32
2f1b6248 198 256,
fb0e7942 199#endif
e53ef38d 200#ifdef CONFIG_HIGHMEM
2a1e274a 201 32,
e53ef38d 202#endif
2a1e274a 203 32,
2f1b6248 204};
1da177e4
LT
205
206EXPORT_SYMBOL(totalram_pages);
1da177e4 207
15ad7cdc 208static char * const zone_names[MAX_NR_ZONES] = {
4b51d669 209#ifdef CONFIG_ZONE_DMA
2f1b6248 210 "DMA",
4b51d669 211#endif
fb0e7942 212#ifdef CONFIG_ZONE_DMA32
2f1b6248 213 "DMA32",
fb0e7942 214#endif
2f1b6248 215 "Normal",
e53ef38d 216#ifdef CONFIG_HIGHMEM
2a1e274a 217 "HighMem",
e53ef38d 218#endif
2a1e274a 219 "Movable",
033fbae9
DW
220#ifdef CONFIG_ZONE_DEVICE
221 "Device",
222#endif
2f1b6248
CL
223};
224
f1e61557
KS
225static void free_compound_page(struct page *page);
226compound_page_dtor * const compound_page_dtors[] = {
227 NULL,
228 free_compound_page,
229#ifdef CONFIG_HUGETLB_PAGE
230 free_huge_page,
231#endif
232};
233
1da177e4 234int min_free_kbytes = 1024;
42aa83cb 235int user_min_free_kbytes = -1;
1da177e4 236
2c85f51d
JB
237static unsigned long __meminitdata nr_kernel_pages;
238static unsigned long __meminitdata nr_all_pages;
a3142c8e 239static unsigned long __meminitdata dma_reserve;
1da177e4 240
0ee332c1
TH
241#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
242static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
243static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
244static unsigned long __initdata required_kernelcore;
245static unsigned long __initdata required_movablecore;
246static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
247
248/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
249int movable_zone;
250EXPORT_SYMBOL(movable_zone);
251#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 252
418508c1
MS
253#if MAX_NUMNODES > 1
254int nr_node_ids __read_mostly = MAX_NUMNODES;
62bc62a8 255int nr_online_nodes __read_mostly = 1;
418508c1 256EXPORT_SYMBOL(nr_node_ids);
62bc62a8 257EXPORT_SYMBOL(nr_online_nodes);
418508c1
MS
258#endif
259
9ef9acb0
MG
260int page_group_by_mobility_disabled __read_mostly;
261
3a80a7fa
MG
262#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
263static inline void reset_deferred_meminit(pg_data_t *pgdat)
264{
265 pgdat->first_deferred_pfn = ULONG_MAX;
266}
267
268/* Returns true if the struct page for the pfn is uninitialised */
0e1cc95b 269static inline bool __meminit early_page_uninitialised(unsigned long pfn)
3a80a7fa 270{
ae026b2a 271 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
3a80a7fa
MG
272 return true;
273
274 return false;
275}
276
7e18adb4
MG
277static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
278{
279 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
280 return true;
281
282 return false;
283}
284
3a80a7fa
MG
285/*
286 * Returns false when the remaining initialisation should be deferred until
287 * later in the boot cycle when it can be parallelised.
288 */
289static inline bool update_defer_init(pg_data_t *pgdat,
290 unsigned long pfn, unsigned long zone_end,
291 unsigned long *nr_initialised)
292{
293 /* Always populate low zones for address-contrained allocations */
294 if (zone_end < pgdat_end_pfn(pgdat))
295 return true;
296
297 /* Initialise at least 2G of the highest zone */
298 (*nr_initialised)++;
299 if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
300 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
301 pgdat->first_deferred_pfn = pfn;
302 return false;
303 }
304
305 return true;
306}
307#else
308static inline void reset_deferred_meminit(pg_data_t *pgdat)
309{
310}
311
312static inline bool early_page_uninitialised(unsigned long pfn)
313{
314 return false;
315}
316
7e18adb4
MG
317static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
318{
319 return false;
320}
321
3a80a7fa
MG
322static inline bool update_defer_init(pg_data_t *pgdat,
323 unsigned long pfn, unsigned long zone_end,
324 unsigned long *nr_initialised)
325{
326 return true;
327}
328#endif
329
330
ee6f509c 331void set_pageblock_migratetype(struct page *page, int migratetype)
b2a0ac88 332{
5d0f3f72
KM
333 if (unlikely(page_group_by_mobility_disabled &&
334 migratetype < MIGRATE_PCPTYPES))
49255c61
MG
335 migratetype = MIGRATE_UNMOVABLE;
336
b2a0ac88
MG
337 set_pageblock_flags_group(page, (unsigned long)migratetype,
338 PB_migrate, PB_migrate_end);
339}
340
13e7444b 341#ifdef CONFIG_DEBUG_VM
c6a57e19 342static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
1da177e4 343{
bdc8cb98
DH
344 int ret = 0;
345 unsigned seq;
346 unsigned long pfn = page_to_pfn(page);
b5e6a5a2 347 unsigned long sp, start_pfn;
c6a57e19 348
bdc8cb98
DH
349 do {
350 seq = zone_span_seqbegin(zone);
b5e6a5a2
CS
351 start_pfn = zone->zone_start_pfn;
352 sp = zone->spanned_pages;
108bcc96 353 if (!zone_spans_pfn(zone, pfn))
bdc8cb98
DH
354 ret = 1;
355 } while (zone_span_seqretry(zone, seq));
356
b5e6a5a2 357 if (ret)
613813e8
DH
358 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
359 pfn, zone_to_nid(zone), zone->name,
360 start_pfn, start_pfn + sp);
b5e6a5a2 361
bdc8cb98 362 return ret;
c6a57e19
DH
363}
364
365static int page_is_consistent(struct zone *zone, struct page *page)
366{
14e07298 367 if (!pfn_valid_within(page_to_pfn(page)))
c6a57e19 368 return 0;
1da177e4 369 if (zone != page_zone(page))
c6a57e19
DH
370 return 0;
371
372 return 1;
373}
374/*
375 * Temporary debugging check for pages not lying within a given zone.
376 */
377static int bad_range(struct zone *zone, struct page *page)
378{
379 if (page_outside_zone_boundaries(zone, page))
1da177e4 380 return 1;
c6a57e19
DH
381 if (!page_is_consistent(zone, page))
382 return 1;
383
1da177e4
LT
384 return 0;
385}
13e7444b
NP
386#else
387static inline int bad_range(struct zone *zone, struct page *page)
388{
389 return 0;
390}
391#endif
392
d230dec1
KS
393static void bad_page(struct page *page, const char *reason,
394 unsigned long bad_flags)
1da177e4 395{
d936cf9b
HD
396 static unsigned long resume;
397 static unsigned long nr_shown;
398 static unsigned long nr_unshown;
399
2a7684a2
WF
400 /* Don't complain about poisoned pages */
401 if (PageHWPoison(page)) {
22b751c3 402 page_mapcount_reset(page); /* remove PageBuddy */
2a7684a2
WF
403 return;
404 }
405
d936cf9b
HD
406 /*
407 * Allow a burst of 60 reports, then keep quiet for that minute;
408 * or allow a steady drip of one report per second.
409 */
410 if (nr_shown == 60) {
411 if (time_before(jiffies, resume)) {
412 nr_unshown++;
413 goto out;
414 }
415 if (nr_unshown) {
1e9e6365
HD
416 printk(KERN_ALERT
417 "BUG: Bad page state: %lu messages suppressed\n",
d936cf9b
HD
418 nr_unshown);
419 nr_unshown = 0;
420 }
421 nr_shown = 0;
422 }
423 if (nr_shown++ == 0)
424 resume = jiffies + 60 * HZ;
425
1e9e6365 426 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
3dc14741 427 current->comm, page_to_pfn(page));
f0b791a3 428 dump_page_badflags(page, reason, bad_flags);
3dc14741 429
4f31888c 430 print_modules();
1da177e4 431 dump_stack();
d936cf9b 432out:
8cc3b392 433 /* Leave bad fields for debug, except PageBuddy could make trouble */
22b751c3 434 page_mapcount_reset(page); /* remove PageBuddy */
373d4d09 435 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1da177e4
LT
436}
437
1da177e4
LT
438/*
439 * Higher-order pages are called "compound pages". They are structured thusly:
440 *
1d798ca3 441 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
1da177e4 442 *
1d798ca3
KS
443 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
444 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
1da177e4 445 *
1d798ca3
KS
446 * The first tail page's ->compound_dtor holds the offset in array of compound
447 * page destructors. See compound_page_dtors.
1da177e4 448 *
1d798ca3 449 * The first tail page's ->compound_order holds the order of allocation.
41d78ba5 450 * This usage means that zero-order pages may not be compound.
1da177e4 451 */
d98c7a09
HD
452
453static void free_compound_page(struct page *page)
454{
d85f3385 455 __free_pages_ok(page, compound_order(page));
d98c7a09
HD
456}
457
d00181b9 458void prep_compound_page(struct page *page, unsigned int order)
18229df5
AW
459{
460 int i;
461 int nr_pages = 1 << order;
462
f1e61557 463 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
18229df5
AW
464 set_compound_order(page, order);
465 __SetPageHead(page);
466 for (i = 1; i < nr_pages; i++) {
467 struct page *p = page + i;
58a84aa9 468 set_page_count(p, 0);
1d798ca3 469 set_compound_head(p, page);
18229df5
AW
470 }
471}
472
c0a32fc5
SG
473#ifdef CONFIG_DEBUG_PAGEALLOC
474unsigned int _debug_guardpage_minorder;
031bc574 475bool _debug_pagealloc_enabled __read_mostly;
e30825f1
JK
476bool _debug_guardpage_enabled __read_mostly;
477
031bc574
JK
478static int __init early_debug_pagealloc(char *buf)
479{
480 if (!buf)
481 return -EINVAL;
482
483 if (strcmp(buf, "on") == 0)
484 _debug_pagealloc_enabled = true;
485
486 return 0;
487}
488early_param("debug_pagealloc", early_debug_pagealloc);
489
e30825f1
JK
490static bool need_debug_guardpage(void)
491{
031bc574
JK
492 /* If we don't use debug_pagealloc, we don't need guard page */
493 if (!debug_pagealloc_enabled())
494 return false;
495
e30825f1
JK
496 return true;
497}
498
499static void init_debug_guardpage(void)
500{
031bc574
JK
501 if (!debug_pagealloc_enabled())
502 return;
503
e30825f1
JK
504 _debug_guardpage_enabled = true;
505}
506
507struct page_ext_operations debug_guardpage_ops = {
508 .need = need_debug_guardpage,
509 .init = init_debug_guardpage,
510};
c0a32fc5
SG
511
512static int __init debug_guardpage_minorder_setup(char *buf)
513{
514 unsigned long res;
515
516 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
517 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
518 return 0;
519 }
520 _debug_guardpage_minorder = res;
521 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
522 return 0;
523}
524__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
525
2847cf95
JK
526static inline void set_page_guard(struct zone *zone, struct page *page,
527 unsigned int order, int migratetype)
c0a32fc5 528{
e30825f1
JK
529 struct page_ext *page_ext;
530
531 if (!debug_guardpage_enabled())
532 return;
533
534 page_ext = lookup_page_ext(page);
535 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
536
2847cf95
JK
537 INIT_LIST_HEAD(&page->lru);
538 set_page_private(page, order);
539 /* Guard pages are not available for any usage */
540 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
c0a32fc5
SG
541}
542
2847cf95
JK
543static inline void clear_page_guard(struct zone *zone, struct page *page,
544 unsigned int order, int migratetype)
c0a32fc5 545{
e30825f1
JK
546 struct page_ext *page_ext;
547
548 if (!debug_guardpage_enabled())
549 return;
550
551 page_ext = lookup_page_ext(page);
552 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
553
2847cf95
JK
554 set_page_private(page, 0);
555 if (!is_migrate_isolate(migratetype))
556 __mod_zone_freepage_state(zone, (1 << order), migratetype);
c0a32fc5
SG
557}
558#else
e30825f1 559struct page_ext_operations debug_guardpage_ops = { NULL, };
2847cf95
JK
560static inline void set_page_guard(struct zone *zone, struct page *page,
561 unsigned int order, int migratetype) {}
562static inline void clear_page_guard(struct zone *zone, struct page *page,
563 unsigned int order, int migratetype) {}
c0a32fc5
SG
564#endif
565
7aeb09f9 566static inline void set_page_order(struct page *page, unsigned int order)
6aa3001b 567{
4c21e2f2 568 set_page_private(page, order);
676165a8 569 __SetPageBuddy(page);
1da177e4
LT
570}
571
572static inline void rmv_page_order(struct page *page)
573{
676165a8 574 __ClearPageBuddy(page);
4c21e2f2 575 set_page_private(page, 0);
1da177e4
LT
576}
577
1da177e4
LT
578/*
579 * This function checks whether a page is free && is the buddy
580 * we can do coalesce a page and its buddy if
13e7444b 581 * (a) the buddy is not in a hole &&
676165a8 582 * (b) the buddy is in the buddy system &&
cb2b95e1
AW
583 * (c) a page and its buddy have the same order &&
584 * (d) a page and its buddy are in the same zone.
676165a8 585 *
cf6fe945
WSH
586 * For recording whether a page is in the buddy system, we set ->_mapcount
587 * PAGE_BUDDY_MAPCOUNT_VALUE.
588 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
589 * serialized by zone->lock.
1da177e4 590 *
676165a8 591 * For recording page's order, we use page_private(page).
1da177e4 592 */
cb2b95e1 593static inline int page_is_buddy(struct page *page, struct page *buddy,
7aeb09f9 594 unsigned int order)
1da177e4 595{
14e07298 596 if (!pfn_valid_within(page_to_pfn(buddy)))
13e7444b 597 return 0;
13e7444b 598
c0a32fc5 599 if (page_is_guard(buddy) && page_order(buddy) == order) {
d34c5fa0
MG
600 if (page_zone_id(page) != page_zone_id(buddy))
601 return 0;
602
4c5018ce
WY
603 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
604
c0a32fc5
SG
605 return 1;
606 }
607
cb2b95e1 608 if (PageBuddy(buddy) && page_order(buddy) == order) {
d34c5fa0
MG
609 /*
610 * zone check is done late to avoid uselessly
611 * calculating zone/node ids for pages that could
612 * never merge.
613 */
614 if (page_zone_id(page) != page_zone_id(buddy))
615 return 0;
616
4c5018ce
WY
617 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
618
6aa3001b 619 return 1;
676165a8 620 }
6aa3001b 621 return 0;
1da177e4
LT
622}
623
624/*
625 * Freeing function for a buddy system allocator.
626 *
627 * The concept of a buddy system is to maintain direct-mapped table
628 * (containing bit values) for memory blocks of various "orders".
629 * The bottom level table contains the map for the smallest allocatable
630 * units of memory (here, pages), and each level above it describes
631 * pairs of units from the levels below, hence, "buddies".
632 * At a high level, all that happens here is marking the table entry
633 * at the bottom level available, and propagating the changes upward
634 * as necessary, plus some accounting needed to play nicely with other
635 * parts of the VM system.
636 * At each level, we keep a list of pages, which are heads of continuous
cf6fe945
WSH
637 * free pages of length of (1 << order) and marked with _mapcount
638 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
639 * field.
1da177e4 640 * So when we are allocating or freeing one, we can derive the state of the
5f63b720
MN
641 * other. That is, if we allocate a small block, and both were
642 * free, the remainder of the region must be split into blocks.
1da177e4 643 * If a block is freed, and its buddy is also free, then this
5f63b720 644 * triggers coalescing into a block of larger size.
1da177e4 645 *
6d49e352 646 * -- nyc
1da177e4
LT
647 */
648
48db57f8 649static inline void __free_one_page(struct page *page,
dc4b0caf 650 unsigned long pfn,
ed0ae21d
MG
651 struct zone *zone, unsigned int order,
652 int migratetype)
1da177e4
LT
653{
654 unsigned long page_idx;
6dda9d55 655 unsigned long combined_idx;
43506fad 656 unsigned long uninitialized_var(buddy_idx);
6dda9d55 657 struct page *buddy;
d00181b9 658 unsigned int max_order = MAX_ORDER;
1da177e4 659
d29bb978 660 VM_BUG_ON(!zone_is_initialized(zone));
6e9f0d58 661 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1da177e4 662
ed0ae21d 663 VM_BUG_ON(migratetype == -1);
3c605096
JK
664 if (is_migrate_isolate(migratetype)) {
665 /*
666 * We restrict max order of merging to prevent merge
667 * between freepages on isolate pageblock and normal
668 * pageblock. Without this, pageblock isolation
669 * could cause incorrect freepage accounting.
670 */
d00181b9 671 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
3c605096 672 } else {
8f82b55d 673 __mod_zone_freepage_state(zone, 1 << order, migratetype);
3c605096 674 }
ed0ae21d 675
3c605096 676 page_idx = pfn & ((1 << max_order) - 1);
1da177e4 677
309381fe
SL
678 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
679 VM_BUG_ON_PAGE(bad_range(zone, page), page);
1da177e4 680
3c605096 681 while (order < max_order - 1) {
43506fad
KC
682 buddy_idx = __find_buddy_index(page_idx, order);
683 buddy = page + (buddy_idx - page_idx);
cb2b95e1 684 if (!page_is_buddy(page, buddy, order))
3c82d0ce 685 break;
c0a32fc5
SG
686 /*
687 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
688 * merge with it and move up one order.
689 */
690 if (page_is_guard(buddy)) {
2847cf95 691 clear_page_guard(zone, buddy, order, migratetype);
c0a32fc5
SG
692 } else {
693 list_del(&buddy->lru);
694 zone->free_area[order].nr_free--;
695 rmv_page_order(buddy);
696 }
43506fad 697 combined_idx = buddy_idx & page_idx;
1da177e4
LT
698 page = page + (combined_idx - page_idx);
699 page_idx = combined_idx;
700 order++;
701 }
702 set_page_order(page, order);
6dda9d55
CZ
703
704 /*
705 * If this is not the largest possible page, check if the buddy
706 * of the next-highest order is free. If it is, it's possible
707 * that pages are being freed that will coalesce soon. In case,
708 * that is happening, add the free page to the tail of the list
709 * so it's less likely to be used soon and more likely to be merged
710 * as a higher order page
711 */
b7f50cfa 712 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
6dda9d55 713 struct page *higher_page, *higher_buddy;
43506fad
KC
714 combined_idx = buddy_idx & page_idx;
715 higher_page = page + (combined_idx - page_idx);
716 buddy_idx = __find_buddy_index(combined_idx, order + 1);
0ba8f2d5 717 higher_buddy = higher_page + (buddy_idx - combined_idx);
6dda9d55
CZ
718 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
719 list_add_tail(&page->lru,
720 &zone->free_area[order].free_list[migratetype]);
721 goto out;
722 }
723 }
724
725 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
726out:
1da177e4
LT
727 zone->free_area[order].nr_free++;
728}
729
224abf92 730static inline int free_pages_check(struct page *page)
1da177e4 731{
d230dec1 732 const char *bad_reason = NULL;
f0b791a3
DH
733 unsigned long bad_flags = 0;
734
735 if (unlikely(page_mapcount(page)))
736 bad_reason = "nonzero mapcount";
737 if (unlikely(page->mapping != NULL))
738 bad_reason = "non-NULL mapping";
739 if (unlikely(atomic_read(&page->_count) != 0))
740 bad_reason = "nonzero _count";
741 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
742 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
743 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
744 }
9edad6ea
JW
745#ifdef CONFIG_MEMCG
746 if (unlikely(page->mem_cgroup))
747 bad_reason = "page still charged to cgroup";
748#endif
f0b791a3
DH
749 if (unlikely(bad_reason)) {
750 bad_page(page, bad_reason, bad_flags);
79f4b7bf 751 return 1;
8cc3b392 752 }
90572890 753 page_cpupid_reset_last(page);
79f4b7bf
HD
754 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
755 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
756 return 0;
1da177e4
LT
757}
758
759/*
5f8dcc21 760 * Frees a number of pages from the PCP lists
1da177e4 761 * Assumes all pages on list are in same zone, and of same order.
207f36ee 762 * count is the number of pages to free.
1da177e4
LT
763 *
764 * If the zone was previously in an "all pages pinned" state then look to
765 * see if this freeing clears that state.
766 *
767 * And clear the zone's pages_scanned counter, to hold off the "all pages are
768 * pinned" detection logic.
769 */
5f8dcc21
MG
770static void free_pcppages_bulk(struct zone *zone, int count,
771 struct per_cpu_pages *pcp)
1da177e4 772{
5f8dcc21 773 int migratetype = 0;
a6f9edd6 774 int batch_free = 0;
72853e29 775 int to_free = count;
0d5d823a 776 unsigned long nr_scanned;
5f8dcc21 777
c54ad30c 778 spin_lock(&zone->lock);
0d5d823a
MG
779 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
780 if (nr_scanned)
781 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
f2260e6b 782
72853e29 783 while (to_free) {
48db57f8 784 struct page *page;
5f8dcc21
MG
785 struct list_head *list;
786
787 /*
a6f9edd6
MG
788 * Remove pages from lists in a round-robin fashion. A
789 * batch_free count is maintained that is incremented when an
790 * empty list is encountered. This is so more pages are freed
791 * off fuller lists instead of spinning excessively around empty
792 * lists
5f8dcc21
MG
793 */
794 do {
a6f9edd6 795 batch_free++;
5f8dcc21
MG
796 if (++migratetype == MIGRATE_PCPTYPES)
797 migratetype = 0;
798 list = &pcp->lists[migratetype];
799 } while (list_empty(list));
48db57f8 800
1d16871d
NK
801 /* This is the only non-empty list. Free them all. */
802 if (batch_free == MIGRATE_PCPTYPES)
803 batch_free = to_free;
804
a6f9edd6 805 do {
770c8aaa
BZ
806 int mt; /* migratetype of the to-be-freed page */
807
a6f9edd6
MG
808 page = list_entry(list->prev, struct page, lru);
809 /* must delete as __free_one_page list manipulates */
810 list_del(&page->lru);
aa016d14 811
bb14c2c7 812 mt = get_pcppage_migratetype(page);
aa016d14
VB
813 /* MIGRATE_ISOLATE page should not go to pcplists */
814 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
815 /* Pageblock could have been isolated meanwhile */
8f82b55d 816 if (unlikely(has_isolate_pageblock(zone)))
51bb1a40 817 mt = get_pageblock_migratetype(page);
51bb1a40 818
dc4b0caf 819 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
770c8aaa 820 trace_mm_page_pcpu_drain(page, 0, mt);
72853e29 821 } while (--to_free && --batch_free && !list_empty(list));
1da177e4 822 }
c54ad30c 823 spin_unlock(&zone->lock);
1da177e4
LT
824}
825
dc4b0caf
MG
826static void free_one_page(struct zone *zone,
827 struct page *page, unsigned long pfn,
7aeb09f9 828 unsigned int order,
ed0ae21d 829 int migratetype)
1da177e4 830{
0d5d823a 831 unsigned long nr_scanned;
006d22d9 832 spin_lock(&zone->lock);
0d5d823a
MG
833 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
834 if (nr_scanned)
835 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
f2260e6b 836
ad53f92e
JK
837 if (unlikely(has_isolate_pageblock(zone) ||
838 is_migrate_isolate(migratetype))) {
839 migratetype = get_pfnblock_migratetype(page, pfn);
ad53f92e 840 }
dc4b0caf 841 __free_one_page(page, pfn, zone, order, migratetype);
006d22d9 842 spin_unlock(&zone->lock);
48db57f8
NP
843}
844
81422f29
KS
845static int free_tail_pages_check(struct page *head_page, struct page *page)
846{
1d798ca3
KS
847 int ret = 1;
848
849 /*
850 * We rely page->lru.next never has bit 0 set, unless the page
851 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
852 */
853 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
854
855 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
856 ret = 0;
857 goto out;
858 }
81422f29
KS
859 if (unlikely(!PageTail(page))) {
860 bad_page(page, "PageTail not set", 0);
1d798ca3 861 goto out;
81422f29 862 }
1d798ca3
KS
863 if (unlikely(compound_head(page) != head_page)) {
864 bad_page(page, "compound_head not consistent", 0);
865 goto out;
81422f29 866 }
1d798ca3
KS
867 ret = 0;
868out:
869 clear_compound_head(page);
870 return ret;
81422f29
KS
871}
872
1e8ce83c
RH
873static void __meminit __init_single_page(struct page *page, unsigned long pfn,
874 unsigned long zone, int nid)
875{
1e8ce83c 876 set_page_links(page, zone, nid, pfn);
1e8ce83c
RH
877 init_page_count(page);
878 page_mapcount_reset(page);
879 page_cpupid_reset_last(page);
1e8ce83c 880
1e8ce83c
RH
881 INIT_LIST_HEAD(&page->lru);
882#ifdef WANT_PAGE_VIRTUAL
883 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
884 if (!is_highmem_idx(zone))
885 set_page_address(page, __va(pfn << PAGE_SHIFT));
886#endif
887}
888
889static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
890 int nid)
891{
892 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
893}
894
7e18adb4
MG
895#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
896static void init_reserved_page(unsigned long pfn)
897{
898 pg_data_t *pgdat;
899 int nid, zid;
900
901 if (!early_page_uninitialised(pfn))
902 return;
903
904 nid = early_pfn_to_nid(pfn);
905 pgdat = NODE_DATA(nid);
906
907 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
908 struct zone *zone = &pgdat->node_zones[zid];
909
910 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
911 break;
912 }
913 __init_single_pfn(pfn, zid, nid);
914}
915#else
916static inline void init_reserved_page(unsigned long pfn)
917{
918}
919#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
920
92923ca3
NZ
921/*
922 * Initialised pages do not have PageReserved set. This function is
923 * called for each range allocated by the bootmem allocator and
924 * marks the pages PageReserved. The remaining valid pages are later
925 * sent to the buddy page allocator.
926 */
7e18adb4 927void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
92923ca3
NZ
928{
929 unsigned long start_pfn = PFN_DOWN(start);
930 unsigned long end_pfn = PFN_UP(end);
931
7e18adb4
MG
932 for (; start_pfn < end_pfn; start_pfn++) {
933 if (pfn_valid(start_pfn)) {
934 struct page *page = pfn_to_page(start_pfn);
935
936 init_reserved_page(start_pfn);
1d798ca3
KS
937
938 /* Avoid false-positive PageTail() */
939 INIT_LIST_HEAD(&page->lru);
940
7e18adb4
MG
941 SetPageReserved(page);
942 }
943 }
92923ca3
NZ
944}
945
ec95f53a 946static bool free_pages_prepare(struct page *page, unsigned int order)
48db57f8 947{
81422f29
KS
948 bool compound = PageCompound(page);
949 int i, bad = 0;
1da177e4 950
ab1f306f 951 VM_BUG_ON_PAGE(PageTail(page), page);
81422f29 952 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
ab1f306f 953
b413d48a 954 trace_mm_page_free(page, order);
b1eeab67 955 kmemcheck_free_shadow(page, order);
b8c73fc2 956 kasan_free_pages(page, order);
b1eeab67 957
8dd60a3a
AA
958 if (PageAnon(page))
959 page->mapping = NULL;
81422f29
KS
960 bad += free_pages_check(page);
961 for (i = 1; i < (1 << order); i++) {
962 if (compound)
963 bad += free_tail_pages_check(page, page + i);
8dd60a3a 964 bad += free_pages_check(page + i);
81422f29 965 }
8cc3b392 966 if (bad)
ec95f53a 967 return false;
689bcebf 968
48c96a36
JK
969 reset_page_owner(page, order);
970
3ac7fe5a 971 if (!PageHighMem(page)) {
b8af2941
PK
972 debug_check_no_locks_freed(page_address(page),
973 PAGE_SIZE << order);
3ac7fe5a
TG
974 debug_check_no_obj_freed(page_address(page),
975 PAGE_SIZE << order);
976 }
dafb1367 977 arch_free_page(page, order);
48db57f8 978 kernel_map_pages(page, 1 << order, 0);
dafb1367 979
ec95f53a
KM
980 return true;
981}
982
983static void __free_pages_ok(struct page *page, unsigned int order)
984{
985 unsigned long flags;
95e34412 986 int migratetype;
dc4b0caf 987 unsigned long pfn = page_to_pfn(page);
ec95f53a
KM
988
989 if (!free_pages_prepare(page, order))
990 return;
991
cfc47a28 992 migratetype = get_pfnblock_migratetype(page, pfn);
c54ad30c 993 local_irq_save(flags);
f8891e5e 994 __count_vm_events(PGFREE, 1 << order);
dc4b0caf 995 free_one_page(page_zone(page), page, pfn, order, migratetype);
c54ad30c 996 local_irq_restore(flags);
1da177e4
LT
997}
998
0e1cc95b 999static void __init __free_pages_boot_core(struct page *page,
3a80a7fa 1000 unsigned long pfn, unsigned int order)
a226f6c8 1001{
c3993076 1002 unsigned int nr_pages = 1 << order;
e2d0bd2b 1003 struct page *p = page;
c3993076 1004 unsigned int loop;
a226f6c8 1005
e2d0bd2b
YL
1006 prefetchw(p);
1007 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1008 prefetchw(p + 1);
c3993076
JW
1009 __ClearPageReserved(p);
1010 set_page_count(p, 0);
a226f6c8 1011 }
e2d0bd2b
YL
1012 __ClearPageReserved(p);
1013 set_page_count(p, 0);
c3993076 1014
e2d0bd2b 1015 page_zone(page)->managed_pages += nr_pages;
c3993076
JW
1016 set_page_refcounted(page);
1017 __free_pages(page, order);
a226f6c8
DH
1018}
1019
75a592a4
MG
1020#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1021 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
7ace9917 1022
75a592a4
MG
1023static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1024
1025int __meminit early_pfn_to_nid(unsigned long pfn)
1026{
7ace9917 1027 static DEFINE_SPINLOCK(early_pfn_lock);
75a592a4
MG
1028 int nid;
1029
7ace9917 1030 spin_lock(&early_pfn_lock);
75a592a4 1031 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
7ace9917
MG
1032 if (nid < 0)
1033 nid = 0;
1034 spin_unlock(&early_pfn_lock);
1035
1036 return nid;
75a592a4
MG
1037}
1038#endif
1039
1040#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1041static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1042 struct mminit_pfnnid_cache *state)
1043{
1044 int nid;
1045
1046 nid = __early_pfn_to_nid(pfn, state);
1047 if (nid >= 0 && nid != node)
1048 return false;
1049 return true;
1050}
1051
1052/* Only safe to use early in boot when initialisation is single-threaded */
1053static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1054{
1055 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1056}
1057
1058#else
1059
1060static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1061{
1062 return true;
1063}
1064static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1065 struct mminit_pfnnid_cache *state)
1066{
1067 return true;
1068}
1069#endif
1070
1071
0e1cc95b 1072void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
3a80a7fa
MG
1073 unsigned int order)
1074{
1075 if (early_page_uninitialised(pfn))
1076 return;
1077 return __free_pages_boot_core(page, pfn, order);
1078}
1079
7e18adb4 1080#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
0e1cc95b 1081static void __init deferred_free_range(struct page *page,
a4de83dd
MG
1082 unsigned long pfn, int nr_pages)
1083{
1084 int i;
1085
1086 if (!page)
1087 return;
1088
1089 /* Free a large naturally-aligned chunk if possible */
1090 if (nr_pages == MAX_ORDER_NR_PAGES &&
1091 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
ac5d2539 1092 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
a4de83dd
MG
1093 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1094 return;
1095 }
1096
1097 for (i = 0; i < nr_pages; i++, page++, pfn++)
1098 __free_pages_boot_core(page, pfn, 0);
1099}
1100
d3cd131d
NS
1101/* Completion tracking for deferred_init_memmap() threads */
1102static atomic_t pgdat_init_n_undone __initdata;
1103static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1104
1105static inline void __init pgdat_init_report_one_done(void)
1106{
1107 if (atomic_dec_and_test(&pgdat_init_n_undone))
1108 complete(&pgdat_init_all_done_comp);
1109}
0e1cc95b 1110
7e18adb4 1111/* Initialise remaining memory on a node */
0e1cc95b 1112static int __init deferred_init_memmap(void *data)
7e18adb4 1113{
0e1cc95b
MG
1114 pg_data_t *pgdat = data;
1115 int nid = pgdat->node_id;
7e18adb4
MG
1116 struct mminit_pfnnid_cache nid_init_state = { };
1117 unsigned long start = jiffies;
1118 unsigned long nr_pages = 0;
1119 unsigned long walk_start, walk_end;
1120 int i, zid;
1121 struct zone *zone;
7e18adb4 1122 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
0e1cc95b 1123 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
7e18adb4 1124
0e1cc95b 1125 if (first_init_pfn == ULONG_MAX) {
d3cd131d 1126 pgdat_init_report_one_done();
0e1cc95b
MG
1127 return 0;
1128 }
1129
1130 /* Bind memory initialisation thread to a local node if possible */
1131 if (!cpumask_empty(cpumask))
1132 set_cpus_allowed_ptr(current, cpumask);
7e18adb4
MG
1133
1134 /* Sanity check boundaries */
1135 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1136 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1137 pgdat->first_deferred_pfn = ULONG_MAX;
1138
1139 /* Only the highest zone is deferred so find it */
1140 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1141 zone = pgdat->node_zones + zid;
1142 if (first_init_pfn < zone_end_pfn(zone))
1143 break;
1144 }
1145
1146 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1147 unsigned long pfn, end_pfn;
54608c3f 1148 struct page *page = NULL;
a4de83dd
MG
1149 struct page *free_base_page = NULL;
1150 unsigned long free_base_pfn = 0;
1151 int nr_to_free = 0;
7e18adb4
MG
1152
1153 end_pfn = min(walk_end, zone_end_pfn(zone));
1154 pfn = first_init_pfn;
1155 if (pfn < walk_start)
1156 pfn = walk_start;
1157 if (pfn < zone->zone_start_pfn)
1158 pfn = zone->zone_start_pfn;
1159
1160 for (; pfn < end_pfn; pfn++) {
54608c3f 1161 if (!pfn_valid_within(pfn))
a4de83dd 1162 goto free_range;
7e18adb4 1163
54608c3f
MG
1164 /*
1165 * Ensure pfn_valid is checked every
1166 * MAX_ORDER_NR_PAGES for memory holes
1167 */
1168 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1169 if (!pfn_valid(pfn)) {
1170 page = NULL;
a4de83dd 1171 goto free_range;
54608c3f
MG
1172 }
1173 }
1174
1175 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1176 page = NULL;
a4de83dd 1177 goto free_range;
54608c3f
MG
1178 }
1179
1180 /* Minimise pfn page lookups and scheduler checks */
1181 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1182 page++;
1183 } else {
a4de83dd
MG
1184 nr_pages += nr_to_free;
1185 deferred_free_range(free_base_page,
1186 free_base_pfn, nr_to_free);
1187 free_base_page = NULL;
1188 free_base_pfn = nr_to_free = 0;
1189
54608c3f
MG
1190 page = pfn_to_page(pfn);
1191 cond_resched();
1192 }
7e18adb4
MG
1193
1194 if (page->flags) {
1195 VM_BUG_ON(page_zone(page) != zone);
a4de83dd 1196 goto free_range;
7e18adb4
MG
1197 }
1198
1199 __init_single_page(page, pfn, zid, nid);
a4de83dd
MG
1200 if (!free_base_page) {
1201 free_base_page = page;
1202 free_base_pfn = pfn;
1203 nr_to_free = 0;
1204 }
1205 nr_to_free++;
1206
1207 /* Where possible, batch up pages for a single free */
1208 continue;
1209free_range:
1210 /* Free the current block of pages to allocator */
1211 nr_pages += nr_to_free;
1212 deferred_free_range(free_base_page, free_base_pfn,
1213 nr_to_free);
1214 free_base_page = NULL;
1215 free_base_pfn = nr_to_free = 0;
7e18adb4 1216 }
a4de83dd 1217
7e18adb4
MG
1218 first_init_pfn = max(end_pfn, first_init_pfn);
1219 }
1220
1221 /* Sanity check that the next zone really is unpopulated */
1222 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1223
0e1cc95b 1224 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
7e18adb4 1225 jiffies_to_msecs(jiffies - start));
d3cd131d
NS
1226
1227 pgdat_init_report_one_done();
0e1cc95b
MG
1228 return 0;
1229}
1230
1231void __init page_alloc_init_late(void)
1232{
1233 int nid;
1234
d3cd131d
NS
1235 /* There will be num_node_state(N_MEMORY) threads */
1236 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
0e1cc95b 1237 for_each_node_state(nid, N_MEMORY) {
0e1cc95b
MG
1238 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1239 }
1240
1241 /* Block until all are initialised */
d3cd131d 1242 wait_for_completion(&pgdat_init_all_done_comp);
4248b0da
MG
1243
1244 /* Reinit limits that are based on free pages after the kernel is up */
1245 files_maxfiles_init();
7e18adb4
MG
1246}
1247#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1248
47118af0 1249#ifdef CONFIG_CMA
9cf510a5 1250/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
47118af0
MN
1251void __init init_cma_reserved_pageblock(struct page *page)
1252{
1253 unsigned i = pageblock_nr_pages;
1254 struct page *p = page;
1255
1256 do {
1257 __ClearPageReserved(p);
1258 set_page_count(p, 0);
1259 } while (++p, --i);
1260
47118af0 1261 set_pageblock_migratetype(page, MIGRATE_CMA);
dc78327c
MN
1262
1263 if (pageblock_order >= MAX_ORDER) {
1264 i = pageblock_nr_pages;
1265 p = page;
1266 do {
1267 set_page_refcounted(p);
1268 __free_pages(p, MAX_ORDER - 1);
1269 p += MAX_ORDER_NR_PAGES;
1270 } while (i -= MAX_ORDER_NR_PAGES);
1271 } else {
1272 set_page_refcounted(page);
1273 __free_pages(page, pageblock_order);
1274 }
1275
3dcc0571 1276 adjust_managed_page_count(page, pageblock_nr_pages);
47118af0
MN
1277}
1278#endif
1da177e4
LT
1279
1280/*
1281 * The order of subdivision here is critical for the IO subsystem.
1282 * Please do not alter this order without good reasons and regression
1283 * testing. Specifically, as large blocks of memory are subdivided,
1284 * the order in which smaller blocks are delivered depends on the order
1285 * they're subdivided in this function. This is the primary factor
1286 * influencing the order in which pages are delivered to the IO
1287 * subsystem according to empirical testing, and this is also justified
1288 * by considering the behavior of a buddy system containing a single
1289 * large block of memory acted on by a series of small allocations.
1290 * This behavior is a critical factor in sglist merging's success.
1291 *
6d49e352 1292 * -- nyc
1da177e4 1293 */
085cc7d5 1294static inline void expand(struct zone *zone, struct page *page,
b2a0ac88
MG
1295 int low, int high, struct free_area *area,
1296 int migratetype)
1da177e4
LT
1297{
1298 unsigned long size = 1 << high;
1299
1300 while (high > low) {
1301 area--;
1302 high--;
1303 size >>= 1;
309381fe 1304 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
c0a32fc5 1305
2847cf95 1306 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
e30825f1 1307 debug_guardpage_enabled() &&
2847cf95 1308 high < debug_guardpage_minorder()) {
c0a32fc5
SG
1309 /*
1310 * Mark as guard pages (or page), that will allow to
1311 * merge back to allocator when buddy will be freed.
1312 * Corresponding page table entries will not be touched,
1313 * pages will stay not present in virtual address space
1314 */
2847cf95 1315 set_page_guard(zone, &page[size], high, migratetype);
c0a32fc5
SG
1316 continue;
1317 }
b2a0ac88 1318 list_add(&page[size].lru, &area->free_list[migratetype]);
1da177e4
LT
1319 area->nr_free++;
1320 set_page_order(&page[size], high);
1321 }
1da177e4
LT
1322}
1323
1da177e4
LT
1324/*
1325 * This page is about to be returned from the page allocator
1326 */
2a7684a2 1327static inline int check_new_page(struct page *page)
1da177e4 1328{
d230dec1 1329 const char *bad_reason = NULL;
f0b791a3
DH
1330 unsigned long bad_flags = 0;
1331
1332 if (unlikely(page_mapcount(page)))
1333 bad_reason = "nonzero mapcount";
1334 if (unlikely(page->mapping != NULL))
1335 bad_reason = "non-NULL mapping";
1336 if (unlikely(atomic_read(&page->_count) != 0))
1337 bad_reason = "nonzero _count";
f4c18e6f
NH
1338 if (unlikely(page->flags & __PG_HWPOISON)) {
1339 bad_reason = "HWPoisoned (hardware-corrupted)";
1340 bad_flags = __PG_HWPOISON;
1341 }
f0b791a3
DH
1342 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1343 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1344 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1345 }
9edad6ea
JW
1346#ifdef CONFIG_MEMCG
1347 if (unlikely(page->mem_cgroup))
1348 bad_reason = "page still charged to cgroup";
1349#endif
f0b791a3
DH
1350 if (unlikely(bad_reason)) {
1351 bad_page(page, bad_reason, bad_flags);
689bcebf 1352 return 1;
8cc3b392 1353 }
2a7684a2
WF
1354 return 0;
1355}
1356
75379191
VB
1357static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1358 int alloc_flags)
2a7684a2
WF
1359{
1360 int i;
1361
1362 for (i = 0; i < (1 << order); i++) {
1363 struct page *p = page + i;
1364 if (unlikely(check_new_page(p)))
1365 return 1;
1366 }
689bcebf 1367
4c21e2f2 1368 set_page_private(page, 0);
7835e98b 1369 set_page_refcounted(page);
cc102509
NP
1370
1371 arch_alloc_page(page, order);
1da177e4 1372 kernel_map_pages(page, 1 << order, 1);
b8c73fc2 1373 kasan_alloc_pages(page, order);
17cf4406
NP
1374
1375 if (gfp_flags & __GFP_ZERO)
f4d2897b
AA
1376 for (i = 0; i < (1 << order); i++)
1377 clear_highpage(page + i);
17cf4406
NP
1378
1379 if (order && (gfp_flags & __GFP_COMP))
1380 prep_compound_page(page, order);
1381
48c96a36
JK
1382 set_page_owner(page, order, gfp_flags);
1383
75379191 1384 /*
2f064f34 1385 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
75379191
VB
1386 * allocate the page. The expectation is that the caller is taking
1387 * steps that will free more memory. The caller should avoid the page
1388 * being used for !PFMEMALLOC purposes.
1389 */
2f064f34
MH
1390 if (alloc_flags & ALLOC_NO_WATERMARKS)
1391 set_page_pfmemalloc(page);
1392 else
1393 clear_page_pfmemalloc(page);
75379191 1394
689bcebf 1395 return 0;
1da177e4
LT
1396}
1397
56fd56b8
MG
1398/*
1399 * Go through the free lists for the given migratetype and remove
1400 * the smallest available page from the freelists
1401 */
728ec980
MG
1402static inline
1403struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
56fd56b8
MG
1404 int migratetype)
1405{
1406 unsigned int current_order;
b8af2941 1407 struct free_area *area;
56fd56b8
MG
1408 struct page *page;
1409
1410 /* Find a page of the appropriate size in the preferred list */
1411 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1412 area = &(zone->free_area[current_order]);
1413 if (list_empty(&area->free_list[migratetype]))
1414 continue;
1415
1416 page = list_entry(area->free_list[migratetype].next,
1417 struct page, lru);
1418 list_del(&page->lru);
1419 rmv_page_order(page);
1420 area->nr_free--;
56fd56b8 1421 expand(zone, page, order, current_order, area, migratetype);
bb14c2c7 1422 set_pcppage_migratetype(page, migratetype);
56fd56b8
MG
1423 return page;
1424 }
1425
1426 return NULL;
1427}
1428
1429
b2a0ac88
MG
1430/*
1431 * This array describes the order lists are fallen back to when
1432 * the free lists for the desirable migrate type are depleted
1433 */
47118af0 1434static int fallbacks[MIGRATE_TYPES][4] = {
974a786e
MG
1435 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1436 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1437 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
47118af0 1438#ifdef CONFIG_CMA
974a786e 1439 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
47118af0 1440#endif
194159fb 1441#ifdef CONFIG_MEMORY_ISOLATION
974a786e 1442 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
194159fb 1443#endif
b2a0ac88
MG
1444};
1445
dc67647b
JK
1446#ifdef CONFIG_CMA
1447static struct page *__rmqueue_cma_fallback(struct zone *zone,
1448 unsigned int order)
1449{
1450 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1451}
1452#else
1453static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1454 unsigned int order) { return NULL; }
1455#endif
1456
c361be55
MG
1457/*
1458 * Move the free pages in a range to the free lists of the requested type.
d9c23400 1459 * Note that start_page and end_pages are not aligned on a pageblock
c361be55
MG
1460 * boundary. If alignment is required, use move_freepages_block()
1461 */
435b405c 1462int move_freepages(struct zone *zone,
b69a7288
AB
1463 struct page *start_page, struct page *end_page,
1464 int migratetype)
c361be55
MG
1465{
1466 struct page *page;
d00181b9 1467 unsigned int order;
d100313f 1468 int pages_moved = 0;
c361be55
MG
1469
1470#ifndef CONFIG_HOLES_IN_ZONE
1471 /*
1472 * page_zone is not safe to call in this context when
1473 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1474 * anyway as we check zone boundaries in move_freepages_block().
1475 * Remove at a later date when no bug reports exist related to
ac0e5b7a 1476 * grouping pages by mobility
c361be55 1477 */
97ee4ba7 1478 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
c361be55
MG
1479#endif
1480
1481 for (page = start_page; page <= end_page;) {
344c790e 1482 /* Make sure we are not inadvertently changing nodes */
309381fe 1483 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
344c790e 1484
c361be55
MG
1485 if (!pfn_valid_within(page_to_pfn(page))) {
1486 page++;
1487 continue;
1488 }
1489
1490 if (!PageBuddy(page)) {
1491 page++;
1492 continue;
1493 }
1494
1495 order = page_order(page);
84be48d8
KS
1496 list_move(&page->lru,
1497 &zone->free_area[order].free_list[migratetype]);
c361be55 1498 page += 1 << order;
d100313f 1499 pages_moved += 1 << order;
c361be55
MG
1500 }
1501
d100313f 1502 return pages_moved;
c361be55
MG
1503}
1504
ee6f509c 1505int move_freepages_block(struct zone *zone, struct page *page,
68e3e926 1506 int migratetype)
c361be55
MG
1507{
1508 unsigned long start_pfn, end_pfn;
1509 struct page *start_page, *end_page;
1510
1511 start_pfn = page_to_pfn(page);
d9c23400 1512 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
c361be55 1513 start_page = pfn_to_page(start_pfn);
d9c23400
MG
1514 end_page = start_page + pageblock_nr_pages - 1;
1515 end_pfn = start_pfn + pageblock_nr_pages - 1;
c361be55
MG
1516
1517 /* Do not cross zone boundaries */
108bcc96 1518 if (!zone_spans_pfn(zone, start_pfn))
c361be55 1519 start_page = page;
108bcc96 1520 if (!zone_spans_pfn(zone, end_pfn))
c361be55
MG
1521 return 0;
1522
1523 return move_freepages(zone, start_page, end_page, migratetype);
1524}
1525
2f66a68f
MG
1526static void change_pageblock_range(struct page *pageblock_page,
1527 int start_order, int migratetype)
1528{
1529 int nr_pageblocks = 1 << (start_order - pageblock_order);
1530
1531 while (nr_pageblocks--) {
1532 set_pageblock_migratetype(pageblock_page, migratetype);
1533 pageblock_page += pageblock_nr_pages;
1534 }
1535}
1536
fef903ef 1537/*
9c0415eb
VB
1538 * When we are falling back to another migratetype during allocation, try to
1539 * steal extra free pages from the same pageblocks to satisfy further
1540 * allocations, instead of polluting multiple pageblocks.
1541 *
1542 * If we are stealing a relatively large buddy page, it is likely there will
1543 * be more free pages in the pageblock, so try to steal them all. For
1544 * reclaimable and unmovable allocations, we steal regardless of page size,
1545 * as fragmentation caused by those allocations polluting movable pageblocks
1546 * is worse than movable allocations stealing from unmovable and reclaimable
1547 * pageblocks.
fef903ef 1548 */
4eb7dce6
JK
1549static bool can_steal_fallback(unsigned int order, int start_mt)
1550{
1551 /*
1552 * Leaving this order check is intended, although there is
1553 * relaxed order check in next check. The reason is that
1554 * we can actually steal whole pageblock if this condition met,
1555 * but, below check doesn't guarantee it and that is just heuristic
1556 * so could be changed anytime.
1557 */
1558 if (order >= pageblock_order)
1559 return true;
1560
1561 if (order >= pageblock_order / 2 ||
1562 start_mt == MIGRATE_RECLAIMABLE ||
1563 start_mt == MIGRATE_UNMOVABLE ||
1564 page_group_by_mobility_disabled)
1565 return true;
1566
1567 return false;
1568}
1569
1570/*
1571 * This function implements actual steal behaviour. If order is large enough,
1572 * we can steal whole pageblock. If not, we first move freepages in this
1573 * pageblock and check whether half of pages are moved or not. If half of
1574 * pages are moved, we can change migratetype of pageblock and permanently
1575 * use it's pages as requested migratetype in the future.
1576 */
1577static void steal_suitable_fallback(struct zone *zone, struct page *page,
1578 int start_type)
fef903ef 1579{
d00181b9 1580 unsigned int current_order = page_order(page);
4eb7dce6 1581 int pages;
fef903ef 1582
fef903ef
SB
1583 /* Take ownership for orders >= pageblock_order */
1584 if (current_order >= pageblock_order) {
1585 change_pageblock_range(page, current_order, start_type);
3a1086fb 1586 return;
fef903ef
SB
1587 }
1588
4eb7dce6 1589 pages = move_freepages_block(zone, page, start_type);
fef903ef 1590
4eb7dce6
JK
1591 /* Claim the whole block if over half of it is free */
1592 if (pages >= (1 << (pageblock_order-1)) ||
1593 page_group_by_mobility_disabled)
1594 set_pageblock_migratetype(page, start_type);
1595}
1596
2149cdae
JK
1597/*
1598 * Check whether there is a suitable fallback freepage with requested order.
1599 * If only_stealable is true, this function returns fallback_mt only if
1600 * we can steal other freepages all together. This would help to reduce
1601 * fragmentation due to mixed migratetype pages in one pageblock.
1602 */
1603int find_suitable_fallback(struct free_area *area, unsigned int order,
1604 int migratetype, bool only_stealable, bool *can_steal)
4eb7dce6
JK
1605{
1606 int i;
1607 int fallback_mt;
1608
1609 if (area->nr_free == 0)
1610 return -1;
1611
1612 *can_steal = false;
1613 for (i = 0;; i++) {
1614 fallback_mt = fallbacks[migratetype][i];
974a786e 1615 if (fallback_mt == MIGRATE_TYPES)
4eb7dce6
JK
1616 break;
1617
1618 if (list_empty(&area->free_list[fallback_mt]))
1619 continue;
fef903ef 1620
4eb7dce6
JK
1621 if (can_steal_fallback(order, migratetype))
1622 *can_steal = true;
1623
2149cdae
JK
1624 if (!only_stealable)
1625 return fallback_mt;
1626
1627 if (*can_steal)
1628 return fallback_mt;
fef903ef 1629 }
4eb7dce6
JK
1630
1631 return -1;
fef903ef
SB
1632}
1633
0aaa29a5
MG
1634/*
1635 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1636 * there are no empty page blocks that contain a page with a suitable order
1637 */
1638static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1639 unsigned int alloc_order)
1640{
1641 int mt;
1642 unsigned long max_managed, flags;
1643
1644 /*
1645 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1646 * Check is race-prone but harmless.
1647 */
1648 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1649 if (zone->nr_reserved_highatomic >= max_managed)
1650 return;
1651
1652 spin_lock_irqsave(&zone->lock, flags);
1653
1654 /* Recheck the nr_reserved_highatomic limit under the lock */
1655 if (zone->nr_reserved_highatomic >= max_managed)
1656 goto out_unlock;
1657
1658 /* Yoink! */
1659 mt = get_pageblock_migratetype(page);
1660 if (mt != MIGRATE_HIGHATOMIC &&
1661 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1662 zone->nr_reserved_highatomic += pageblock_nr_pages;
1663 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1664 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1665 }
1666
1667out_unlock:
1668 spin_unlock_irqrestore(&zone->lock, flags);
1669}
1670
1671/*
1672 * Used when an allocation is about to fail under memory pressure. This
1673 * potentially hurts the reliability of high-order allocations when under
1674 * intense memory pressure but failed atomic allocations should be easier
1675 * to recover from than an OOM.
1676 */
1677static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1678{
1679 struct zonelist *zonelist = ac->zonelist;
1680 unsigned long flags;
1681 struct zoneref *z;
1682 struct zone *zone;
1683 struct page *page;
1684 int order;
1685
1686 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1687 ac->nodemask) {
1688 /* Preserve at least one pageblock */
1689 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1690 continue;
1691
1692 spin_lock_irqsave(&zone->lock, flags);
1693 for (order = 0; order < MAX_ORDER; order++) {
1694 struct free_area *area = &(zone->free_area[order]);
1695
1696 if (list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
1697 continue;
1698
1699 page = list_entry(area->free_list[MIGRATE_HIGHATOMIC].next,
1700 struct page, lru);
1701
1702 /*
1703 * It should never happen but changes to locking could
1704 * inadvertently allow a per-cpu drain to add pages
1705 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1706 * and watch for underflows.
1707 */
1708 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1709 zone->nr_reserved_highatomic);
1710
1711 /*
1712 * Convert to ac->migratetype and avoid the normal
1713 * pageblock stealing heuristics. Minimally, the caller
1714 * is doing the work and needs the pages. More
1715 * importantly, if the block was always converted to
1716 * MIGRATE_UNMOVABLE or another type then the number
1717 * of pageblocks that cannot be completely freed
1718 * may increase.
1719 */
1720 set_pageblock_migratetype(page, ac->migratetype);
1721 move_freepages_block(zone, page, ac->migratetype);
1722 spin_unlock_irqrestore(&zone->lock, flags);
1723 return;
1724 }
1725 spin_unlock_irqrestore(&zone->lock, flags);
1726 }
1727}
1728
b2a0ac88 1729/* Remove an element from the buddy allocator from the fallback list */
0ac3a409 1730static inline struct page *
7aeb09f9 1731__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
b2a0ac88 1732{
b8af2941 1733 struct free_area *area;
7aeb09f9 1734 unsigned int current_order;
b2a0ac88 1735 struct page *page;
4eb7dce6
JK
1736 int fallback_mt;
1737 bool can_steal;
b2a0ac88
MG
1738
1739 /* Find the largest possible block of pages in the other list */
7aeb09f9
MG
1740 for (current_order = MAX_ORDER-1;
1741 current_order >= order && current_order <= MAX_ORDER-1;
1742 --current_order) {
4eb7dce6
JK
1743 area = &(zone->free_area[current_order]);
1744 fallback_mt = find_suitable_fallback(area, current_order,
2149cdae 1745 start_migratetype, false, &can_steal);
4eb7dce6
JK
1746 if (fallback_mt == -1)
1747 continue;
b2a0ac88 1748
4eb7dce6
JK
1749 page = list_entry(area->free_list[fallback_mt].next,
1750 struct page, lru);
1751 if (can_steal)
1752 steal_suitable_fallback(zone, page, start_migratetype);
b2a0ac88 1753
4eb7dce6
JK
1754 /* Remove the page from the freelists */
1755 area->nr_free--;
1756 list_del(&page->lru);
1757 rmv_page_order(page);
3a1086fb 1758
4eb7dce6
JK
1759 expand(zone, page, order, current_order, area,
1760 start_migratetype);
1761 /*
bb14c2c7 1762 * The pcppage_migratetype may differ from pageblock's
4eb7dce6 1763 * migratetype depending on the decisions in
bb14c2c7
VB
1764 * find_suitable_fallback(). This is OK as long as it does not
1765 * differ for MIGRATE_CMA pageblocks. Those can be used as
1766 * fallback only via special __rmqueue_cma_fallback() function
4eb7dce6 1767 */
bb14c2c7 1768 set_pcppage_migratetype(page, start_migratetype);
e0fff1bd 1769
4eb7dce6
JK
1770 trace_mm_page_alloc_extfrag(page, order, current_order,
1771 start_migratetype, fallback_mt);
e0fff1bd 1772
4eb7dce6 1773 return page;
b2a0ac88
MG
1774 }
1775
728ec980 1776 return NULL;
b2a0ac88
MG
1777}
1778
56fd56b8 1779/*
1da177e4
LT
1780 * Do the hard work of removing an element from the buddy allocator.
1781 * Call me with the zone->lock already held.
1782 */
b2a0ac88 1783static struct page *__rmqueue(struct zone *zone, unsigned int order,
0aaa29a5 1784 int migratetype, gfp_t gfp_flags)
1da177e4 1785{
1da177e4
LT
1786 struct page *page;
1787
56fd56b8 1788 page = __rmqueue_smallest(zone, order, migratetype);
974a786e 1789 if (unlikely(!page)) {
dc67647b
JK
1790 if (migratetype == MIGRATE_MOVABLE)
1791 page = __rmqueue_cma_fallback(zone, order);
1792
1793 if (!page)
1794 page = __rmqueue_fallback(zone, order, migratetype);
728ec980
MG
1795 }
1796
0d3d062a 1797 trace_mm_page_alloc_zone_locked(page, order, migratetype);
b2a0ac88 1798 return page;
1da177e4
LT
1799}
1800
5f63b720 1801/*
1da177e4
LT
1802 * Obtain a specified number of elements from the buddy allocator, all under
1803 * a single hold of the lock, for efficiency. Add them to the supplied list.
1804 * Returns the number of new pages which were placed at *list.
1805 */
5f63b720 1806static int rmqueue_bulk(struct zone *zone, unsigned int order,
b2a0ac88 1807 unsigned long count, struct list_head *list,
b745bc85 1808 int migratetype, bool cold)
1da177e4 1809{
5bcc9f86 1810 int i;
5f63b720 1811
c54ad30c 1812 spin_lock(&zone->lock);
1da177e4 1813 for (i = 0; i < count; ++i) {
0aaa29a5 1814 struct page *page = __rmqueue(zone, order, migratetype, 0);
085cc7d5 1815 if (unlikely(page == NULL))
1da177e4 1816 break;
81eabcbe
MG
1817
1818 /*
1819 * Split buddy pages returned by expand() are received here
1820 * in physical page order. The page is added to the callers and
1821 * list and the list head then moves forward. From the callers
1822 * perspective, the linked list is ordered by page number in
1823 * some conditions. This is useful for IO devices that can
1824 * merge IO requests if the physical pages are ordered
1825 * properly.
1826 */
b745bc85 1827 if (likely(!cold))
e084b2d9
MG
1828 list_add(&page->lru, list);
1829 else
1830 list_add_tail(&page->lru, list);
81eabcbe 1831 list = &page->lru;
bb14c2c7 1832 if (is_migrate_cma(get_pcppage_migratetype(page)))
d1ce749a
BZ
1833 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1834 -(1 << order));
1da177e4 1835 }
f2260e6b 1836 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
c54ad30c 1837 spin_unlock(&zone->lock);
085cc7d5 1838 return i;
1da177e4
LT
1839}
1840
4ae7c039 1841#ifdef CONFIG_NUMA
8fce4d8e 1842/*
4037d452
CL
1843 * Called from the vmstat counter updater to drain pagesets of this
1844 * currently executing processor on remote nodes after they have
1845 * expired.
1846 *
879336c3
CL
1847 * Note that this function must be called with the thread pinned to
1848 * a single processor.
8fce4d8e 1849 */
4037d452 1850void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
4ae7c039 1851{
4ae7c039 1852 unsigned long flags;
7be12fc9 1853 int to_drain, batch;
4ae7c039 1854
4037d452 1855 local_irq_save(flags);
4db0c3c2 1856 batch = READ_ONCE(pcp->batch);
7be12fc9 1857 to_drain = min(pcp->count, batch);
2a13515c
KM
1858 if (to_drain > 0) {
1859 free_pcppages_bulk(zone, to_drain, pcp);
1860 pcp->count -= to_drain;
1861 }
4037d452 1862 local_irq_restore(flags);
4ae7c039
CL
1863}
1864#endif
1865
9f8f2172 1866/*
93481ff0 1867 * Drain pcplists of the indicated processor and zone.
9f8f2172
CL
1868 *
1869 * The processor must either be the current processor and the
1870 * thread pinned to the current processor or a processor that
1871 * is not online.
1872 */
93481ff0 1873static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1da177e4 1874{
c54ad30c 1875 unsigned long flags;
93481ff0
VB
1876 struct per_cpu_pageset *pset;
1877 struct per_cpu_pages *pcp;
1da177e4 1878
93481ff0
VB
1879 local_irq_save(flags);
1880 pset = per_cpu_ptr(zone->pageset, cpu);
1da177e4 1881
93481ff0
VB
1882 pcp = &pset->pcp;
1883 if (pcp->count) {
1884 free_pcppages_bulk(zone, pcp->count, pcp);
1885 pcp->count = 0;
1886 }
1887 local_irq_restore(flags);
1888}
3dfa5721 1889
93481ff0
VB
1890/*
1891 * Drain pcplists of all zones on the indicated processor.
1892 *
1893 * The processor must either be the current processor and the
1894 * thread pinned to the current processor or a processor that
1895 * is not online.
1896 */
1897static void drain_pages(unsigned int cpu)
1898{
1899 struct zone *zone;
1900
1901 for_each_populated_zone(zone) {
1902 drain_pages_zone(cpu, zone);
1da177e4
LT
1903 }
1904}
1da177e4 1905
9f8f2172
CL
1906/*
1907 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
93481ff0
VB
1908 *
1909 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1910 * the single zone's pages.
9f8f2172 1911 */
93481ff0 1912void drain_local_pages(struct zone *zone)
9f8f2172 1913{
93481ff0
VB
1914 int cpu = smp_processor_id();
1915
1916 if (zone)
1917 drain_pages_zone(cpu, zone);
1918 else
1919 drain_pages(cpu);
9f8f2172
CL
1920}
1921
1922/*
74046494
GBY
1923 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1924 *
93481ff0
VB
1925 * When zone parameter is non-NULL, spill just the single zone's pages.
1926 *
74046494
GBY
1927 * Note that this code is protected against sending an IPI to an offline
1928 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1929 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1930 * nothing keeps CPUs from showing up after we populated the cpumask and
1931 * before the call to on_each_cpu_mask().
9f8f2172 1932 */
93481ff0 1933void drain_all_pages(struct zone *zone)
9f8f2172 1934{
74046494 1935 int cpu;
74046494
GBY
1936
1937 /*
1938 * Allocate in the BSS so we wont require allocation in
1939 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1940 */
1941 static cpumask_t cpus_with_pcps;
1942
1943 /*
1944 * We don't care about racing with CPU hotplug event
1945 * as offline notification will cause the notified
1946 * cpu to drain that CPU pcps and on_each_cpu_mask
1947 * disables preemption as part of its processing
1948 */
1949 for_each_online_cpu(cpu) {
93481ff0
VB
1950 struct per_cpu_pageset *pcp;
1951 struct zone *z;
74046494 1952 bool has_pcps = false;
93481ff0
VB
1953
1954 if (zone) {
74046494 1955 pcp = per_cpu_ptr(zone->pageset, cpu);
93481ff0 1956 if (pcp->pcp.count)
74046494 1957 has_pcps = true;
93481ff0
VB
1958 } else {
1959 for_each_populated_zone(z) {
1960 pcp = per_cpu_ptr(z->pageset, cpu);
1961 if (pcp->pcp.count) {
1962 has_pcps = true;
1963 break;
1964 }
74046494
GBY
1965 }
1966 }
93481ff0 1967
74046494
GBY
1968 if (has_pcps)
1969 cpumask_set_cpu(cpu, &cpus_with_pcps);
1970 else
1971 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1972 }
93481ff0
VB
1973 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1974 zone, 1);
9f8f2172
CL
1975}
1976
296699de 1977#ifdef CONFIG_HIBERNATION
1da177e4
LT
1978
1979void mark_free_pages(struct zone *zone)
1980{
f623f0db
RW
1981 unsigned long pfn, max_zone_pfn;
1982 unsigned long flags;
7aeb09f9 1983 unsigned int order, t;
1da177e4
LT
1984 struct list_head *curr;
1985
8080fc03 1986 if (zone_is_empty(zone))
1da177e4
LT
1987 return;
1988
1989 spin_lock_irqsave(&zone->lock, flags);
f623f0db 1990
108bcc96 1991 max_zone_pfn = zone_end_pfn(zone);
f623f0db
RW
1992 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1993 if (pfn_valid(pfn)) {
1994 struct page *page = pfn_to_page(pfn);
1995
7be98234
RW
1996 if (!swsusp_page_is_forbidden(page))
1997 swsusp_unset_page_free(page);
f623f0db 1998 }
1da177e4 1999
b2a0ac88
MG
2000 for_each_migratetype_order(order, t) {
2001 list_for_each(curr, &zone->free_area[order].free_list[t]) {
f623f0db 2002 unsigned long i;
1da177e4 2003
f623f0db
RW
2004 pfn = page_to_pfn(list_entry(curr, struct page, lru));
2005 for (i = 0; i < (1UL << order); i++)
7be98234 2006 swsusp_set_page_free(pfn_to_page(pfn + i));
f623f0db 2007 }
b2a0ac88 2008 }
1da177e4
LT
2009 spin_unlock_irqrestore(&zone->lock, flags);
2010}
e2c55dc8 2011#endif /* CONFIG_PM */
1da177e4 2012
1da177e4
LT
2013/*
2014 * Free a 0-order page
b745bc85 2015 * cold == true ? free a cold page : free a hot page
1da177e4 2016 */
b745bc85 2017void free_hot_cold_page(struct page *page, bool cold)
1da177e4
LT
2018{
2019 struct zone *zone = page_zone(page);
2020 struct per_cpu_pages *pcp;
2021 unsigned long flags;
dc4b0caf 2022 unsigned long pfn = page_to_pfn(page);
5f8dcc21 2023 int migratetype;
1da177e4 2024
ec95f53a 2025 if (!free_pages_prepare(page, 0))
689bcebf
HD
2026 return;
2027
dc4b0caf 2028 migratetype = get_pfnblock_migratetype(page, pfn);
bb14c2c7 2029 set_pcppage_migratetype(page, migratetype);
1da177e4 2030 local_irq_save(flags);
f8891e5e 2031 __count_vm_event(PGFREE);
da456f14 2032
5f8dcc21
MG
2033 /*
2034 * We only track unmovable, reclaimable and movable on pcp lists.
2035 * Free ISOLATE pages back to the allocator because they are being
2036 * offlined but treat RESERVE as movable pages so we can get those
2037 * areas back if necessary. Otherwise, we may have to free
2038 * excessively into the page allocator
2039 */
2040 if (migratetype >= MIGRATE_PCPTYPES) {
194159fb 2041 if (unlikely(is_migrate_isolate(migratetype))) {
dc4b0caf 2042 free_one_page(zone, page, pfn, 0, migratetype);
5f8dcc21
MG
2043 goto out;
2044 }
2045 migratetype = MIGRATE_MOVABLE;
2046 }
2047
99dcc3e5 2048 pcp = &this_cpu_ptr(zone->pageset)->pcp;
b745bc85 2049 if (!cold)
5f8dcc21 2050 list_add(&page->lru, &pcp->lists[migratetype]);
b745bc85
MG
2051 else
2052 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1da177e4 2053 pcp->count++;
48db57f8 2054 if (pcp->count >= pcp->high) {
4db0c3c2 2055 unsigned long batch = READ_ONCE(pcp->batch);
998d39cb
CS
2056 free_pcppages_bulk(zone, batch, pcp);
2057 pcp->count -= batch;
48db57f8 2058 }
5f8dcc21
MG
2059
2060out:
1da177e4 2061 local_irq_restore(flags);
1da177e4
LT
2062}
2063
cc59850e
KK
2064/*
2065 * Free a list of 0-order pages
2066 */
b745bc85 2067void free_hot_cold_page_list(struct list_head *list, bool cold)
cc59850e
KK
2068{
2069 struct page *page, *next;
2070
2071 list_for_each_entry_safe(page, next, list, lru) {
b413d48a 2072 trace_mm_page_free_batched(page, cold);
cc59850e
KK
2073 free_hot_cold_page(page, cold);
2074 }
2075}
2076
8dfcc9ba
NP
2077/*
2078 * split_page takes a non-compound higher-order page, and splits it into
2079 * n (1<<order) sub-pages: page[0..n]
2080 * Each sub-page must be freed individually.
2081 *
2082 * Note: this is probably too low level an operation for use in drivers.
2083 * Please consult with lkml before using this in your driver.
2084 */
2085void split_page(struct page *page, unsigned int order)
2086{
2087 int i;
e2cfc911 2088 gfp_t gfp_mask;
8dfcc9ba 2089
309381fe
SL
2090 VM_BUG_ON_PAGE(PageCompound(page), page);
2091 VM_BUG_ON_PAGE(!page_count(page), page);
b1eeab67
VN
2092
2093#ifdef CONFIG_KMEMCHECK
2094 /*
2095 * Split shadow pages too, because free(page[0]) would
2096 * otherwise free the whole shadow.
2097 */
2098 if (kmemcheck_page_is_tracked(page))
2099 split_page(virt_to_page(page[0].shadow), order);
2100#endif
2101
e2cfc911
JK
2102 gfp_mask = get_page_owner_gfp(page);
2103 set_page_owner(page, 0, gfp_mask);
48c96a36 2104 for (i = 1; i < (1 << order); i++) {
7835e98b 2105 set_page_refcounted(page + i);
e2cfc911 2106 set_page_owner(page + i, 0, gfp_mask);
48c96a36 2107 }
8dfcc9ba 2108}
5853ff23 2109EXPORT_SYMBOL_GPL(split_page);
8dfcc9ba 2110
3c605096 2111int __isolate_free_page(struct page *page, unsigned int order)
748446bb 2112{
748446bb
MG
2113 unsigned long watermark;
2114 struct zone *zone;
2139cbe6 2115 int mt;
748446bb
MG
2116
2117 BUG_ON(!PageBuddy(page));
2118
2119 zone = page_zone(page);
2e30abd1 2120 mt = get_pageblock_migratetype(page);
748446bb 2121
194159fb 2122 if (!is_migrate_isolate(mt)) {
2e30abd1
MS
2123 /* Obey watermarks as if the page was being allocated */
2124 watermark = low_wmark_pages(zone) + (1 << order);
2125 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2126 return 0;
2127
8fb74b9f 2128 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2e30abd1 2129 }
748446bb
MG
2130
2131 /* Remove page from free list */
2132 list_del(&page->lru);
2133 zone->free_area[order].nr_free--;
2134 rmv_page_order(page);
2139cbe6 2135
e2cfc911 2136 set_page_owner(page, order, __GFP_MOVABLE);
f3a14ced 2137
8fb74b9f 2138 /* Set the pageblock if the isolated page is at least a pageblock */
748446bb
MG
2139 if (order >= pageblock_order - 1) {
2140 struct page *endpage = page + (1 << order) - 1;
47118af0
MN
2141 for (; page < endpage; page += pageblock_nr_pages) {
2142 int mt = get_pageblock_migratetype(page);
194159fb 2143 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
47118af0
MN
2144 set_pageblock_migratetype(page,
2145 MIGRATE_MOVABLE);
2146 }
748446bb
MG
2147 }
2148
f3a14ced 2149
8fb74b9f 2150 return 1UL << order;
1fb3f8ca
MG
2151}
2152
2153/*
2154 * Similar to split_page except the page is already free. As this is only
2155 * being used for migration, the migratetype of the block also changes.
2156 * As this is called with interrupts disabled, the caller is responsible
2157 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2158 * are enabled.
2159 *
2160 * Note: this is probably too low level an operation for use in drivers.
2161 * Please consult with lkml before using this in your driver.
2162 */
2163int split_free_page(struct page *page)
2164{
2165 unsigned int order;
2166 int nr_pages;
2167
1fb3f8ca
MG
2168 order = page_order(page);
2169
8fb74b9f 2170 nr_pages = __isolate_free_page(page, order);
1fb3f8ca
MG
2171 if (!nr_pages)
2172 return 0;
2173
2174 /* Split into individual pages */
2175 set_page_refcounted(page);
2176 split_page(page, order);
2177 return nr_pages;
748446bb
MG
2178}
2179
1da177e4 2180/*
75379191 2181 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
1da177e4 2182 */
0a15c3e9
MG
2183static inline
2184struct page *buffered_rmqueue(struct zone *preferred_zone,
7aeb09f9 2185 struct zone *zone, unsigned int order,
0aaa29a5 2186 gfp_t gfp_flags, int alloc_flags, int migratetype)
1da177e4
LT
2187{
2188 unsigned long flags;
689bcebf 2189 struct page *page;
b745bc85 2190 bool cold = ((gfp_flags & __GFP_COLD) != 0);
1da177e4 2191
48db57f8 2192 if (likely(order == 0)) {
1da177e4 2193 struct per_cpu_pages *pcp;
5f8dcc21 2194 struct list_head *list;
1da177e4 2195
1da177e4 2196 local_irq_save(flags);
99dcc3e5
CL
2197 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2198 list = &pcp->lists[migratetype];
5f8dcc21 2199 if (list_empty(list)) {
535131e6 2200 pcp->count += rmqueue_bulk(zone, 0,
5f8dcc21 2201 pcp->batch, list,
e084b2d9 2202 migratetype, cold);
5f8dcc21 2203 if (unlikely(list_empty(list)))
6fb332fa 2204 goto failed;
535131e6 2205 }
b92a6edd 2206
5f8dcc21
MG
2207 if (cold)
2208 page = list_entry(list->prev, struct page, lru);
2209 else
2210 page = list_entry(list->next, struct page, lru);
2211
b92a6edd
MG
2212 list_del(&page->lru);
2213 pcp->count--;
7fb1d9fc 2214 } else {
dab48dab
AM
2215 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2216 /*
2217 * __GFP_NOFAIL is not to be used in new code.
2218 *
2219 * All __GFP_NOFAIL callers should be fixed so that they
2220 * properly detect and handle allocation failures.
2221 *
2222 * We most definitely don't want callers attempting to
4923abf9 2223 * allocate greater than order-1 page units with
dab48dab
AM
2224 * __GFP_NOFAIL.
2225 */
4923abf9 2226 WARN_ON_ONCE(order > 1);
dab48dab 2227 }
1da177e4 2228 spin_lock_irqsave(&zone->lock, flags);
0aaa29a5
MG
2229
2230 page = NULL;
2231 if (alloc_flags & ALLOC_HARDER) {
2232 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2233 if (page)
2234 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2235 }
2236 if (!page)
2237 page = __rmqueue(zone, order, migratetype, gfp_flags);
a74609fa
NP
2238 spin_unlock(&zone->lock);
2239 if (!page)
2240 goto failed;
d1ce749a 2241 __mod_zone_freepage_state(zone, -(1 << order),
bb14c2c7 2242 get_pcppage_migratetype(page));
1da177e4
LT
2243 }
2244
3a025760 2245 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
abe5f972 2246 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
57054651
JW
2247 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2248 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
27329369 2249
f8891e5e 2250 __count_zone_vm_events(PGALLOC, zone, 1 << order);
78afd561 2251 zone_statistics(preferred_zone, zone, gfp_flags);
a74609fa 2252 local_irq_restore(flags);
1da177e4 2253
309381fe 2254 VM_BUG_ON_PAGE(bad_range(zone, page), page);
1da177e4 2255 return page;
a74609fa
NP
2256
2257failed:
2258 local_irq_restore(flags);
a74609fa 2259 return NULL;
1da177e4
LT
2260}
2261
933e312e
AM
2262#ifdef CONFIG_FAIL_PAGE_ALLOC
2263
b2588c4b 2264static struct {
933e312e
AM
2265 struct fault_attr attr;
2266
621a5f7a 2267 bool ignore_gfp_highmem;
71baba4b 2268 bool ignore_gfp_reclaim;
54114994 2269 u32 min_order;
933e312e
AM
2270} fail_page_alloc = {
2271 .attr = FAULT_ATTR_INITIALIZER,
71baba4b 2272 .ignore_gfp_reclaim = true,
621a5f7a 2273 .ignore_gfp_highmem = true,
54114994 2274 .min_order = 1,
933e312e
AM
2275};
2276
2277static int __init setup_fail_page_alloc(char *str)
2278{
2279 return setup_fault_attr(&fail_page_alloc.attr, str);
2280}
2281__setup("fail_page_alloc=", setup_fail_page_alloc);
2282
deaf386e 2283static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
933e312e 2284{
54114994 2285 if (order < fail_page_alloc.min_order)
deaf386e 2286 return false;
933e312e 2287 if (gfp_mask & __GFP_NOFAIL)
deaf386e 2288 return false;
933e312e 2289 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
deaf386e 2290 return false;
71baba4b
MG
2291 if (fail_page_alloc.ignore_gfp_reclaim &&
2292 (gfp_mask & __GFP_DIRECT_RECLAIM))
deaf386e 2293 return false;
933e312e
AM
2294
2295 return should_fail(&fail_page_alloc.attr, 1 << order);
2296}
2297
2298#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2299
2300static int __init fail_page_alloc_debugfs(void)
2301{
f4ae40a6 2302 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
933e312e 2303 struct dentry *dir;
933e312e 2304
dd48c085
AM
2305 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2306 &fail_page_alloc.attr);
2307 if (IS_ERR(dir))
2308 return PTR_ERR(dir);
933e312e 2309
b2588c4b 2310 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
71baba4b 2311 &fail_page_alloc.ignore_gfp_reclaim))
b2588c4b
AM
2312 goto fail;
2313 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2314 &fail_page_alloc.ignore_gfp_highmem))
2315 goto fail;
2316 if (!debugfs_create_u32("min-order", mode, dir,
2317 &fail_page_alloc.min_order))
2318 goto fail;
2319
2320 return 0;
2321fail:
dd48c085 2322 debugfs_remove_recursive(dir);
933e312e 2323
b2588c4b 2324 return -ENOMEM;
933e312e
AM
2325}
2326
2327late_initcall(fail_page_alloc_debugfs);
2328
2329#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2330
2331#else /* CONFIG_FAIL_PAGE_ALLOC */
2332
deaf386e 2333static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
933e312e 2334{
deaf386e 2335 return false;
933e312e
AM
2336}
2337
2338#endif /* CONFIG_FAIL_PAGE_ALLOC */
2339
1da177e4 2340/*
97a16fc8
MG
2341 * Return true if free base pages are above 'mark'. For high-order checks it
2342 * will return true of the order-0 watermark is reached and there is at least
2343 * one free page of a suitable size. Checking now avoids taking the zone lock
2344 * to check in the allocation paths if no pages are free.
1da177e4 2345 */
7aeb09f9
MG
2346static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2347 unsigned long mark, int classzone_idx, int alloc_flags,
2348 long free_pages)
1da177e4 2349{
d23ad423 2350 long min = mark;
1da177e4 2351 int o;
97a16fc8 2352 const int alloc_harder = (alloc_flags & ALLOC_HARDER);
1da177e4 2353
0aaa29a5 2354 /* free_pages may go negative - that's OK */
df0a6daa 2355 free_pages -= (1 << order) - 1;
0aaa29a5 2356
7fb1d9fc 2357 if (alloc_flags & ALLOC_HIGH)
1da177e4 2358 min -= min / 2;
0aaa29a5
MG
2359
2360 /*
2361 * If the caller does not have rights to ALLOC_HARDER then subtract
2362 * the high-atomic reserves. This will over-estimate the size of the
2363 * atomic reserve but it avoids a search.
2364 */
97a16fc8 2365 if (likely(!alloc_harder))
0aaa29a5
MG
2366 free_pages -= z->nr_reserved_highatomic;
2367 else
1da177e4 2368 min -= min / 4;
e2b19197 2369
d95ea5d1
BZ
2370#ifdef CONFIG_CMA
2371 /* If allocation can't use CMA areas don't use free CMA pages */
2372 if (!(alloc_flags & ALLOC_CMA))
97a16fc8 2373 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
d95ea5d1 2374#endif
026b0814 2375
97a16fc8
MG
2376 /*
2377 * Check watermarks for an order-0 allocation request. If these
2378 * are not met, then a high-order request also cannot go ahead
2379 * even if a suitable page happened to be free.
2380 */
2381 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
88f5acf8 2382 return false;
1da177e4 2383
97a16fc8
MG
2384 /* If this is an order-0 request then the watermark is fine */
2385 if (!order)
2386 return true;
2387
2388 /* For a high-order request, check at least one suitable page is free */
2389 for (o = order; o < MAX_ORDER; o++) {
2390 struct free_area *area = &z->free_area[o];
2391 int mt;
2392
2393 if (!area->nr_free)
2394 continue;
2395
2396 if (alloc_harder)
2397 return true;
1da177e4 2398
97a16fc8
MG
2399 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2400 if (!list_empty(&area->free_list[mt]))
2401 return true;
2402 }
2403
2404#ifdef CONFIG_CMA
2405 if ((alloc_flags & ALLOC_CMA) &&
2406 !list_empty(&area->free_list[MIGRATE_CMA])) {
2407 return true;
2408 }
2409#endif
1da177e4 2410 }
97a16fc8 2411 return false;
88f5acf8
MG
2412}
2413
7aeb09f9 2414bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
88f5acf8
MG
2415 int classzone_idx, int alloc_flags)
2416{
2417 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2418 zone_page_state(z, NR_FREE_PAGES));
2419}
2420
7aeb09f9 2421bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
e2b19197 2422 unsigned long mark, int classzone_idx)
88f5acf8
MG
2423{
2424 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2425
2426 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2427 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2428
e2b19197 2429 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
88f5acf8 2430 free_pages);
1da177e4
LT
2431}
2432
9276b1bc 2433#ifdef CONFIG_NUMA
81c0a2bb
JW
2434static bool zone_local(struct zone *local_zone, struct zone *zone)
2435{
fff4068c 2436 return local_zone->node == zone->node;
81c0a2bb
JW
2437}
2438
957f822a
DR
2439static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2440{
5f7a75ac
MG
2441 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2442 RECLAIM_DISTANCE;
957f822a 2443}
9276b1bc 2444#else /* CONFIG_NUMA */
81c0a2bb
JW
2445static bool zone_local(struct zone *local_zone, struct zone *zone)
2446{
2447 return true;
2448}
2449
957f822a
DR
2450static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2451{
2452 return true;
2453}
9276b1bc
PJ
2454#endif /* CONFIG_NUMA */
2455
4ffeaf35
MG
2456static void reset_alloc_batches(struct zone *preferred_zone)
2457{
2458 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2459
2460 do {
2461 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2462 high_wmark_pages(zone) - low_wmark_pages(zone) -
2463 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
57054651 2464 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
4ffeaf35
MG
2465 } while (zone++ != preferred_zone);
2466}
2467
7fb1d9fc 2468/*
0798e519 2469 * get_page_from_freelist goes through the zonelist trying to allocate
7fb1d9fc
RS
2470 * a page.
2471 */
2472static struct page *
a9263751
VB
2473get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2474 const struct alloc_context *ac)
753ee728 2475{
a9263751 2476 struct zonelist *zonelist = ac->zonelist;
dd1a239f 2477 struct zoneref *z;
7fb1d9fc 2478 struct page *page = NULL;
5117f45d 2479 struct zone *zone;
4ffeaf35
MG
2480 int nr_fair_skipped = 0;
2481 bool zonelist_rescan;
54a6eb5c 2482
9276b1bc 2483zonelist_scan:
4ffeaf35
MG
2484 zonelist_rescan = false;
2485
7fb1d9fc 2486 /*
9276b1bc 2487 * Scan zonelist, looking for a zone with enough free.
344736f2 2488 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
7fb1d9fc 2489 */
a9263751
VB
2490 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2491 ac->nodemask) {
e085dbc5
JW
2492 unsigned long mark;
2493
664eedde
MG
2494 if (cpusets_enabled() &&
2495 (alloc_flags & ALLOC_CPUSET) &&
344736f2 2496 !cpuset_zone_allowed(zone, gfp_mask))
cd38b115 2497 continue;
81c0a2bb
JW
2498 /*
2499 * Distribute pages in proportion to the individual
2500 * zone size to ensure fair page aging. The zone a
2501 * page was allocated in should have no effect on the
2502 * time the page has in memory before being reclaimed.
81c0a2bb 2503 */
3a025760 2504 if (alloc_flags & ALLOC_FAIR) {
a9263751 2505 if (!zone_local(ac->preferred_zone, zone))
f7b5d647 2506 break;
57054651 2507 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
4ffeaf35 2508 nr_fair_skipped++;
3a025760 2509 continue;
4ffeaf35 2510 }
81c0a2bb 2511 }
a756cf59
JW
2512 /*
2513 * When allocating a page cache page for writing, we
2514 * want to get it from a zone that is within its dirty
2515 * limit, such that no single zone holds more than its
2516 * proportional share of globally allowed dirty pages.
2517 * The dirty limits take into account the zone's
2518 * lowmem reserves and high watermark so that kswapd
2519 * should be able to balance it without having to
2520 * write pages from its LRU list.
2521 *
2522 * This may look like it could increase pressure on
2523 * lower zones by failing allocations in higher zones
2524 * before they are full. But the pages that do spill
2525 * over are limited as the lower zones are protected
2526 * by this very same mechanism. It should not become
2527 * a practical burden to them.
2528 *
2529 * XXX: For now, allow allocations to potentially
2530 * exceed the per-zone dirty limit in the slowpath
c9ab0c4f 2531 * (spread_dirty_pages unset) before going into reclaim,
a756cf59
JW
2532 * which is important when on a NUMA setup the allowed
2533 * zones are together not big enough to reach the
2534 * global limit. The proper fix for these situations
2535 * will require awareness of zones in the
2536 * dirty-throttling and the flusher threads.
2537 */
c9ab0c4f 2538 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
800a1e75 2539 continue;
7fb1d9fc 2540
e085dbc5
JW
2541 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2542 if (!zone_watermark_ok(zone, order, mark,
a9263751 2543 ac->classzone_idx, alloc_flags)) {
fa5e084e
MG
2544 int ret;
2545
5dab2911
MG
2546 /* Checked here to keep the fast path fast */
2547 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2548 if (alloc_flags & ALLOC_NO_WATERMARKS)
2549 goto try_this_zone;
2550
957f822a 2551 if (zone_reclaim_mode == 0 ||
a9263751 2552 !zone_allows_reclaim(ac->preferred_zone, zone))
cd38b115
MG
2553 continue;
2554
fa5e084e
MG
2555 ret = zone_reclaim(zone, gfp_mask, order);
2556 switch (ret) {
2557 case ZONE_RECLAIM_NOSCAN:
2558 /* did not scan */
cd38b115 2559 continue;
fa5e084e
MG
2560 case ZONE_RECLAIM_FULL:
2561 /* scanned but unreclaimable */
cd38b115 2562 continue;
fa5e084e
MG
2563 default:
2564 /* did we reclaim enough */
fed2719e 2565 if (zone_watermark_ok(zone, order, mark,
a9263751 2566 ac->classzone_idx, alloc_flags))
fed2719e
MG
2567 goto try_this_zone;
2568
fed2719e 2569 continue;
0798e519 2570 }
7fb1d9fc
RS
2571 }
2572
fa5e084e 2573try_this_zone:
a9263751 2574 page = buffered_rmqueue(ac->preferred_zone, zone, order,
0aaa29a5 2575 gfp_mask, alloc_flags, ac->migratetype);
75379191
VB
2576 if (page) {
2577 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2578 goto try_this_zone;
0aaa29a5
MG
2579
2580 /*
2581 * If this is a high-order atomic allocation then check
2582 * if the pageblock should be reserved for the future
2583 */
2584 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2585 reserve_highatomic_pageblock(page, zone, order);
2586
75379191
VB
2587 return page;
2588 }
54a6eb5c 2589 }
9276b1bc 2590
4ffeaf35
MG
2591 /*
2592 * The first pass makes sure allocations are spread fairly within the
2593 * local node. However, the local node might have free pages left
2594 * after the fairness batches are exhausted, and remote zones haven't
2595 * even been considered yet. Try once more without fairness, and
2596 * include remote zones now, before entering the slowpath and waking
2597 * kswapd: prefer spilling to a remote zone over swapping locally.
2598 */
2599 if (alloc_flags & ALLOC_FAIR) {
2600 alloc_flags &= ~ALLOC_FAIR;
2601 if (nr_fair_skipped) {
2602 zonelist_rescan = true;
a9263751 2603 reset_alloc_batches(ac->preferred_zone);
4ffeaf35
MG
2604 }
2605 if (nr_online_nodes > 1)
2606 zonelist_rescan = true;
2607 }
2608
4ffeaf35
MG
2609 if (zonelist_rescan)
2610 goto zonelist_scan;
2611
2612 return NULL;
753ee728
MH
2613}
2614
29423e77
DR
2615/*
2616 * Large machines with many possible nodes should not always dump per-node
2617 * meminfo in irq context.
2618 */
2619static inline bool should_suppress_show_mem(void)
2620{
2621 bool ret = false;
2622
2623#if NODES_SHIFT > 8
2624 ret = in_interrupt();
2625#endif
2626 return ret;
2627}
2628
a238ab5b
DH
2629static DEFINE_RATELIMIT_STATE(nopage_rs,
2630 DEFAULT_RATELIMIT_INTERVAL,
2631 DEFAULT_RATELIMIT_BURST);
2632
d00181b9 2633void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
a238ab5b 2634{
a238ab5b
DH
2635 unsigned int filter = SHOW_MEM_FILTER_NODES;
2636
c0a32fc5
SG
2637 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2638 debug_guardpage_minorder() > 0)
a238ab5b
DH
2639 return;
2640
2641 /*
2642 * This documents exceptions given to allocations in certain
2643 * contexts that are allowed to allocate outside current's set
2644 * of allowed nodes.
2645 */
2646 if (!(gfp_mask & __GFP_NOMEMALLOC))
2647 if (test_thread_flag(TIF_MEMDIE) ||
2648 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2649 filter &= ~SHOW_MEM_FILTER_NODES;
d0164adc 2650 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
a238ab5b
DH
2651 filter &= ~SHOW_MEM_FILTER_NODES;
2652
2653 if (fmt) {
3ee9a4f0
JP
2654 struct va_format vaf;
2655 va_list args;
2656
a238ab5b 2657 va_start(args, fmt);
3ee9a4f0
JP
2658
2659 vaf.fmt = fmt;
2660 vaf.va = &args;
2661
2662 pr_warn("%pV", &vaf);
2663
a238ab5b
DH
2664 va_end(args);
2665 }
2666
d00181b9 2667 pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
3ee9a4f0 2668 current->comm, order, gfp_mask);
a238ab5b
DH
2669
2670 dump_stack();
2671 if (!should_suppress_show_mem())
2672 show_mem(filter);
2673}
2674
11e33f6a
MG
2675static inline struct page *
2676__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
a9263751 2677 const struct alloc_context *ac, unsigned long *did_some_progress)
11e33f6a 2678{
6e0fc46d
DR
2679 struct oom_control oc = {
2680 .zonelist = ac->zonelist,
2681 .nodemask = ac->nodemask,
2682 .gfp_mask = gfp_mask,
2683 .order = order,
6e0fc46d 2684 };
11e33f6a
MG
2685 struct page *page;
2686
9879de73
JW
2687 *did_some_progress = 0;
2688
9879de73 2689 /*
dc56401f
JW
2690 * Acquire the oom lock. If that fails, somebody else is
2691 * making progress for us.
9879de73 2692 */
dc56401f 2693 if (!mutex_trylock(&oom_lock)) {
9879de73 2694 *did_some_progress = 1;
11e33f6a 2695 schedule_timeout_uninterruptible(1);
1da177e4
LT
2696 return NULL;
2697 }
6b1de916 2698
11e33f6a
MG
2699 /*
2700 * Go through the zonelist yet one more time, keep very high watermark
2701 * here, this is only to catch a parallel oom killing, we must fail if
2702 * we're still under heavy pressure.
2703 */
a9263751
VB
2704 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2705 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
7fb1d9fc 2706 if (page)
11e33f6a
MG
2707 goto out;
2708
4365a567 2709 if (!(gfp_mask & __GFP_NOFAIL)) {
9879de73
JW
2710 /* Coredumps can quickly deplete all memory reserves */
2711 if (current->flags & PF_DUMPCORE)
2712 goto out;
4365a567
KH
2713 /* The OOM killer will not help higher order allocs */
2714 if (order > PAGE_ALLOC_COSTLY_ORDER)
2715 goto out;
03668b3c 2716 /* The OOM killer does not needlessly kill tasks for lowmem */
a9263751 2717 if (ac->high_zoneidx < ZONE_NORMAL)
03668b3c 2718 goto out;
9083905a 2719 /* The OOM killer does not compensate for IO-less reclaim */
cc873177
JW
2720 if (!(gfp_mask & __GFP_FS)) {
2721 /*
2722 * XXX: Page reclaim didn't yield anything,
2723 * and the OOM killer can't be invoked, but
9083905a 2724 * keep looping as per tradition.
cc873177
JW
2725 */
2726 *did_some_progress = 1;
9879de73 2727 goto out;
cc873177 2728 }
9083905a
JW
2729 if (pm_suspended_storage())
2730 goto out;
4167e9b2 2731 /* The OOM killer may not free memory on a specific node */
4365a567
KH
2732 if (gfp_mask & __GFP_THISNODE)
2733 goto out;
2734 }
11e33f6a 2735 /* Exhausted what can be done so it's blamo time */
6e0fc46d 2736 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
c32b3cbe 2737 *did_some_progress = 1;
11e33f6a 2738out:
dc56401f 2739 mutex_unlock(&oom_lock);
11e33f6a
MG
2740 return page;
2741}
2742
56de7263
MG
2743#ifdef CONFIG_COMPACTION
2744/* Try memory compaction for high-order allocations before reclaim */
2745static struct page *
2746__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
a9263751
VB
2747 int alloc_flags, const struct alloc_context *ac,
2748 enum migrate_mode mode, int *contended_compaction,
2749 bool *deferred_compaction)
56de7263 2750{
53853e2d 2751 unsigned long compact_result;
98dd3b48 2752 struct page *page;
53853e2d
VB
2753
2754 if (!order)
66199712 2755 return NULL;
66199712 2756
c06b1fca 2757 current->flags |= PF_MEMALLOC;
1a6d53a1
VB
2758 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2759 mode, contended_compaction);
c06b1fca 2760 current->flags &= ~PF_MEMALLOC;
56de7263 2761
98dd3b48
VB
2762 switch (compact_result) {
2763 case COMPACT_DEFERRED:
53853e2d 2764 *deferred_compaction = true;
98dd3b48
VB
2765 /* fall-through */
2766 case COMPACT_SKIPPED:
2767 return NULL;
2768 default:
2769 break;
2770 }
53853e2d 2771
98dd3b48
VB
2772 /*
2773 * At least in one zone compaction wasn't deferred or skipped, so let's
2774 * count a compaction stall
2775 */
2776 count_vm_event(COMPACTSTALL);
8fb74b9f 2777
a9263751
VB
2778 page = get_page_from_freelist(gfp_mask, order,
2779 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
53853e2d 2780
98dd3b48
VB
2781 if (page) {
2782 struct zone *zone = page_zone(page);
53853e2d 2783
98dd3b48
VB
2784 zone->compact_blockskip_flush = false;
2785 compaction_defer_reset(zone, order, true);
2786 count_vm_event(COMPACTSUCCESS);
2787 return page;
2788 }
56de7263 2789
98dd3b48
VB
2790 /*
2791 * It's bad if compaction run occurs and fails. The most likely reason
2792 * is that pages exist, but not enough to satisfy watermarks.
2793 */
2794 count_vm_event(COMPACTFAIL);
66199712 2795
98dd3b48 2796 cond_resched();
56de7263
MG
2797
2798 return NULL;
2799}
2800#else
2801static inline struct page *
2802__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
a9263751
VB
2803 int alloc_flags, const struct alloc_context *ac,
2804 enum migrate_mode mode, int *contended_compaction,
2805 bool *deferred_compaction)
56de7263
MG
2806{
2807 return NULL;
2808}
2809#endif /* CONFIG_COMPACTION */
2810
bba90710
MS
2811/* Perform direct synchronous page reclaim */
2812static int
a9263751
VB
2813__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2814 const struct alloc_context *ac)
11e33f6a 2815{
11e33f6a 2816 struct reclaim_state reclaim_state;
bba90710 2817 int progress;
11e33f6a
MG
2818
2819 cond_resched();
2820
2821 /* We now go into synchronous reclaim */
2822 cpuset_memory_pressure_bump();
c06b1fca 2823 current->flags |= PF_MEMALLOC;
11e33f6a
MG
2824 lockdep_set_current_reclaim_state(gfp_mask);
2825 reclaim_state.reclaimed_slab = 0;
c06b1fca 2826 current->reclaim_state = &reclaim_state;
11e33f6a 2827
a9263751
VB
2828 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2829 ac->nodemask);
11e33f6a 2830
c06b1fca 2831 current->reclaim_state = NULL;
11e33f6a 2832 lockdep_clear_current_reclaim_state();
c06b1fca 2833 current->flags &= ~PF_MEMALLOC;
11e33f6a
MG
2834
2835 cond_resched();
2836
bba90710
MS
2837 return progress;
2838}
2839
2840/* The really slow allocator path where we enter direct reclaim */
2841static inline struct page *
2842__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
a9263751
VB
2843 int alloc_flags, const struct alloc_context *ac,
2844 unsigned long *did_some_progress)
bba90710
MS
2845{
2846 struct page *page = NULL;
2847 bool drained = false;
2848
a9263751 2849 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
9ee493ce
MG
2850 if (unlikely(!(*did_some_progress)))
2851 return NULL;
11e33f6a 2852
9ee493ce 2853retry:
a9263751
VB
2854 page = get_page_from_freelist(gfp_mask, order,
2855 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
9ee493ce
MG
2856
2857 /*
2858 * If an allocation failed after direct reclaim, it could be because
0aaa29a5
MG
2859 * pages are pinned on the per-cpu lists or in high alloc reserves.
2860 * Shrink them them and try again
9ee493ce
MG
2861 */
2862 if (!page && !drained) {
0aaa29a5 2863 unreserve_highatomic_pageblock(ac);
93481ff0 2864 drain_all_pages(NULL);
9ee493ce
MG
2865 drained = true;
2866 goto retry;
2867 }
2868
11e33f6a
MG
2869 return page;
2870}
2871
a9263751 2872static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3a025760
JW
2873{
2874 struct zoneref *z;
2875 struct zone *zone;
2876
a9263751
VB
2877 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2878 ac->high_zoneidx, ac->nodemask)
2879 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
3a025760
JW
2880}
2881
341ce06f
PZ
2882static inline int
2883gfp_to_alloc_flags(gfp_t gfp_mask)
2884{
341ce06f 2885 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1da177e4 2886
a56f57ff 2887 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
e6223a3b 2888 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
933e312e 2889
341ce06f
PZ
2890 /*
2891 * The caller may dip into page reserves a bit more if the caller
2892 * cannot run direct reclaim, or if the caller has realtime scheduling
2893 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
d0164adc 2894 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
341ce06f 2895 */
e6223a3b 2896 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
1da177e4 2897
d0164adc 2898 if (gfp_mask & __GFP_ATOMIC) {
5c3240d9 2899 /*
b104a35d
DR
2900 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2901 * if it can't schedule.
5c3240d9 2902 */
b104a35d 2903 if (!(gfp_mask & __GFP_NOMEMALLOC))
5c3240d9 2904 alloc_flags |= ALLOC_HARDER;
523b9458 2905 /*
b104a35d 2906 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
344736f2 2907 * comment for __cpuset_node_allowed().
523b9458 2908 */
341ce06f 2909 alloc_flags &= ~ALLOC_CPUSET;
c06b1fca 2910 } else if (unlikely(rt_task(current)) && !in_interrupt())
341ce06f
PZ
2911 alloc_flags |= ALLOC_HARDER;
2912
b37f1dd0
MG
2913 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2914 if (gfp_mask & __GFP_MEMALLOC)
2915 alloc_flags |= ALLOC_NO_WATERMARKS;
907aed48
MG
2916 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2917 alloc_flags |= ALLOC_NO_WATERMARKS;
2918 else if (!in_interrupt() &&
2919 ((current->flags & PF_MEMALLOC) ||
2920 unlikely(test_thread_flag(TIF_MEMDIE))))
341ce06f 2921 alloc_flags |= ALLOC_NO_WATERMARKS;
1da177e4 2922 }
d95ea5d1 2923#ifdef CONFIG_CMA
43e7a34d 2924 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
d95ea5d1
BZ
2925 alloc_flags |= ALLOC_CMA;
2926#endif
341ce06f
PZ
2927 return alloc_flags;
2928}
2929
072bb0aa
MG
2930bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2931{
b37f1dd0 2932 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
072bb0aa
MG
2933}
2934
d0164adc
MG
2935static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
2936{
2937 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
2938}
2939
11e33f6a
MG
2940static inline struct page *
2941__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
a9263751 2942 struct alloc_context *ac)
11e33f6a 2943{
d0164adc 2944 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
11e33f6a
MG
2945 struct page *page = NULL;
2946 int alloc_flags;
2947 unsigned long pages_reclaimed = 0;
2948 unsigned long did_some_progress;
e0b9daeb 2949 enum migrate_mode migration_mode = MIGRATE_ASYNC;
66199712 2950 bool deferred_compaction = false;
1f9efdef 2951 int contended_compaction = COMPACT_CONTENDED_NONE;
1da177e4 2952
72807a74
MG
2953 /*
2954 * In the slowpath, we sanity check order to avoid ever trying to
2955 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2956 * be using allocators in order of preference for an area that is
2957 * too large.
2958 */
1fc28b70
MG
2959 if (order >= MAX_ORDER) {
2960 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
72807a74 2961 return NULL;
1fc28b70 2962 }
1da177e4 2963
d0164adc
MG
2964 /*
2965 * We also sanity check to catch abuse of atomic reserves being used by
2966 * callers that are not in atomic context.
2967 */
2968 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
2969 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
2970 gfp_mask &= ~__GFP_ATOMIC;
2971
952f3b51 2972 /*
4167e9b2
DR
2973 * If this allocation cannot block and it is for a specific node, then
2974 * fail early. There's no need to wakeup kswapd or retry for a
2975 * speculative node-specific allocation.
952f3b51 2976 */
d0164adc 2977 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !can_direct_reclaim)
952f3b51
CL
2978 goto nopage;
2979
9879de73 2980retry:
d0164adc 2981 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
a9263751 2982 wake_all_kswapds(order, ac);
1da177e4 2983
9bf2229f 2984 /*
7fb1d9fc
RS
2985 * OK, we're below the kswapd watermark and have kicked background
2986 * reclaim. Now things get more complex, so set up alloc_flags according
2987 * to how we want to proceed.
9bf2229f 2988 */
341ce06f 2989 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1da177e4 2990
f33261d7
DR
2991 /*
2992 * Find the true preferred zone if the allocation is unconstrained by
2993 * cpusets.
2994 */
a9263751 2995 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
d8846374 2996 struct zoneref *preferred_zoneref;
a9263751
VB
2997 preferred_zoneref = first_zones_zonelist(ac->zonelist,
2998 ac->high_zoneidx, NULL, &ac->preferred_zone);
2999 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
d8846374 3000 }
f33261d7 3001
341ce06f 3002 /* This is the last chance, in general, before the goto nopage. */
a9263751
VB
3003 page = get_page_from_freelist(gfp_mask, order,
3004 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
7fb1d9fc
RS
3005 if (page)
3006 goto got_pg;
1da177e4 3007
11e33f6a 3008 /* Allocate without watermarks if the context allows */
341ce06f 3009 if (alloc_flags & ALLOC_NO_WATERMARKS) {
183f6371
MG
3010 /*
3011 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3012 * the allocation is high priority and these type of
3013 * allocations are system rather than user orientated
3014 */
a9263751 3015 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
33d53103
MH
3016 page = get_page_from_freelist(gfp_mask, order,
3017 ALLOC_NO_WATERMARKS, ac);
3018 if (page)
3019 goto got_pg;
1da177e4
LT
3020 }
3021
d0164adc
MG
3022 /* Caller is not willing to reclaim, we can't balance anything */
3023 if (!can_direct_reclaim) {
aed0a0e3 3024 /*
33d53103
MH
3025 * All existing users of the __GFP_NOFAIL are blockable, so warn
3026 * of any new users that actually allow this type of allocation
3027 * to fail.
aed0a0e3
DR
3028 */
3029 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
1da177e4 3030 goto nopage;
aed0a0e3 3031 }
1da177e4 3032
341ce06f 3033 /* Avoid recursion of direct reclaim */
33d53103
MH
3034 if (current->flags & PF_MEMALLOC) {
3035 /*
3036 * __GFP_NOFAIL request from this context is rather bizarre
3037 * because we cannot reclaim anything and only can loop waiting
3038 * for somebody to do a work for us.
3039 */
3040 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3041 cond_resched();
3042 goto retry;
3043 }
341ce06f 3044 goto nopage;
33d53103 3045 }
341ce06f 3046
6583bb64
DR
3047 /* Avoid allocations with no watermarks from looping endlessly */
3048 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3049 goto nopage;
3050
77f1fe6b
MG
3051 /*
3052 * Try direct compaction. The first pass is asynchronous. Subsequent
3053 * attempts after direct reclaim are synchronous
3054 */
a9263751
VB
3055 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3056 migration_mode,
3057 &contended_compaction,
53853e2d 3058 &deferred_compaction);
56de7263
MG
3059 if (page)
3060 goto got_pg;
75f30861 3061
1f9efdef 3062 /* Checks for THP-specific high-order allocations */
d0164adc 3063 if (is_thp_gfp_mask(gfp_mask)) {
1f9efdef
VB
3064 /*
3065 * If compaction is deferred for high-order allocations, it is
3066 * because sync compaction recently failed. If this is the case
3067 * and the caller requested a THP allocation, we do not want
3068 * to heavily disrupt the system, so we fail the allocation
3069 * instead of entering direct reclaim.
3070 */
3071 if (deferred_compaction)
3072 goto nopage;
3073
3074 /*
3075 * In all zones where compaction was attempted (and not
3076 * deferred or skipped), lock contention has been detected.
3077 * For THP allocation we do not want to disrupt the others
3078 * so we fallback to base pages instead.
3079 */
3080 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3081 goto nopage;
3082
3083 /*
3084 * If compaction was aborted due to need_resched(), we do not
3085 * want to further increase allocation latency, unless it is
3086 * khugepaged trying to collapse.
3087 */
3088 if (contended_compaction == COMPACT_CONTENDED_SCHED
3089 && !(current->flags & PF_KTHREAD))
3090 goto nopage;
3091 }
66199712 3092
8fe78048
DR
3093 /*
3094 * It can become very expensive to allocate transparent hugepages at
3095 * fault, so use asynchronous memory compaction for THP unless it is
3096 * khugepaged trying to collapse.
3097 */
d0164adc 3098 if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
8fe78048
DR
3099 migration_mode = MIGRATE_SYNC_LIGHT;
3100
11e33f6a 3101 /* Try direct reclaim and then allocating */
a9263751
VB
3102 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3103 &did_some_progress);
11e33f6a
MG
3104 if (page)
3105 goto got_pg;
1da177e4 3106
9083905a
JW
3107 /* Do not loop if specifically requested */
3108 if (gfp_mask & __GFP_NORETRY)
3109 goto noretry;
3110
3111 /* Keep reclaiming pages as long as there is reasonable progress */
a41f24ea 3112 pages_reclaimed += did_some_progress;
9083905a
JW
3113 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3114 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
11e33f6a 3115 /* Wait for some write requests to complete then retry */
a9263751 3116 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
9879de73 3117 goto retry;
1da177e4
LT
3118 }
3119
9083905a
JW
3120 /* Reclaim has failed us, start killing things */
3121 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3122 if (page)
3123 goto got_pg;
3124
3125 /* Retry as long as the OOM killer is making progress */
3126 if (did_some_progress)
3127 goto retry;
3128
3129noretry:
3130 /*
3131 * High-order allocations do not necessarily loop after
3132 * direct reclaim and reclaim/compaction depends on compaction
3133 * being called after reclaim so call directly if necessary
3134 */
3135 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3136 ac, migration_mode,
3137 &contended_compaction,
3138 &deferred_compaction);
3139 if (page)
3140 goto got_pg;
1da177e4 3141nopage:
a238ab5b 3142 warn_alloc_failed(gfp_mask, order, NULL);
1da177e4 3143got_pg:
072bb0aa 3144 return page;
1da177e4 3145}
11e33f6a
MG
3146
3147/*
3148 * This is the 'heart' of the zoned buddy allocator.
3149 */
3150struct page *
3151__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3152 struct zonelist *zonelist, nodemask_t *nodemask)
3153{
d8846374 3154 struct zoneref *preferred_zoneref;
cc9a6c87 3155 struct page *page = NULL;
cc9a6c87 3156 unsigned int cpuset_mems_cookie;
3a025760 3157 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
91fbdc0f 3158 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
a9263751
VB
3159 struct alloc_context ac = {
3160 .high_zoneidx = gfp_zone(gfp_mask),
3161 .nodemask = nodemask,
3162 .migratetype = gfpflags_to_migratetype(gfp_mask),
3163 };
11e33f6a 3164
dcce284a
BH
3165 gfp_mask &= gfp_allowed_mask;
3166
11e33f6a
MG
3167 lockdep_trace_alloc(gfp_mask);
3168
d0164adc 3169 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
11e33f6a
MG
3170
3171 if (should_fail_alloc_page(gfp_mask, order))
3172 return NULL;
3173
3174 /*
3175 * Check the zones suitable for the gfp_mask contain at least one
3176 * valid zone. It's possible to have an empty zonelist as a result
4167e9b2 3177 * of __GFP_THISNODE and a memoryless node
11e33f6a
MG
3178 */
3179 if (unlikely(!zonelist->_zonerefs->zone))
3180 return NULL;
3181
a9263751 3182 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
21bb9bd1
VB
3183 alloc_flags |= ALLOC_CMA;
3184
cc9a6c87 3185retry_cpuset:
d26914d1 3186 cpuset_mems_cookie = read_mems_allowed_begin();
cc9a6c87 3187
a9263751
VB
3188 /* We set it here, as __alloc_pages_slowpath might have changed it */
3189 ac.zonelist = zonelist;
c9ab0c4f
MG
3190
3191 /* Dirty zone balancing only done in the fast path */
3192 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3193
5117f45d 3194 /* The preferred zone is used for statistics later */
a9263751
VB
3195 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3196 ac.nodemask ? : &cpuset_current_mems_allowed,
3197 &ac.preferred_zone);
3198 if (!ac.preferred_zone)
cc9a6c87 3199 goto out;
a9263751 3200 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
5117f45d
MG
3201
3202 /* First allocation attempt */
91fbdc0f 3203 alloc_mask = gfp_mask|__GFP_HARDWALL;
a9263751 3204 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
21caf2fc
ML
3205 if (unlikely(!page)) {
3206 /*
3207 * Runtime PM, block IO and its error handling path
3208 * can deadlock because I/O on the device might not
3209 * complete.
3210 */
91fbdc0f 3211 alloc_mask = memalloc_noio_flags(gfp_mask);
c9ab0c4f 3212 ac.spread_dirty_pages = false;
91fbdc0f 3213
a9263751 3214 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
21caf2fc 3215 }
11e33f6a 3216
23f086f9
XQ
3217 if (kmemcheck_enabled && page)
3218 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3219
a9263751 3220 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
cc9a6c87
MG
3221
3222out:
3223 /*
3224 * When updating a task's mems_allowed, it is possible to race with
3225 * parallel threads in such a way that an allocation can fail while
3226 * the mask is being updated. If a page allocation is about to fail,
3227 * check if the cpuset changed during allocation and if so, retry.
3228 */
d26914d1 3229 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
cc9a6c87
MG
3230 goto retry_cpuset;
3231
11e33f6a 3232 return page;
1da177e4 3233}
d239171e 3234EXPORT_SYMBOL(__alloc_pages_nodemask);
1da177e4
LT
3235
3236/*
3237 * Common helper functions.
3238 */
920c7a5d 3239unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1da177e4 3240{
945a1113
AM
3241 struct page *page;
3242
3243 /*
3244 * __get_free_pages() returns a 32-bit address, which cannot represent
3245 * a highmem page
3246 */
3247 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3248
1da177e4
LT
3249 page = alloc_pages(gfp_mask, order);
3250 if (!page)
3251 return 0;
3252 return (unsigned long) page_address(page);
3253}
1da177e4
LT
3254EXPORT_SYMBOL(__get_free_pages);
3255
920c7a5d 3256unsigned long get_zeroed_page(gfp_t gfp_mask)
1da177e4 3257{
945a1113 3258 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
1da177e4 3259}
1da177e4
LT
3260EXPORT_SYMBOL(get_zeroed_page);
3261
920c7a5d 3262void __free_pages(struct page *page, unsigned int order)
1da177e4 3263{
b5810039 3264 if (put_page_testzero(page)) {
1da177e4 3265 if (order == 0)
b745bc85 3266 free_hot_cold_page(page, false);
1da177e4
LT
3267 else
3268 __free_pages_ok(page, order);
3269 }
3270}
3271
3272EXPORT_SYMBOL(__free_pages);
3273
920c7a5d 3274void free_pages(unsigned long addr, unsigned int order)
1da177e4
LT
3275{
3276 if (addr != 0) {
725d704e 3277 VM_BUG_ON(!virt_addr_valid((void *)addr));
1da177e4
LT
3278 __free_pages(virt_to_page((void *)addr), order);
3279 }
3280}
3281
3282EXPORT_SYMBOL(free_pages);
3283
b63ae8ca
AD
3284/*
3285 * Page Fragment:
3286 * An arbitrary-length arbitrary-offset area of memory which resides
3287 * within a 0 or higher order page. Multiple fragments within that page
3288 * are individually refcounted, in the page's reference counter.
3289 *
3290 * The page_frag functions below provide a simple allocation framework for
3291 * page fragments. This is used by the network stack and network device
3292 * drivers to provide a backing region of memory for use as either an
3293 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3294 */
3295static struct page *__page_frag_refill(struct page_frag_cache *nc,
3296 gfp_t gfp_mask)
3297{
3298 struct page *page = NULL;
3299 gfp_t gfp = gfp_mask;
3300
3301#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3302 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3303 __GFP_NOMEMALLOC;
3304 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3305 PAGE_FRAG_CACHE_MAX_ORDER);
3306 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3307#endif
3308 if (unlikely(!page))
3309 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3310
3311 nc->va = page ? page_address(page) : NULL;
3312
3313 return page;
3314}
3315
3316void *__alloc_page_frag(struct page_frag_cache *nc,
3317 unsigned int fragsz, gfp_t gfp_mask)
3318{
3319 unsigned int size = PAGE_SIZE;
3320 struct page *page;
3321 int offset;
3322
3323 if (unlikely(!nc->va)) {
3324refill:
3325 page = __page_frag_refill(nc, gfp_mask);
3326 if (!page)
3327 return NULL;
3328
3329#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3330 /* if size can vary use size else just use PAGE_SIZE */
3331 size = nc->size;
3332#endif
3333 /* Even if we own the page, we do not use atomic_set().
3334 * This would break get_page_unless_zero() users.
3335 */
3336 atomic_add(size - 1, &page->_count);
3337
3338 /* reset page count bias and offset to start of new frag */
2f064f34 3339 nc->pfmemalloc = page_is_pfmemalloc(page);
b63ae8ca
AD
3340 nc->pagecnt_bias = size;
3341 nc->offset = size;
3342 }
3343
3344 offset = nc->offset - fragsz;
3345 if (unlikely(offset < 0)) {
3346 page = virt_to_page(nc->va);
3347
3348 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3349 goto refill;
3350
3351#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3352 /* if size can vary use size else just use PAGE_SIZE */
3353 size = nc->size;
3354#endif
3355 /* OK, page count is 0, we can safely set it */
3356 atomic_set(&page->_count, size);
3357
3358 /* reset page count bias and offset to start of new frag */
3359 nc->pagecnt_bias = size;
3360 offset = size - fragsz;
3361 }
3362
3363 nc->pagecnt_bias--;
3364 nc->offset = offset;
3365
3366 return nc->va + offset;
3367}
3368EXPORT_SYMBOL(__alloc_page_frag);
3369
3370/*
3371 * Frees a page fragment allocated out of either a compound or order 0 page.
3372 */
3373void __free_page_frag(void *addr)
3374{
3375 struct page *page = virt_to_head_page(addr);
3376
3377 if (unlikely(put_page_testzero(page)))
3378 __free_pages_ok(page, compound_order(page));
3379}
3380EXPORT_SYMBOL(__free_page_frag);
3381
6a1a0d3b 3382/*
52383431 3383 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
a9bb7e62
VD
3384 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3385 * equivalent to alloc_pages.
6a1a0d3b 3386 *
52383431
VD
3387 * It should be used when the caller would like to use kmalloc, but since the
3388 * allocation is large, it has to fall back to the page allocator.
3389 */
3390struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3391{
3392 struct page *page;
52383431 3393
52383431 3394 page = alloc_pages(gfp_mask, order);
d05e83a6
VD
3395 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3396 __free_pages(page, order);
3397 page = NULL;
3398 }
52383431
VD
3399 return page;
3400}
3401
3402struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3403{
3404 struct page *page;
52383431 3405
52383431 3406 page = alloc_pages_node(nid, gfp_mask, order);
d05e83a6
VD
3407 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3408 __free_pages(page, order);
3409 page = NULL;
3410 }
52383431
VD
3411 return page;
3412}
3413
3414/*
3415 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3416 * alloc_kmem_pages.
6a1a0d3b 3417 */
52383431 3418void __free_kmem_pages(struct page *page, unsigned int order)
6a1a0d3b 3419{
d05e83a6 3420 memcg_kmem_uncharge(page, order);
6a1a0d3b
GC
3421 __free_pages(page, order);
3422}
3423
52383431 3424void free_kmem_pages(unsigned long addr, unsigned int order)
6a1a0d3b
GC
3425{
3426 if (addr != 0) {
3427 VM_BUG_ON(!virt_addr_valid((void *)addr));
52383431 3428 __free_kmem_pages(virt_to_page((void *)addr), order);
6a1a0d3b
GC
3429 }
3430}
3431
d00181b9
KS
3432static void *make_alloc_exact(unsigned long addr, unsigned int order,
3433 size_t size)
ee85c2e1
AK
3434{
3435 if (addr) {
3436 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3437 unsigned long used = addr + PAGE_ALIGN(size);
3438
3439 split_page(virt_to_page((void *)addr), order);
3440 while (used < alloc_end) {
3441 free_page(used);
3442 used += PAGE_SIZE;
3443 }
3444 }
3445 return (void *)addr;
3446}
3447
2be0ffe2
TT
3448/**
3449 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3450 * @size: the number of bytes to allocate
3451 * @gfp_mask: GFP flags for the allocation
3452 *
3453 * This function is similar to alloc_pages(), except that it allocates the
3454 * minimum number of pages to satisfy the request. alloc_pages() can only
3455 * allocate memory in power-of-two pages.
3456 *
3457 * This function is also limited by MAX_ORDER.
3458 *
3459 * Memory allocated by this function must be released by free_pages_exact().
3460 */
3461void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3462{
3463 unsigned int order = get_order(size);
3464 unsigned long addr;
3465
3466 addr = __get_free_pages(gfp_mask, order);
ee85c2e1 3467 return make_alloc_exact(addr, order, size);
2be0ffe2
TT
3468}
3469EXPORT_SYMBOL(alloc_pages_exact);
3470
ee85c2e1
AK
3471/**
3472 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3473 * pages on a node.
b5e6ab58 3474 * @nid: the preferred node ID where memory should be allocated
ee85c2e1
AK
3475 * @size: the number of bytes to allocate
3476 * @gfp_mask: GFP flags for the allocation
3477 *
3478 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3479 * back.
ee85c2e1 3480 */
e1931811 3481void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
ee85c2e1 3482{
d00181b9 3483 unsigned int order = get_order(size);
ee85c2e1
AK
3484 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3485 if (!p)
3486 return NULL;
3487 return make_alloc_exact((unsigned long)page_address(p), order, size);
3488}
ee85c2e1 3489
2be0ffe2
TT
3490/**
3491 * free_pages_exact - release memory allocated via alloc_pages_exact()
3492 * @virt: the value returned by alloc_pages_exact.
3493 * @size: size of allocation, same value as passed to alloc_pages_exact().
3494 *
3495 * Release the memory allocated by a previous call to alloc_pages_exact.
3496 */
3497void free_pages_exact(void *virt, size_t size)
3498{
3499 unsigned long addr = (unsigned long)virt;
3500 unsigned long end = addr + PAGE_ALIGN(size);
3501
3502 while (addr < end) {
3503 free_page(addr);
3504 addr += PAGE_SIZE;
3505 }
3506}
3507EXPORT_SYMBOL(free_pages_exact);
3508
e0fb5815
ZY
3509/**
3510 * nr_free_zone_pages - count number of pages beyond high watermark
3511 * @offset: The zone index of the highest zone
3512 *
3513 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3514 * high watermark within all zones at or below a given zone index. For each
3515 * zone, the number of pages is calculated as:
834405c3 3516 * managed_pages - high_pages
e0fb5815 3517 */
ebec3862 3518static unsigned long nr_free_zone_pages(int offset)
1da177e4 3519{
dd1a239f 3520 struct zoneref *z;
54a6eb5c
MG
3521 struct zone *zone;
3522
e310fd43 3523 /* Just pick one node, since fallback list is circular */
ebec3862 3524 unsigned long sum = 0;
1da177e4 3525
0e88460d 3526 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1da177e4 3527
54a6eb5c 3528 for_each_zone_zonelist(zone, z, zonelist, offset) {
b40da049 3529 unsigned long size = zone->managed_pages;
41858966 3530 unsigned long high = high_wmark_pages(zone);
e310fd43
MB
3531 if (size > high)
3532 sum += size - high;
1da177e4
LT
3533 }
3534
3535 return sum;
3536}
3537
e0fb5815
ZY
3538/**
3539 * nr_free_buffer_pages - count number of pages beyond high watermark
3540 *
3541 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3542 * watermark within ZONE_DMA and ZONE_NORMAL.
1da177e4 3543 */
ebec3862 3544unsigned long nr_free_buffer_pages(void)
1da177e4 3545{
af4ca457 3546 return nr_free_zone_pages(gfp_zone(GFP_USER));
1da177e4 3547}
c2f1a551 3548EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1da177e4 3549
e0fb5815
ZY
3550/**
3551 * nr_free_pagecache_pages - count number of pages beyond high watermark
3552 *
3553 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3554 * high watermark within all zones.
1da177e4 3555 */
ebec3862 3556unsigned long nr_free_pagecache_pages(void)
1da177e4 3557{
2a1e274a 3558 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1da177e4 3559}
08e0f6a9
CL
3560
3561static inline void show_node(struct zone *zone)
1da177e4 3562{
e5adfffc 3563 if (IS_ENABLED(CONFIG_NUMA))
25ba77c1 3564 printk("Node %d ", zone_to_nid(zone));
1da177e4 3565}
1da177e4 3566
1da177e4
LT
3567void si_meminfo(struct sysinfo *val)
3568{
3569 val->totalram = totalram_pages;
cc7452b6 3570 val->sharedram = global_page_state(NR_SHMEM);
d23ad423 3571 val->freeram = global_page_state(NR_FREE_PAGES);
1da177e4 3572 val->bufferram = nr_blockdev_pages();
1da177e4
LT
3573 val->totalhigh = totalhigh_pages;
3574 val->freehigh = nr_free_highpages();
1da177e4
LT
3575 val->mem_unit = PAGE_SIZE;
3576}
3577
3578EXPORT_SYMBOL(si_meminfo);
3579
3580#ifdef CONFIG_NUMA
3581void si_meminfo_node(struct sysinfo *val, int nid)
3582{
cdd91a77
JL
3583 int zone_type; /* needs to be signed */
3584 unsigned long managed_pages = 0;
1da177e4
LT
3585 pg_data_t *pgdat = NODE_DATA(nid);
3586
cdd91a77
JL
3587 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3588 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3589 val->totalram = managed_pages;
cc7452b6 3590 val->sharedram = node_page_state(nid, NR_SHMEM);
d23ad423 3591 val->freeram = node_page_state(nid, NR_FREE_PAGES);
98d2b0eb 3592#ifdef CONFIG_HIGHMEM
b40da049 3593 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
d23ad423
CL
3594 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3595 NR_FREE_PAGES);
98d2b0eb
CL
3596#else
3597 val->totalhigh = 0;
3598 val->freehigh = 0;
3599#endif
1da177e4
LT
3600 val->mem_unit = PAGE_SIZE;
3601}
3602#endif
3603
ddd588b5 3604/*
7bf02ea2
DR
3605 * Determine whether the node should be displayed or not, depending on whether
3606 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
ddd588b5 3607 */
7bf02ea2 3608bool skip_free_areas_node(unsigned int flags, int nid)
ddd588b5
DR
3609{
3610 bool ret = false;
cc9a6c87 3611 unsigned int cpuset_mems_cookie;
ddd588b5
DR
3612
3613 if (!(flags & SHOW_MEM_FILTER_NODES))
3614 goto out;
3615
cc9a6c87 3616 do {
d26914d1 3617 cpuset_mems_cookie = read_mems_allowed_begin();
cc9a6c87 3618 ret = !node_isset(nid, cpuset_current_mems_allowed);
d26914d1 3619 } while (read_mems_allowed_retry(cpuset_mems_cookie));
ddd588b5
DR
3620out:
3621 return ret;
3622}
3623
1da177e4
LT
3624#define K(x) ((x) << (PAGE_SHIFT-10))
3625
377e4f16
RV
3626static void show_migration_types(unsigned char type)
3627{
3628 static const char types[MIGRATE_TYPES] = {
3629 [MIGRATE_UNMOVABLE] = 'U',
377e4f16 3630 [MIGRATE_MOVABLE] = 'M',
475a2f90
VB
3631 [MIGRATE_RECLAIMABLE] = 'E',
3632 [MIGRATE_HIGHATOMIC] = 'H',
377e4f16
RV
3633#ifdef CONFIG_CMA
3634 [MIGRATE_CMA] = 'C',
3635#endif
194159fb 3636#ifdef CONFIG_MEMORY_ISOLATION
377e4f16 3637 [MIGRATE_ISOLATE] = 'I',
194159fb 3638#endif
377e4f16
RV
3639 };
3640 char tmp[MIGRATE_TYPES + 1];
3641 char *p = tmp;
3642 int i;
3643
3644 for (i = 0; i < MIGRATE_TYPES; i++) {
3645 if (type & (1 << i))
3646 *p++ = types[i];
3647 }
3648
3649 *p = '\0';
3650 printk("(%s) ", tmp);
3651}
3652
1da177e4
LT
3653/*
3654 * Show free area list (used inside shift_scroll-lock stuff)
3655 * We also calculate the percentage fragmentation. We do this by counting the
3656 * memory on each free list with the exception of the first item on the list.
d1bfcdb8
KK
3657 *
3658 * Bits in @filter:
3659 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3660 * cpuset.
1da177e4 3661 */
7bf02ea2 3662void show_free_areas(unsigned int filter)
1da177e4 3663{
d1bfcdb8 3664 unsigned long free_pcp = 0;
c7241913 3665 int cpu;
1da177e4
LT
3666 struct zone *zone;
3667
ee99c71c 3668 for_each_populated_zone(zone) {
7bf02ea2 3669 if (skip_free_areas_node(filter, zone_to_nid(zone)))
ddd588b5 3670 continue;
d1bfcdb8 3671
761b0677
KK
3672 for_each_online_cpu(cpu)
3673 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
1da177e4
LT
3674 }
3675
a731286d
KM
3676 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3677 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
d1bfcdb8
KK
3678 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3679 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
d1ce749a 3680 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
d1bfcdb8 3681 " free:%lu free_pcp:%lu free_cma:%lu\n",
4f98a2fe 3682 global_page_state(NR_ACTIVE_ANON),
4f98a2fe 3683 global_page_state(NR_INACTIVE_ANON),
a731286d
KM
3684 global_page_state(NR_ISOLATED_ANON),
3685 global_page_state(NR_ACTIVE_FILE),
4f98a2fe 3686 global_page_state(NR_INACTIVE_FILE),
a731286d 3687 global_page_state(NR_ISOLATED_FILE),
7b854121 3688 global_page_state(NR_UNEVICTABLE),
b1e7a8fd 3689 global_page_state(NR_FILE_DIRTY),
ce866b34 3690 global_page_state(NR_WRITEBACK),
fd39fc85 3691 global_page_state(NR_UNSTABLE_NFS),
3701b033
KM
3692 global_page_state(NR_SLAB_RECLAIMABLE),
3693 global_page_state(NR_SLAB_UNRECLAIMABLE),
65ba55f5 3694 global_page_state(NR_FILE_MAPPED),
4b02108a 3695 global_page_state(NR_SHMEM),
a25700a5 3696 global_page_state(NR_PAGETABLE),
d1ce749a 3697 global_page_state(NR_BOUNCE),
d1bfcdb8
KK
3698 global_page_state(NR_FREE_PAGES),
3699 free_pcp,
d1ce749a 3700 global_page_state(NR_FREE_CMA_PAGES));
1da177e4 3701
ee99c71c 3702 for_each_populated_zone(zone) {
1da177e4
LT
3703 int i;
3704
7bf02ea2 3705 if (skip_free_areas_node(filter, zone_to_nid(zone)))
ddd588b5 3706 continue;
d1bfcdb8
KK
3707
3708 free_pcp = 0;
3709 for_each_online_cpu(cpu)
3710 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3711
1da177e4
LT
3712 show_node(zone);
3713 printk("%s"
3714 " free:%lukB"
3715 " min:%lukB"
3716 " low:%lukB"
3717 " high:%lukB"
4f98a2fe
RR
3718 " active_anon:%lukB"
3719 " inactive_anon:%lukB"
3720 " active_file:%lukB"
3721 " inactive_file:%lukB"
7b854121 3722 " unevictable:%lukB"
a731286d
KM
3723 " isolated(anon):%lukB"
3724 " isolated(file):%lukB"
1da177e4 3725 " present:%lukB"
9feedc9d 3726 " managed:%lukB"
4a0aa73f
KM
3727 " mlocked:%lukB"
3728 " dirty:%lukB"
3729 " writeback:%lukB"
3730 " mapped:%lukB"
4b02108a 3731 " shmem:%lukB"
4a0aa73f
KM
3732 " slab_reclaimable:%lukB"
3733 " slab_unreclaimable:%lukB"
c6a7f572 3734 " kernel_stack:%lukB"
4a0aa73f
KM
3735 " pagetables:%lukB"
3736 " unstable:%lukB"
3737 " bounce:%lukB"
d1bfcdb8
KK
3738 " free_pcp:%lukB"
3739 " local_pcp:%ukB"
d1ce749a 3740 " free_cma:%lukB"
4a0aa73f 3741 " writeback_tmp:%lukB"
1da177e4
LT
3742 " pages_scanned:%lu"
3743 " all_unreclaimable? %s"
3744 "\n",
3745 zone->name,
88f5acf8 3746 K(zone_page_state(zone, NR_FREE_PAGES)),
41858966
MG
3747 K(min_wmark_pages(zone)),
3748 K(low_wmark_pages(zone)),
3749 K(high_wmark_pages(zone)),
4f98a2fe
RR
3750 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3751 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3752 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3753 K(zone_page_state(zone, NR_INACTIVE_FILE)),
7b854121 3754 K(zone_page_state(zone, NR_UNEVICTABLE)),
a731286d
KM
3755 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3756 K(zone_page_state(zone, NR_ISOLATED_FILE)),
1da177e4 3757 K(zone->present_pages),
9feedc9d 3758 K(zone->managed_pages),
4a0aa73f
KM
3759 K(zone_page_state(zone, NR_MLOCK)),
3760 K(zone_page_state(zone, NR_FILE_DIRTY)),
3761 K(zone_page_state(zone, NR_WRITEBACK)),
3762 K(zone_page_state(zone, NR_FILE_MAPPED)),
4b02108a 3763 K(zone_page_state(zone, NR_SHMEM)),
4a0aa73f
KM
3764 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3765 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
c6a7f572
KM
3766 zone_page_state(zone, NR_KERNEL_STACK) *
3767 THREAD_SIZE / 1024,
4a0aa73f
KM
3768 K(zone_page_state(zone, NR_PAGETABLE)),
3769 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3770 K(zone_page_state(zone, NR_BOUNCE)),
d1bfcdb8
KK
3771 K(free_pcp),
3772 K(this_cpu_read(zone->pageset->pcp.count)),
d1ce749a 3773 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
4a0aa73f 3774 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
0d5d823a 3775 K(zone_page_state(zone, NR_PAGES_SCANNED)),
6e543d57 3776 (!zone_reclaimable(zone) ? "yes" : "no")
1da177e4
LT
3777 );
3778 printk("lowmem_reserve[]:");
3779 for (i = 0; i < MAX_NR_ZONES; i++)
3484b2de 3780 printk(" %ld", zone->lowmem_reserve[i]);
1da177e4
LT
3781 printk("\n");
3782 }
3783
ee99c71c 3784 for_each_populated_zone(zone) {
d00181b9
KS
3785 unsigned int order;
3786 unsigned long nr[MAX_ORDER], flags, total = 0;
377e4f16 3787 unsigned char types[MAX_ORDER];
1da177e4 3788
7bf02ea2 3789 if (skip_free_areas_node(filter, zone_to_nid(zone)))
ddd588b5 3790 continue;
1da177e4
LT
3791 show_node(zone);
3792 printk("%s: ", zone->name);
1da177e4
LT
3793
3794 spin_lock_irqsave(&zone->lock, flags);
3795 for (order = 0; order < MAX_ORDER; order++) {
377e4f16
RV
3796 struct free_area *area = &zone->free_area[order];
3797 int type;
3798
3799 nr[order] = area->nr_free;
8f9de51a 3800 total += nr[order] << order;
377e4f16
RV
3801
3802 types[order] = 0;
3803 for (type = 0; type < MIGRATE_TYPES; type++) {
3804 if (!list_empty(&area->free_list[type]))
3805 types[order] |= 1 << type;
3806 }
1da177e4
LT
3807 }
3808 spin_unlock_irqrestore(&zone->lock, flags);
377e4f16 3809 for (order = 0; order < MAX_ORDER; order++) {
8f9de51a 3810 printk("%lu*%lukB ", nr[order], K(1UL) << order);
377e4f16
RV
3811 if (nr[order])
3812 show_migration_types(types[order]);
3813 }
1da177e4
LT
3814 printk("= %lukB\n", K(total));
3815 }
3816
949f7ec5
DR
3817 hugetlb_show_meminfo();
3818
e6f3602d
LW
3819 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3820
1da177e4
LT
3821 show_swap_cache_info();
3822}
3823
19770b32
MG
3824static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3825{
3826 zoneref->zone = zone;
3827 zoneref->zone_idx = zone_idx(zone);
3828}
3829
1da177e4
LT
3830/*
3831 * Builds allocation fallback zone lists.
1a93205b
CL
3832 *
3833 * Add all populated zones of a node to the zonelist.
1da177e4 3834 */
f0c0b2b8 3835static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
bc732f1d 3836 int nr_zones)
1da177e4 3837{
1a93205b 3838 struct zone *zone;
bc732f1d 3839 enum zone_type zone_type = MAX_NR_ZONES;
02a68a5e
CL
3840
3841 do {
2f6726e5 3842 zone_type--;
070f8032 3843 zone = pgdat->node_zones + zone_type;
1a93205b 3844 if (populated_zone(zone)) {
dd1a239f
MG
3845 zoneref_set_zone(zone,
3846 &zonelist->_zonerefs[nr_zones++]);
070f8032 3847 check_highest_zone(zone_type);
1da177e4 3848 }
2f6726e5 3849 } while (zone_type);
bc732f1d 3850
070f8032 3851 return nr_zones;
1da177e4
LT
3852}
3853
f0c0b2b8
KH
3854
3855/*
3856 * zonelist_order:
3857 * 0 = automatic detection of better ordering.
3858 * 1 = order by ([node] distance, -zonetype)
3859 * 2 = order by (-zonetype, [node] distance)
3860 *
3861 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3862 * the same zonelist. So only NUMA can configure this param.
3863 */
3864#define ZONELIST_ORDER_DEFAULT 0
3865#define ZONELIST_ORDER_NODE 1
3866#define ZONELIST_ORDER_ZONE 2
3867
3868/* zonelist order in the kernel.
3869 * set_zonelist_order() will set this to NODE or ZONE.
3870 */
3871static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3872static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3873
3874
1da177e4 3875#ifdef CONFIG_NUMA
f0c0b2b8
KH
3876/* The value user specified ....changed by config */
3877static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3878/* string for sysctl */
3879#define NUMA_ZONELIST_ORDER_LEN 16
3880char numa_zonelist_order[16] = "default";
3881
3882/*
3883 * interface for configure zonelist ordering.
3884 * command line option "numa_zonelist_order"
3885 * = "[dD]efault - default, automatic configuration.
3886 * = "[nN]ode - order by node locality, then by zone within node
3887 * = "[zZ]one - order by zone, then by locality within zone
3888 */
3889
3890static int __parse_numa_zonelist_order(char *s)
3891{
3892 if (*s == 'd' || *s == 'D') {
3893 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3894 } else if (*s == 'n' || *s == 'N') {
3895 user_zonelist_order = ZONELIST_ORDER_NODE;
3896 } else if (*s == 'z' || *s == 'Z') {
3897 user_zonelist_order = ZONELIST_ORDER_ZONE;
3898 } else {
3899 printk(KERN_WARNING
3900 "Ignoring invalid numa_zonelist_order value: "
3901 "%s\n", s);
3902 return -EINVAL;
3903 }
3904 return 0;
3905}
3906
3907static __init int setup_numa_zonelist_order(char *s)
3908{
ecb256f8
VL
3909 int ret;
3910
3911 if (!s)
3912 return 0;
3913
3914 ret = __parse_numa_zonelist_order(s);
3915 if (ret == 0)
3916 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3917
3918 return ret;
f0c0b2b8
KH
3919}
3920early_param("numa_zonelist_order", setup_numa_zonelist_order);
3921
3922/*
3923 * sysctl handler for numa_zonelist_order
3924 */
cccad5b9 3925int numa_zonelist_order_handler(struct ctl_table *table, int write,
8d65af78 3926 void __user *buffer, size_t *length,
f0c0b2b8
KH
3927 loff_t *ppos)
3928{
3929 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3930 int ret;
443c6f14 3931 static DEFINE_MUTEX(zl_order_mutex);
f0c0b2b8 3932
443c6f14 3933 mutex_lock(&zl_order_mutex);
dacbde09
CG
3934 if (write) {
3935 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3936 ret = -EINVAL;
3937 goto out;
3938 }
3939 strcpy(saved_string, (char *)table->data);
3940 }
8d65af78 3941 ret = proc_dostring(table, write, buffer, length, ppos);
f0c0b2b8 3942 if (ret)
443c6f14 3943 goto out;
f0c0b2b8
KH
3944 if (write) {
3945 int oldval = user_zonelist_order;
dacbde09
CG
3946
3947 ret = __parse_numa_zonelist_order((char *)table->data);
3948 if (ret) {
f0c0b2b8
KH
3949 /*
3950 * bogus value. restore saved string
3951 */
dacbde09 3952 strncpy((char *)table->data, saved_string,
f0c0b2b8
KH
3953 NUMA_ZONELIST_ORDER_LEN);
3954 user_zonelist_order = oldval;
4eaf3f64
HL
3955 } else if (oldval != user_zonelist_order) {
3956 mutex_lock(&zonelists_mutex);
9adb62a5 3957 build_all_zonelists(NULL, NULL);
4eaf3f64
HL
3958 mutex_unlock(&zonelists_mutex);
3959 }
f0c0b2b8 3960 }
443c6f14
AK
3961out:
3962 mutex_unlock(&zl_order_mutex);
3963 return ret;
f0c0b2b8
KH
3964}
3965
3966
62bc62a8 3967#define MAX_NODE_LOAD (nr_online_nodes)
f0c0b2b8
KH
3968static int node_load[MAX_NUMNODES];
3969
1da177e4 3970/**
4dc3b16b 3971 * find_next_best_node - find the next node that should appear in a given node's fallback list
1da177e4
LT
3972 * @node: node whose fallback list we're appending
3973 * @used_node_mask: nodemask_t of already used nodes
3974 *
3975 * We use a number of factors to determine which is the next node that should
3976 * appear on a given node's fallback list. The node should not have appeared
3977 * already in @node's fallback list, and it should be the next closest node
3978 * according to the distance array (which contains arbitrary distance values
3979 * from each node to each node in the system), and should also prefer nodes
3980 * with no CPUs, since presumably they'll have very little allocation pressure
3981 * on them otherwise.
3982 * It returns -1 if no node is found.
3983 */
f0c0b2b8 3984static int find_next_best_node(int node, nodemask_t *used_node_mask)
1da177e4 3985{
4cf808eb 3986 int n, val;
1da177e4 3987 int min_val = INT_MAX;
00ef2d2f 3988 int best_node = NUMA_NO_NODE;
a70f7302 3989 const struct cpumask *tmp = cpumask_of_node(0);
1da177e4 3990
4cf808eb
LT
3991 /* Use the local node if we haven't already */
3992 if (!node_isset(node, *used_node_mask)) {
3993 node_set(node, *used_node_mask);
3994 return node;
3995 }
1da177e4 3996
4b0ef1fe 3997 for_each_node_state(n, N_MEMORY) {
1da177e4
LT
3998
3999 /* Don't want a node to appear more than once */
4000 if (node_isset(n, *used_node_mask))
4001 continue;
4002
1da177e4
LT
4003 /* Use the distance array to find the distance */
4004 val = node_distance(node, n);
4005
4cf808eb
LT
4006 /* Penalize nodes under us ("prefer the next node") */
4007 val += (n < node);
4008
1da177e4 4009 /* Give preference to headless and unused nodes */
a70f7302
RR
4010 tmp = cpumask_of_node(n);
4011 if (!cpumask_empty(tmp))
1da177e4
LT
4012 val += PENALTY_FOR_NODE_WITH_CPUS;
4013
4014 /* Slight preference for less loaded node */
4015 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4016 val += node_load[n];
4017
4018 if (val < min_val) {
4019 min_val = val;
4020 best_node = n;
4021 }
4022 }
4023
4024 if (best_node >= 0)
4025 node_set(best_node, *used_node_mask);
4026
4027 return best_node;
4028}
4029
f0c0b2b8
KH
4030
4031/*
4032 * Build zonelists ordered by node and zones within node.
4033 * This results in maximum locality--normal zone overflows into local
4034 * DMA zone, if any--but risks exhausting DMA zone.
4035 */
4036static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
1da177e4 4037{
f0c0b2b8 4038 int j;
1da177e4 4039 struct zonelist *zonelist;
f0c0b2b8 4040
54a6eb5c 4041 zonelist = &pgdat->node_zonelists[0];
dd1a239f 4042 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
54a6eb5c 4043 ;
bc732f1d 4044 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
dd1a239f
MG
4045 zonelist->_zonerefs[j].zone = NULL;
4046 zonelist->_zonerefs[j].zone_idx = 0;
f0c0b2b8
KH
4047}
4048
523b9458
CL
4049/*
4050 * Build gfp_thisnode zonelists
4051 */
4052static void build_thisnode_zonelists(pg_data_t *pgdat)
4053{
523b9458
CL
4054 int j;
4055 struct zonelist *zonelist;
4056
54a6eb5c 4057 zonelist = &pgdat->node_zonelists[1];
bc732f1d 4058 j = build_zonelists_node(pgdat, zonelist, 0);
dd1a239f
MG
4059 zonelist->_zonerefs[j].zone = NULL;
4060 zonelist->_zonerefs[j].zone_idx = 0;
523b9458
CL
4061}
4062
f0c0b2b8
KH
4063/*
4064 * Build zonelists ordered by zone and nodes within zones.
4065 * This results in conserving DMA zone[s] until all Normal memory is
4066 * exhausted, but results in overflowing to remote node while memory
4067 * may still exist in local DMA zone.
4068 */
4069static int node_order[MAX_NUMNODES];
4070
4071static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4072{
f0c0b2b8
KH
4073 int pos, j, node;
4074 int zone_type; /* needs to be signed */
4075 struct zone *z;
4076 struct zonelist *zonelist;
4077
54a6eb5c
MG
4078 zonelist = &pgdat->node_zonelists[0];
4079 pos = 0;
4080 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4081 for (j = 0; j < nr_nodes; j++) {
4082 node = node_order[j];
4083 z = &NODE_DATA(node)->node_zones[zone_type];
4084 if (populated_zone(z)) {
dd1a239f
MG
4085 zoneref_set_zone(z,
4086 &zonelist->_zonerefs[pos++]);
54a6eb5c 4087 check_highest_zone(zone_type);
f0c0b2b8
KH
4088 }
4089 }
f0c0b2b8 4090 }
dd1a239f
MG
4091 zonelist->_zonerefs[pos].zone = NULL;
4092 zonelist->_zonerefs[pos].zone_idx = 0;
f0c0b2b8
KH
4093}
4094
3193913c
MG
4095#if defined(CONFIG_64BIT)
4096/*
4097 * Devices that require DMA32/DMA are relatively rare and do not justify a
4098 * penalty to every machine in case the specialised case applies. Default
4099 * to Node-ordering on 64-bit NUMA machines
4100 */
4101static int default_zonelist_order(void)
4102{
4103 return ZONELIST_ORDER_NODE;
4104}
4105#else
4106/*
4107 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4108 * by the kernel. If processes running on node 0 deplete the low memory zone
4109 * then reclaim will occur more frequency increasing stalls and potentially
4110 * be easier to OOM if a large percentage of the zone is under writeback or
4111 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4112 * Hence, default to zone ordering on 32-bit.
4113 */
f0c0b2b8
KH
4114static int default_zonelist_order(void)
4115{
f0c0b2b8
KH
4116 return ZONELIST_ORDER_ZONE;
4117}
3193913c 4118#endif /* CONFIG_64BIT */
f0c0b2b8
KH
4119
4120static void set_zonelist_order(void)
4121{
4122 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4123 current_zonelist_order = default_zonelist_order();
4124 else
4125 current_zonelist_order = user_zonelist_order;
4126}
4127
4128static void build_zonelists(pg_data_t *pgdat)
4129{
c00eb15a 4130 int i, node, load;
1da177e4 4131 nodemask_t used_mask;
f0c0b2b8
KH
4132 int local_node, prev_node;
4133 struct zonelist *zonelist;
d00181b9 4134 unsigned int order = current_zonelist_order;
1da177e4
LT
4135
4136 /* initialize zonelists */
523b9458 4137 for (i = 0; i < MAX_ZONELISTS; i++) {
1da177e4 4138 zonelist = pgdat->node_zonelists + i;
dd1a239f
MG
4139 zonelist->_zonerefs[0].zone = NULL;
4140 zonelist->_zonerefs[0].zone_idx = 0;
1da177e4
LT
4141 }
4142
4143 /* NUMA-aware ordering of nodes */
4144 local_node = pgdat->node_id;
62bc62a8 4145 load = nr_online_nodes;
1da177e4
LT
4146 prev_node = local_node;
4147 nodes_clear(used_mask);
f0c0b2b8 4148
f0c0b2b8 4149 memset(node_order, 0, sizeof(node_order));
c00eb15a 4150 i = 0;
f0c0b2b8 4151
1da177e4
LT
4152 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4153 /*
4154 * We don't want to pressure a particular node.
4155 * So adding penalty to the first node in same
4156 * distance group to make it round-robin.
4157 */
957f822a
DR
4158 if (node_distance(local_node, node) !=
4159 node_distance(local_node, prev_node))
f0c0b2b8
KH
4160 node_load[node] = load;
4161
1da177e4
LT
4162 prev_node = node;
4163 load--;
f0c0b2b8
KH
4164 if (order == ZONELIST_ORDER_NODE)
4165 build_zonelists_in_node_order(pgdat, node);
4166 else
c00eb15a 4167 node_order[i++] = node; /* remember order */
f0c0b2b8 4168 }
1da177e4 4169
f0c0b2b8
KH
4170 if (order == ZONELIST_ORDER_ZONE) {
4171 /* calculate node order -- i.e., DMA last! */
c00eb15a 4172 build_zonelists_in_zone_order(pgdat, i);
1da177e4 4173 }
523b9458
CL
4174
4175 build_thisnode_zonelists(pgdat);
1da177e4
LT
4176}
4177
7aac7898
LS
4178#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4179/*
4180 * Return node id of node used for "local" allocations.
4181 * I.e., first node id of first zone in arg node's generic zonelist.
4182 * Used for initializing percpu 'numa_mem', which is used primarily
4183 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4184 */
4185int local_memory_node(int node)
4186{
4187 struct zone *zone;
4188
4189 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4190 gfp_zone(GFP_KERNEL),
4191 NULL,
4192 &zone);
4193 return zone->node;
4194}
4195#endif
f0c0b2b8 4196
1da177e4
LT
4197#else /* CONFIG_NUMA */
4198
f0c0b2b8
KH
4199static void set_zonelist_order(void)
4200{
4201 current_zonelist_order = ZONELIST_ORDER_ZONE;
4202}
4203
4204static void build_zonelists(pg_data_t *pgdat)
1da177e4 4205{
19655d34 4206 int node, local_node;
54a6eb5c
MG
4207 enum zone_type j;
4208 struct zonelist *zonelist;
1da177e4
LT
4209
4210 local_node = pgdat->node_id;
1da177e4 4211
54a6eb5c 4212 zonelist = &pgdat->node_zonelists[0];
bc732f1d 4213 j = build_zonelists_node(pgdat, zonelist, 0);
1da177e4 4214
54a6eb5c
MG
4215 /*
4216 * Now we build the zonelist so that it contains the zones
4217 * of all the other nodes.
4218 * We don't want to pressure a particular node, so when
4219 * building the zones for node N, we make sure that the
4220 * zones coming right after the local ones are those from
4221 * node N+1 (modulo N)
4222 */
4223 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4224 if (!node_online(node))
4225 continue;
bc732f1d 4226 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
1da177e4 4227 }
54a6eb5c
MG
4228 for (node = 0; node < local_node; node++) {
4229 if (!node_online(node))
4230 continue;
bc732f1d 4231 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
54a6eb5c
MG
4232 }
4233
dd1a239f
MG
4234 zonelist->_zonerefs[j].zone = NULL;
4235 zonelist->_zonerefs[j].zone_idx = 0;
1da177e4
LT
4236}
4237
4238#endif /* CONFIG_NUMA */
4239
99dcc3e5
CL
4240/*
4241 * Boot pageset table. One per cpu which is going to be used for all
4242 * zones and all nodes. The parameters will be set in such a way
4243 * that an item put on a list will immediately be handed over to
4244 * the buddy list. This is safe since pageset manipulation is done
4245 * with interrupts disabled.
4246 *
4247 * The boot_pagesets must be kept even after bootup is complete for
4248 * unused processors and/or zones. They do play a role for bootstrapping
4249 * hotplugged processors.
4250 *
4251 * zoneinfo_show() and maybe other functions do
4252 * not check if the processor is online before following the pageset pointer.
4253 * Other parts of the kernel may not check if the zone is available.
4254 */
4255static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4256static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
1f522509 4257static void setup_zone_pageset(struct zone *zone);
99dcc3e5 4258
4eaf3f64
HL
4259/*
4260 * Global mutex to protect against size modification of zonelists
4261 * as well as to serialize pageset setup for the new populated zone.
4262 */
4263DEFINE_MUTEX(zonelists_mutex);
4264
9b1a4d38 4265/* return values int ....just for stop_machine() */
4ed7e022 4266static int __build_all_zonelists(void *data)
1da177e4 4267{
6811378e 4268 int nid;
99dcc3e5 4269 int cpu;
9adb62a5 4270 pg_data_t *self = data;
9276b1bc 4271
7f9cfb31
BL
4272#ifdef CONFIG_NUMA
4273 memset(node_load, 0, sizeof(node_load));
4274#endif
9adb62a5
JL
4275
4276 if (self && !node_online(self->node_id)) {
4277 build_zonelists(self);
9adb62a5
JL
4278 }
4279
9276b1bc 4280 for_each_online_node(nid) {
7ea1530a
CL
4281 pg_data_t *pgdat = NODE_DATA(nid);
4282
4283 build_zonelists(pgdat);
9276b1bc 4284 }
99dcc3e5
CL
4285
4286 /*
4287 * Initialize the boot_pagesets that are going to be used
4288 * for bootstrapping processors. The real pagesets for
4289 * each zone will be allocated later when the per cpu
4290 * allocator is available.
4291 *
4292 * boot_pagesets are used also for bootstrapping offline
4293 * cpus if the system is already booted because the pagesets
4294 * are needed to initialize allocators on a specific cpu too.
4295 * F.e. the percpu allocator needs the page allocator which
4296 * needs the percpu allocator in order to allocate its pagesets
4297 * (a chicken-egg dilemma).
4298 */
7aac7898 4299 for_each_possible_cpu(cpu) {
99dcc3e5
CL
4300 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4301
7aac7898
LS
4302#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4303 /*
4304 * We now know the "local memory node" for each node--
4305 * i.e., the node of the first zone in the generic zonelist.
4306 * Set up numa_mem percpu variable for on-line cpus. During
4307 * boot, only the boot cpu should be on-line; we'll init the
4308 * secondary cpus' numa_mem as they come on-line. During
4309 * node/memory hotplug, we'll fixup all on-line cpus.
4310 */
4311 if (cpu_online(cpu))
4312 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4313#endif
4314 }
4315
6811378e
YG
4316 return 0;
4317}
4318
061f67bc
RV
4319static noinline void __init
4320build_all_zonelists_init(void)
4321{
4322 __build_all_zonelists(NULL);
4323 mminit_verify_zonelist();
4324 cpuset_init_current_mems_allowed();
4325}
4326
4eaf3f64
HL
4327/*
4328 * Called with zonelists_mutex held always
4329 * unless system_state == SYSTEM_BOOTING.
061f67bc
RV
4330 *
4331 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4332 * [we're only called with non-NULL zone through __meminit paths] and
4333 * (2) call of __init annotated helper build_all_zonelists_init
4334 * [protected by SYSTEM_BOOTING].
4eaf3f64 4335 */
9adb62a5 4336void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
6811378e 4337{
f0c0b2b8
KH
4338 set_zonelist_order();
4339
6811378e 4340 if (system_state == SYSTEM_BOOTING) {
061f67bc 4341 build_all_zonelists_init();
6811378e 4342 } else {
e9959f0f 4343#ifdef CONFIG_MEMORY_HOTPLUG
9adb62a5
JL
4344 if (zone)
4345 setup_zone_pageset(zone);
e9959f0f 4346#endif
dd1895e2
CS
4347 /* we have to stop all cpus to guarantee there is no user
4348 of zonelist */
9adb62a5 4349 stop_machine(__build_all_zonelists, pgdat, NULL);
6811378e
YG
4350 /* cpuset refresh routine should be here */
4351 }
bd1e22b8 4352 vm_total_pages = nr_free_pagecache_pages();
9ef9acb0
MG
4353 /*
4354 * Disable grouping by mobility if the number of pages in the
4355 * system is too low to allow the mechanism to work. It would be
4356 * more accurate, but expensive to check per-zone. This check is
4357 * made on memory-hotadd so a system can start with mobility
4358 * disabled and enable it later
4359 */
d9c23400 4360 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
9ef9acb0
MG
4361 page_group_by_mobility_disabled = 1;
4362 else
4363 page_group_by_mobility_disabled = 0;
4364
f88dfff5 4365 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
9ef9acb0 4366 "Total pages: %ld\n",
62bc62a8 4367 nr_online_nodes,
f0c0b2b8 4368 zonelist_order_name[current_zonelist_order],
9ef9acb0 4369 page_group_by_mobility_disabled ? "off" : "on",
f0c0b2b8
KH
4370 vm_total_pages);
4371#ifdef CONFIG_NUMA
f88dfff5 4372 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
f0c0b2b8 4373#endif
1da177e4
LT
4374}
4375
4376/*
4377 * Helper functions to size the waitqueue hash table.
4378 * Essentially these want to choose hash table sizes sufficiently
4379 * large so that collisions trying to wait on pages are rare.
4380 * But in fact, the number of active page waitqueues on typical
4381 * systems is ridiculously low, less than 200. So this is even
4382 * conservative, even though it seems large.
4383 *
4384 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4385 * waitqueues, i.e. the size of the waitq table given the number of pages.
4386 */
4387#define PAGES_PER_WAITQUEUE 256
4388
cca448fe 4389#ifndef CONFIG_MEMORY_HOTPLUG
02b694de 4390static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1da177e4
LT
4391{
4392 unsigned long size = 1;
4393
4394 pages /= PAGES_PER_WAITQUEUE;
4395
4396 while (size < pages)
4397 size <<= 1;
4398
4399 /*
4400 * Once we have dozens or even hundreds of threads sleeping
4401 * on IO we've got bigger problems than wait queue collision.
4402 * Limit the size of the wait table to a reasonable size.
4403 */
4404 size = min(size, 4096UL);
4405
4406 return max(size, 4UL);
4407}
cca448fe
YG
4408#else
4409/*
4410 * A zone's size might be changed by hot-add, so it is not possible to determine
4411 * a suitable size for its wait_table. So we use the maximum size now.
4412 *
4413 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4414 *
4415 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4416 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4417 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4418 *
4419 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4420 * or more by the traditional way. (See above). It equals:
4421 *
4422 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4423 * ia64(16K page size) : = ( 8G + 4M)byte.
4424 * powerpc (64K page size) : = (32G +16M)byte.
4425 */
4426static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4427{
4428 return 4096UL;
4429}
4430#endif
1da177e4
LT
4431
4432/*
4433 * This is an integer logarithm so that shifts can be used later
4434 * to extract the more random high bits from the multiplicative
4435 * hash function before the remainder is taken.
4436 */
4437static inline unsigned long wait_table_bits(unsigned long size)
4438{
4439 return ffz(~size);
4440}
4441
1da177e4
LT
4442/*
4443 * Initially all pages are reserved - free ones are freed
4444 * up by free_all_bootmem() once the early boot process is
4445 * done. Non-atomic initialization, single-pass.
4446 */
c09b4240 4447void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
a2f3aa02 4448 unsigned long start_pfn, enum memmap_context context)
1da177e4 4449{
3a80a7fa 4450 pg_data_t *pgdat = NODE_DATA(nid);
29751f69
AW
4451 unsigned long end_pfn = start_pfn + size;
4452 unsigned long pfn;
86051ca5 4453 struct zone *z;
3a80a7fa 4454 unsigned long nr_initialised = 0;
1da177e4 4455
22b31eec
HD
4456 if (highest_memmap_pfn < end_pfn - 1)
4457 highest_memmap_pfn = end_pfn - 1;
4458
3a80a7fa 4459 z = &pgdat->node_zones[zone];
cbe8dd4a 4460 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
a2f3aa02
DH
4461 /*
4462 * There can be holes in boot-time mem_map[]s
4463 * handed to this function. They do not
4464 * exist on hotplugged memory.
4465 */
4466 if (context == MEMMAP_EARLY) {
4467 if (!early_pfn_valid(pfn))
4468 continue;
4469 if (!early_pfn_in_nid(pfn, nid))
4470 continue;
3a80a7fa
MG
4471 if (!update_defer_init(pgdat, pfn, end_pfn,
4472 &nr_initialised))
4473 break;
a2f3aa02 4474 }
ac5d2539
MG
4475
4476 /*
4477 * Mark the block movable so that blocks are reserved for
4478 * movable at startup. This will force kernel allocations
4479 * to reserve their blocks rather than leaking throughout
4480 * the address space during boot when many long-lived
974a786e 4481 * kernel allocations are made.
ac5d2539
MG
4482 *
4483 * bitmap is created for zone's valid pfn range. but memmap
4484 * can be created for invalid pages (for alignment)
4485 * check here not to call set_pageblock_migratetype() against
4486 * pfn out of zone.
4487 */
4488 if (!(pfn & (pageblock_nr_pages - 1))) {
4489 struct page *page = pfn_to_page(pfn);
4490
4491 __init_single_page(page, pfn, zone, nid);
4492 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4493 } else {
4494 __init_single_pfn(pfn, zone, nid);
4495 }
1da177e4
LT
4496 }
4497}
4498
1e548deb 4499static void __meminit zone_init_free_lists(struct zone *zone)
1da177e4 4500{
7aeb09f9 4501 unsigned int order, t;
b2a0ac88
MG
4502 for_each_migratetype_order(order, t) {
4503 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1da177e4
LT
4504 zone->free_area[order].nr_free = 0;
4505 }
4506}
4507
4508#ifndef __HAVE_ARCH_MEMMAP_INIT
4509#define memmap_init(size, nid, zone, start_pfn) \
a2f3aa02 4510 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
1da177e4
LT
4511#endif
4512
7cd2b0a3 4513static int zone_batchsize(struct zone *zone)
e7c8d5c9 4514{
3a6be87f 4515#ifdef CONFIG_MMU
e7c8d5c9
CL
4516 int batch;
4517
4518 /*
4519 * The per-cpu-pages pools are set to around 1000th of the
ba56e91c 4520 * size of the zone. But no more than 1/2 of a meg.
e7c8d5c9
CL
4521 *
4522 * OK, so we don't know how big the cache is. So guess.
4523 */
b40da049 4524 batch = zone->managed_pages / 1024;
ba56e91c
SR
4525 if (batch * PAGE_SIZE > 512 * 1024)
4526 batch = (512 * 1024) / PAGE_SIZE;
e7c8d5c9
CL
4527 batch /= 4; /* We effectively *= 4 below */
4528 if (batch < 1)
4529 batch = 1;
4530
4531 /*
0ceaacc9
NP
4532 * Clamp the batch to a 2^n - 1 value. Having a power
4533 * of 2 value was found to be more likely to have
4534 * suboptimal cache aliasing properties in some cases.
e7c8d5c9 4535 *
0ceaacc9
NP
4536 * For example if 2 tasks are alternately allocating
4537 * batches of pages, one task can end up with a lot
4538 * of pages of one half of the possible page colors
4539 * and the other with pages of the other colors.
e7c8d5c9 4540 */
9155203a 4541 batch = rounddown_pow_of_two(batch + batch/2) - 1;
ba56e91c 4542
e7c8d5c9 4543 return batch;
3a6be87f
DH
4544
4545#else
4546 /* The deferral and batching of frees should be suppressed under NOMMU
4547 * conditions.
4548 *
4549 * The problem is that NOMMU needs to be able to allocate large chunks
4550 * of contiguous memory as there's no hardware page translation to
4551 * assemble apparent contiguous memory from discontiguous pages.
4552 *
4553 * Queueing large contiguous runs of pages for batching, however,
4554 * causes the pages to actually be freed in smaller chunks. As there
4555 * can be a significant delay between the individual batches being
4556 * recycled, this leads to the once large chunks of space being
4557 * fragmented and becoming unavailable for high-order allocations.
4558 */
4559 return 0;
4560#endif
e7c8d5c9
CL
4561}
4562
8d7a8fa9
CS
4563/*
4564 * pcp->high and pcp->batch values are related and dependent on one another:
4565 * ->batch must never be higher then ->high.
4566 * The following function updates them in a safe manner without read side
4567 * locking.
4568 *
4569 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4570 * those fields changing asynchronously (acording the the above rule).
4571 *
4572 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4573 * outside of boot time (or some other assurance that no concurrent updaters
4574 * exist).
4575 */
4576static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4577 unsigned long batch)
4578{
4579 /* start with a fail safe value for batch */
4580 pcp->batch = 1;
4581 smp_wmb();
4582
4583 /* Update high, then batch, in order */
4584 pcp->high = high;
4585 smp_wmb();
4586
4587 pcp->batch = batch;
4588}
4589
3664033c 4590/* a companion to pageset_set_high() */
4008bab7
CS
4591static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4592{
8d7a8fa9 4593 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4008bab7
CS
4594}
4595
88c90dbc 4596static void pageset_init(struct per_cpu_pageset *p)
2caaad41
CL
4597{
4598 struct per_cpu_pages *pcp;
5f8dcc21 4599 int migratetype;
2caaad41 4600
1c6fe946
MD
4601 memset(p, 0, sizeof(*p));
4602
3dfa5721 4603 pcp = &p->pcp;
2caaad41 4604 pcp->count = 0;
5f8dcc21
MG
4605 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4606 INIT_LIST_HEAD(&pcp->lists[migratetype]);
2caaad41
CL
4607}
4608
88c90dbc
CS
4609static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4610{
4611 pageset_init(p);
4612 pageset_set_batch(p, batch);
4613}
4614
8ad4b1fb 4615/*
3664033c 4616 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
8ad4b1fb
RS
4617 * to the value high for the pageset p.
4618 */
3664033c 4619static void pageset_set_high(struct per_cpu_pageset *p,
8ad4b1fb
RS
4620 unsigned long high)
4621{
8d7a8fa9
CS
4622 unsigned long batch = max(1UL, high / 4);
4623 if ((high / 4) > (PAGE_SHIFT * 8))
4624 batch = PAGE_SHIFT * 8;
8ad4b1fb 4625
8d7a8fa9 4626 pageset_update(&p->pcp, high, batch);
8ad4b1fb
RS
4627}
4628
7cd2b0a3
DR
4629static void pageset_set_high_and_batch(struct zone *zone,
4630 struct per_cpu_pageset *pcp)
56cef2b8 4631{
56cef2b8 4632 if (percpu_pagelist_fraction)
3664033c 4633 pageset_set_high(pcp,
56cef2b8
CS
4634 (zone->managed_pages /
4635 percpu_pagelist_fraction));
4636 else
4637 pageset_set_batch(pcp, zone_batchsize(zone));
4638}
4639
169f6c19
CS
4640static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4641{
4642 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4643
4644 pageset_init(pcp);
4645 pageset_set_high_and_batch(zone, pcp);
4646}
4647
4ed7e022 4648static void __meminit setup_zone_pageset(struct zone *zone)
319774e2
WF
4649{
4650 int cpu;
319774e2 4651 zone->pageset = alloc_percpu(struct per_cpu_pageset);
56cef2b8
CS
4652 for_each_possible_cpu(cpu)
4653 zone_pageset_init(zone, cpu);
319774e2
WF
4654}
4655
2caaad41 4656/*
99dcc3e5
CL
4657 * Allocate per cpu pagesets and initialize them.
4658 * Before this call only boot pagesets were available.
e7c8d5c9 4659 */
99dcc3e5 4660void __init setup_per_cpu_pageset(void)
e7c8d5c9 4661{
99dcc3e5 4662 struct zone *zone;
e7c8d5c9 4663
319774e2
WF
4664 for_each_populated_zone(zone)
4665 setup_zone_pageset(zone);
e7c8d5c9
CL
4666}
4667
577a32f6 4668static noinline __init_refok
cca448fe 4669int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
ed8ece2e
DH
4670{
4671 int i;
cca448fe 4672 size_t alloc_size;
ed8ece2e
DH
4673
4674 /*
4675 * The per-page waitqueue mechanism uses hashed waitqueues
4676 * per zone.
4677 */
02b694de
YG
4678 zone->wait_table_hash_nr_entries =
4679 wait_table_hash_nr_entries(zone_size_pages);
4680 zone->wait_table_bits =
4681 wait_table_bits(zone->wait_table_hash_nr_entries);
cca448fe
YG
4682 alloc_size = zone->wait_table_hash_nr_entries
4683 * sizeof(wait_queue_head_t);
4684
cd94b9db 4685 if (!slab_is_available()) {
cca448fe 4686 zone->wait_table = (wait_queue_head_t *)
6782832e
SS
4687 memblock_virt_alloc_node_nopanic(
4688 alloc_size, zone->zone_pgdat->node_id);
cca448fe
YG
4689 } else {
4690 /*
4691 * This case means that a zone whose size was 0 gets new memory
4692 * via memory hot-add.
4693 * But it may be the case that a new node was hot-added. In
4694 * this case vmalloc() will not be able to use this new node's
4695 * memory - this wait_table must be initialized to use this new
4696 * node itself as well.
4697 * To use this new node's memory, further consideration will be
4698 * necessary.
4699 */
8691f3a7 4700 zone->wait_table = vmalloc(alloc_size);
cca448fe
YG
4701 }
4702 if (!zone->wait_table)
4703 return -ENOMEM;
ed8ece2e 4704
b8af2941 4705 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
ed8ece2e 4706 init_waitqueue_head(zone->wait_table + i);
cca448fe
YG
4707
4708 return 0;
ed8ece2e
DH
4709}
4710
c09b4240 4711static __meminit void zone_pcp_init(struct zone *zone)
ed8ece2e 4712{
99dcc3e5
CL
4713 /*
4714 * per cpu subsystem is not up at this point. The following code
4715 * relies on the ability of the linker to provide the
4716 * offset of a (static) per cpu variable into the per cpu area.
4717 */
4718 zone->pageset = &boot_pageset;
ed8ece2e 4719
b38a8725 4720 if (populated_zone(zone))
99dcc3e5
CL
4721 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4722 zone->name, zone->present_pages,
4723 zone_batchsize(zone));
ed8ece2e
DH
4724}
4725
4ed7e022 4726int __meminit init_currently_empty_zone(struct zone *zone,
718127cc 4727 unsigned long zone_start_pfn,
b171e409 4728 unsigned long size)
ed8ece2e
DH
4729{
4730 struct pglist_data *pgdat = zone->zone_pgdat;
cca448fe
YG
4731 int ret;
4732 ret = zone_wait_table_init(zone, size);
4733 if (ret)
4734 return ret;
ed8ece2e
DH
4735 pgdat->nr_zones = zone_idx(zone) + 1;
4736
ed8ece2e
DH
4737 zone->zone_start_pfn = zone_start_pfn;
4738
708614e6
MG
4739 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4740 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4741 pgdat->node_id,
4742 (unsigned long)zone_idx(zone),
4743 zone_start_pfn, (zone_start_pfn + size));
4744
1e548deb 4745 zone_init_free_lists(zone);
718127cc
YG
4746
4747 return 0;
ed8ece2e
DH
4748}
4749
0ee332c1 4750#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
c713216d 4751#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
8a942fde 4752
c713216d
MG
4753/*
4754 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
c713216d 4755 */
8a942fde
MG
4756int __meminit __early_pfn_to_nid(unsigned long pfn,
4757 struct mminit_pfnnid_cache *state)
c713216d 4758{
c13291a5 4759 unsigned long start_pfn, end_pfn;
e76b63f8 4760 int nid;
7c243c71 4761
8a942fde
MG
4762 if (state->last_start <= pfn && pfn < state->last_end)
4763 return state->last_nid;
c713216d 4764
e76b63f8
YL
4765 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4766 if (nid != -1) {
8a942fde
MG
4767 state->last_start = start_pfn;
4768 state->last_end = end_pfn;
4769 state->last_nid = nid;
e76b63f8
YL
4770 }
4771
4772 return nid;
c713216d
MG
4773}
4774#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4775
c713216d 4776/**
6782832e 4777 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
88ca3b94 4778 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
6782832e 4779 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
c713216d 4780 *
7d018176
ZZ
4781 * If an architecture guarantees that all ranges registered contain no holes
4782 * and may be freed, this this function may be used instead of calling
4783 * memblock_free_early_nid() manually.
c713216d 4784 */
c13291a5 4785void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
cc289894 4786{
c13291a5
TH
4787 unsigned long start_pfn, end_pfn;
4788 int i, this_nid;
edbe7d23 4789
c13291a5
TH
4790 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4791 start_pfn = min(start_pfn, max_low_pfn);
4792 end_pfn = min(end_pfn, max_low_pfn);
edbe7d23 4793
c13291a5 4794 if (start_pfn < end_pfn)
6782832e
SS
4795 memblock_free_early_nid(PFN_PHYS(start_pfn),
4796 (end_pfn - start_pfn) << PAGE_SHIFT,
4797 this_nid);
edbe7d23 4798 }
edbe7d23 4799}
edbe7d23 4800
c713216d
MG
4801/**
4802 * sparse_memory_present_with_active_regions - Call memory_present for each active range
88ca3b94 4803 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
c713216d 4804 *
7d018176
ZZ
4805 * If an architecture guarantees that all ranges registered contain no holes and may
4806 * be freed, this function may be used instead of calling memory_present() manually.
c713216d
MG
4807 */
4808void __init sparse_memory_present_with_active_regions(int nid)
4809{
c13291a5
TH
4810 unsigned long start_pfn, end_pfn;
4811 int i, this_nid;
c713216d 4812
c13291a5
TH
4813 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4814 memory_present(this_nid, start_pfn, end_pfn);
c713216d
MG
4815}
4816
4817/**
4818 * get_pfn_range_for_nid - Return the start and end page frames for a node
88ca3b94
RD
4819 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4820 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4821 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
c713216d
MG
4822 *
4823 * It returns the start and end page frame of a node based on information
7d018176 4824 * provided by memblock_set_node(). If called for a node
c713216d 4825 * with no available memory, a warning is printed and the start and end
88ca3b94 4826 * PFNs will be 0.
c713216d 4827 */
a3142c8e 4828void __meminit get_pfn_range_for_nid(unsigned int nid,
c713216d
MG
4829 unsigned long *start_pfn, unsigned long *end_pfn)
4830{
c13291a5 4831 unsigned long this_start_pfn, this_end_pfn;
c713216d 4832 int i;
c13291a5 4833
c713216d
MG
4834 *start_pfn = -1UL;
4835 *end_pfn = 0;
4836
c13291a5
TH
4837 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4838 *start_pfn = min(*start_pfn, this_start_pfn);
4839 *end_pfn = max(*end_pfn, this_end_pfn);
c713216d
MG
4840 }
4841
633c0666 4842 if (*start_pfn == -1UL)
c713216d 4843 *start_pfn = 0;
c713216d
MG
4844}
4845
2a1e274a
MG
4846/*
4847 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4848 * assumption is made that zones within a node are ordered in monotonic
4849 * increasing memory addresses so that the "highest" populated zone is used
4850 */
b69a7288 4851static void __init find_usable_zone_for_movable(void)
2a1e274a
MG
4852{
4853 int zone_index;
4854 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4855 if (zone_index == ZONE_MOVABLE)
4856 continue;
4857
4858 if (arch_zone_highest_possible_pfn[zone_index] >
4859 arch_zone_lowest_possible_pfn[zone_index])
4860 break;
4861 }
4862
4863 VM_BUG_ON(zone_index == -1);
4864 movable_zone = zone_index;
4865}
4866
4867/*
4868 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
25985edc 4869 * because it is sized independent of architecture. Unlike the other zones,
2a1e274a
MG
4870 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4871 * in each node depending on the size of each node and how evenly kernelcore
4872 * is distributed. This helper function adjusts the zone ranges
4873 * provided by the architecture for a given node by using the end of the
4874 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4875 * zones within a node are in order of monotonic increases memory addresses
4876 */
b69a7288 4877static void __meminit adjust_zone_range_for_zone_movable(int nid,
2a1e274a
MG
4878 unsigned long zone_type,
4879 unsigned long node_start_pfn,
4880 unsigned long node_end_pfn,
4881 unsigned long *zone_start_pfn,
4882 unsigned long *zone_end_pfn)
4883{
4884 /* Only adjust if ZONE_MOVABLE is on this node */
4885 if (zone_movable_pfn[nid]) {
4886 /* Size ZONE_MOVABLE */
4887 if (zone_type == ZONE_MOVABLE) {
4888 *zone_start_pfn = zone_movable_pfn[nid];
4889 *zone_end_pfn = min(node_end_pfn,
4890 arch_zone_highest_possible_pfn[movable_zone]);
4891
4892 /* Adjust for ZONE_MOVABLE starting within this range */
4893 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4894 *zone_end_pfn > zone_movable_pfn[nid]) {
4895 *zone_end_pfn = zone_movable_pfn[nid];
4896
4897 /* Check if this whole range is within ZONE_MOVABLE */
4898 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4899 *zone_start_pfn = *zone_end_pfn;
4900 }
4901}
4902
c713216d
MG
4903/*
4904 * Return the number of pages a zone spans in a node, including holes
4905 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4906 */
6ea6e688 4907static unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d 4908 unsigned long zone_type,
7960aedd
ZY
4909 unsigned long node_start_pfn,
4910 unsigned long node_end_pfn,
c713216d
MG
4911 unsigned long *ignored)
4912{
c713216d
MG
4913 unsigned long zone_start_pfn, zone_end_pfn;
4914
b5685e92 4915 /* When hotadd a new node from cpu_up(), the node should be empty */
f9126ab9
XQ
4916 if (!node_start_pfn && !node_end_pfn)
4917 return 0;
4918
7960aedd 4919 /* Get the start and end of the zone */
c713216d
MG
4920 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4921 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2a1e274a
MG
4922 adjust_zone_range_for_zone_movable(nid, zone_type,
4923 node_start_pfn, node_end_pfn,
4924 &zone_start_pfn, &zone_end_pfn);
c713216d
MG
4925
4926 /* Check that this node has pages within the zone's required range */
4927 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4928 return 0;
4929
4930 /* Move the zone boundaries inside the node if necessary */
4931 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4932 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4933
4934 /* Return the spanned pages */
4935 return zone_end_pfn - zone_start_pfn;
4936}
4937
4938/*
4939 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
88ca3b94 4940 * then all holes in the requested range will be accounted for.
c713216d 4941 */
32996250 4942unsigned long __meminit __absent_pages_in_range(int nid,
c713216d
MG
4943 unsigned long range_start_pfn,
4944 unsigned long range_end_pfn)
4945{
96e907d1
TH
4946 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4947 unsigned long start_pfn, end_pfn;
4948 int i;
c713216d 4949
96e907d1
TH
4950 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4951 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4952 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4953 nr_absent -= end_pfn - start_pfn;
c713216d 4954 }
96e907d1 4955 return nr_absent;
c713216d
MG
4956}
4957
4958/**
4959 * absent_pages_in_range - Return number of page frames in holes within a range
4960 * @start_pfn: The start PFN to start searching for holes
4961 * @end_pfn: The end PFN to stop searching for holes
4962 *
88ca3b94 4963 * It returns the number of pages frames in memory holes within a range.
c713216d
MG
4964 */
4965unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4966 unsigned long end_pfn)
4967{
4968 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4969}
4970
4971/* Return the number of page frames in holes in a zone on a node */
6ea6e688 4972static unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d 4973 unsigned long zone_type,
7960aedd
ZY
4974 unsigned long node_start_pfn,
4975 unsigned long node_end_pfn,
c713216d
MG
4976 unsigned long *ignored)
4977{
96e907d1
TH
4978 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4979 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
9c7cd687
MG
4980 unsigned long zone_start_pfn, zone_end_pfn;
4981
b5685e92 4982 /* When hotadd a new node from cpu_up(), the node should be empty */
f9126ab9
XQ
4983 if (!node_start_pfn && !node_end_pfn)
4984 return 0;
4985
96e907d1
TH
4986 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4987 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
9c7cd687 4988
2a1e274a
MG
4989 adjust_zone_range_for_zone_movable(nid, zone_type,
4990 node_start_pfn, node_end_pfn,
4991 &zone_start_pfn, &zone_end_pfn);
9c7cd687 4992 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
c713216d 4993}
0e0b864e 4994
0ee332c1 4995#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6ea6e688 4996static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d 4997 unsigned long zone_type,
7960aedd
ZY
4998 unsigned long node_start_pfn,
4999 unsigned long node_end_pfn,
c713216d
MG
5000 unsigned long *zones_size)
5001{
5002 return zones_size[zone_type];
5003}
5004
6ea6e688 5005static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d 5006 unsigned long zone_type,
7960aedd
ZY
5007 unsigned long node_start_pfn,
5008 unsigned long node_end_pfn,
c713216d
MG
5009 unsigned long *zholes_size)
5010{
5011 if (!zholes_size)
5012 return 0;
5013
5014 return zholes_size[zone_type];
5015}
20e6926d 5016
0ee332c1 5017#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 5018
a3142c8e 5019static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
7960aedd
ZY
5020 unsigned long node_start_pfn,
5021 unsigned long node_end_pfn,
5022 unsigned long *zones_size,
5023 unsigned long *zholes_size)
c713216d 5024{
febd5949 5025 unsigned long realtotalpages = 0, totalpages = 0;
c713216d
MG
5026 enum zone_type i;
5027
febd5949
GZ
5028 for (i = 0; i < MAX_NR_ZONES; i++) {
5029 struct zone *zone = pgdat->node_zones + i;
5030 unsigned long size, real_size;
c713216d 5031
febd5949
GZ
5032 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5033 node_start_pfn,
5034 node_end_pfn,
5035 zones_size);
5036 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
7960aedd
ZY
5037 node_start_pfn, node_end_pfn,
5038 zholes_size);
febd5949
GZ
5039 zone->spanned_pages = size;
5040 zone->present_pages = real_size;
5041
5042 totalpages += size;
5043 realtotalpages += real_size;
5044 }
5045
5046 pgdat->node_spanned_pages = totalpages;
c713216d
MG
5047 pgdat->node_present_pages = realtotalpages;
5048 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5049 realtotalpages);
5050}
5051
835c134e
MG
5052#ifndef CONFIG_SPARSEMEM
5053/*
5054 * Calculate the size of the zone->blockflags rounded to an unsigned long
d9c23400
MG
5055 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5056 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
835c134e
MG
5057 * round what is now in bits to nearest long in bits, then return it in
5058 * bytes.
5059 */
7c45512d 5060static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
835c134e
MG
5061{
5062 unsigned long usemapsize;
5063
7c45512d 5064 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
d9c23400
MG
5065 usemapsize = roundup(zonesize, pageblock_nr_pages);
5066 usemapsize = usemapsize >> pageblock_order;
835c134e
MG
5067 usemapsize *= NR_PAGEBLOCK_BITS;
5068 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5069
5070 return usemapsize / 8;
5071}
5072
5073static void __init setup_usemap(struct pglist_data *pgdat,
7c45512d
LT
5074 struct zone *zone,
5075 unsigned long zone_start_pfn,
5076 unsigned long zonesize)
835c134e 5077{
7c45512d 5078 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
835c134e 5079 zone->pageblock_flags = NULL;
58a01a45 5080 if (usemapsize)
6782832e
SS
5081 zone->pageblock_flags =
5082 memblock_virt_alloc_node_nopanic(usemapsize,
5083 pgdat->node_id);
835c134e
MG
5084}
5085#else
7c45512d
LT
5086static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5087 unsigned long zone_start_pfn, unsigned long zonesize) {}
835c134e
MG
5088#endif /* CONFIG_SPARSEMEM */
5089
d9c23400 5090#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
ba72cb8c 5091
d9c23400 5092/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
15ca220e 5093void __paginginit set_pageblock_order(void)
d9c23400 5094{
955c1cd7
AM
5095 unsigned int order;
5096
d9c23400
MG
5097 /* Check that pageblock_nr_pages has not already been setup */
5098 if (pageblock_order)
5099 return;
5100
955c1cd7
AM
5101 if (HPAGE_SHIFT > PAGE_SHIFT)
5102 order = HUGETLB_PAGE_ORDER;
5103 else
5104 order = MAX_ORDER - 1;
5105
d9c23400
MG
5106 /*
5107 * Assume the largest contiguous order of interest is a huge page.
955c1cd7
AM
5108 * This value may be variable depending on boot parameters on IA64 and
5109 * powerpc.
d9c23400
MG
5110 */
5111 pageblock_order = order;
5112}
5113#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5114
ba72cb8c
MG
5115/*
5116 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
955c1cd7
AM
5117 * is unused as pageblock_order is set at compile-time. See
5118 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5119 * the kernel config
ba72cb8c 5120 */
15ca220e 5121void __paginginit set_pageblock_order(void)
ba72cb8c 5122{
ba72cb8c 5123}
d9c23400
MG
5124
5125#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5126
01cefaef
JL
5127static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5128 unsigned long present_pages)
5129{
5130 unsigned long pages = spanned_pages;
5131
5132 /*
5133 * Provide a more accurate estimation if there are holes within
5134 * the zone and SPARSEMEM is in use. If there are holes within the
5135 * zone, each populated memory region may cost us one or two extra
5136 * memmap pages due to alignment because memmap pages for each
5137 * populated regions may not naturally algined on page boundary.
5138 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5139 */
5140 if (spanned_pages > present_pages + (present_pages >> 4) &&
5141 IS_ENABLED(CONFIG_SPARSEMEM))
5142 pages = present_pages;
5143
5144 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5145}
5146
1da177e4
LT
5147/*
5148 * Set up the zone data structures:
5149 * - mark all pages reserved
5150 * - mark all memory queues empty
5151 * - clear the memory bitmaps
6527af5d
MK
5152 *
5153 * NOTE: pgdat should get zeroed by caller.
1da177e4 5154 */
7f3eb55b 5155static void __paginginit free_area_init_core(struct pglist_data *pgdat)
1da177e4 5156{
2f1b6248 5157 enum zone_type j;
ed8ece2e 5158 int nid = pgdat->node_id;
1da177e4 5159 unsigned long zone_start_pfn = pgdat->node_start_pfn;
718127cc 5160 int ret;
1da177e4 5161
208d54e5 5162 pgdat_resize_init(pgdat);
8177a420
AA
5163#ifdef CONFIG_NUMA_BALANCING
5164 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5165 pgdat->numabalancing_migrate_nr_pages = 0;
5166 pgdat->numabalancing_migrate_next_window = jiffies;
5167#endif
1da177e4 5168 init_waitqueue_head(&pgdat->kswapd_wait);
5515061d 5169 init_waitqueue_head(&pgdat->pfmemalloc_wait);
eefa864b 5170 pgdat_page_ext_init(pgdat);
5f63b720 5171
1da177e4
LT
5172 for (j = 0; j < MAX_NR_ZONES; j++) {
5173 struct zone *zone = pgdat->node_zones + j;
9feedc9d 5174 unsigned long size, realsize, freesize, memmap_pages;
1da177e4 5175
febd5949
GZ
5176 size = zone->spanned_pages;
5177 realsize = freesize = zone->present_pages;
1da177e4 5178
0e0b864e 5179 /*
9feedc9d 5180 * Adjust freesize so that it accounts for how much memory
0e0b864e
MG
5181 * is used by this zone for memmap. This affects the watermark
5182 * and per-cpu initialisations
5183 */
01cefaef 5184 memmap_pages = calc_memmap_size(size, realsize);
ba914f48
ZH
5185 if (!is_highmem_idx(j)) {
5186 if (freesize >= memmap_pages) {
5187 freesize -= memmap_pages;
5188 if (memmap_pages)
5189 printk(KERN_DEBUG
5190 " %s zone: %lu pages used for memmap\n",
5191 zone_names[j], memmap_pages);
5192 } else
5193 printk(KERN_WARNING
5194 " %s zone: %lu pages exceeds freesize %lu\n",
5195 zone_names[j], memmap_pages, freesize);
5196 }
0e0b864e 5197
6267276f 5198 /* Account for reserved pages */
9feedc9d
JL
5199 if (j == 0 && freesize > dma_reserve) {
5200 freesize -= dma_reserve;
d903ef9f 5201 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
6267276f 5202 zone_names[0], dma_reserve);
0e0b864e
MG
5203 }
5204
98d2b0eb 5205 if (!is_highmem_idx(j))
9feedc9d 5206 nr_kernel_pages += freesize;
01cefaef
JL
5207 /* Charge for highmem memmap if there are enough kernel pages */
5208 else if (nr_kernel_pages > memmap_pages * 2)
5209 nr_kernel_pages -= memmap_pages;
9feedc9d 5210 nr_all_pages += freesize;
1da177e4 5211
9feedc9d
JL
5212 /*
5213 * Set an approximate value for lowmem here, it will be adjusted
5214 * when the bootmem allocator frees pages into the buddy system.
5215 * And all highmem pages will be managed by the buddy system.
5216 */
5217 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
9614634f 5218#ifdef CONFIG_NUMA
d5f541ed 5219 zone->node = nid;
9feedc9d 5220 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
9614634f 5221 / 100;
9feedc9d 5222 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
9614634f 5223#endif
1da177e4
LT
5224 zone->name = zone_names[j];
5225 spin_lock_init(&zone->lock);
5226 spin_lock_init(&zone->lru_lock);
bdc8cb98 5227 zone_seqlock_init(zone);
1da177e4 5228 zone->zone_pgdat = pgdat;
ed8ece2e 5229 zone_pcp_init(zone);
81c0a2bb
JW
5230
5231 /* For bootup, initialized properly in watermark setup */
5232 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5233
bea8c150 5234 lruvec_init(&zone->lruvec);
1da177e4
LT
5235 if (!size)
5236 continue;
5237
955c1cd7 5238 set_pageblock_order();
7c45512d 5239 setup_usemap(pgdat, zone, zone_start_pfn, size);
b171e409 5240 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
718127cc 5241 BUG_ON(ret);
76cdd58e 5242 memmap_init(size, nid, j, zone_start_pfn);
1da177e4 5243 zone_start_pfn += size;
1da177e4
LT
5244 }
5245}
5246
577a32f6 5247static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
1da177e4 5248{
b0aeba74 5249 unsigned long __maybe_unused start = 0;
a1c34a3b
LA
5250 unsigned long __maybe_unused offset = 0;
5251
1da177e4
LT
5252 /* Skip empty nodes */
5253 if (!pgdat->node_spanned_pages)
5254 return;
5255
d41dee36 5256#ifdef CONFIG_FLAT_NODE_MEM_MAP
b0aeba74
TL
5257 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5258 offset = pgdat->node_start_pfn - start;
1da177e4
LT
5259 /* ia64 gets its own node_mem_map, before this, without bootmem */
5260 if (!pgdat->node_mem_map) {
b0aeba74 5261 unsigned long size, end;
d41dee36
AW
5262 struct page *map;
5263
e984bb43
BP
5264 /*
5265 * The zone's endpoints aren't required to be MAX_ORDER
5266 * aligned but the node_mem_map endpoints must be in order
5267 * for the buddy allocator to function correctly.
5268 */
108bcc96 5269 end = pgdat_end_pfn(pgdat);
e984bb43
BP
5270 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5271 size = (end - start) * sizeof(struct page);
6f167ec7
DH
5272 map = alloc_remap(pgdat->node_id, size);
5273 if (!map)
6782832e
SS
5274 map = memblock_virt_alloc_node_nopanic(size,
5275 pgdat->node_id);
a1c34a3b 5276 pgdat->node_mem_map = map + offset;
1da177e4 5277 }
12d810c1 5278#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
5279 /*
5280 * With no DISCONTIG, the global mem_map is just set as node 0's
5281 */
c713216d 5282 if (pgdat == NODE_DATA(0)) {
1da177e4 5283 mem_map = NODE_DATA(0)->node_mem_map;
a1c34a3b 5284#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
c713216d 5285 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
a1c34a3b 5286 mem_map -= offset;
0ee332c1 5287#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 5288 }
1da177e4 5289#endif
d41dee36 5290#endif /* CONFIG_FLAT_NODE_MEM_MAP */
1da177e4
LT
5291}
5292
9109fb7b
JW
5293void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5294 unsigned long node_start_pfn, unsigned long *zholes_size)
1da177e4 5295{
9109fb7b 5296 pg_data_t *pgdat = NODE_DATA(nid);
7960aedd
ZY
5297 unsigned long start_pfn = 0;
5298 unsigned long end_pfn = 0;
9109fb7b 5299
88fdf75d 5300 /* pg_data_t should be reset to zero when it's allocated */
8783b6e2 5301 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
88fdf75d 5302
3a80a7fa 5303 reset_deferred_meminit(pgdat);
1da177e4
LT
5304 pgdat->node_id = nid;
5305 pgdat->node_start_pfn = node_start_pfn;
7960aedd
ZY
5306#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5307 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
8d29e18a 5308 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
4ada0c5a
ZL
5309 (u64)start_pfn << PAGE_SHIFT,
5310 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
7960aedd
ZY
5311#endif
5312 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5313 zones_size, zholes_size);
1da177e4
LT
5314
5315 alloc_node_mem_map(pgdat);
e8c27ac9
YL
5316#ifdef CONFIG_FLAT_NODE_MEM_MAP
5317 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5318 nid, (unsigned long)pgdat,
5319 (unsigned long)pgdat->node_mem_map);
5320#endif
1da177e4 5321
7f3eb55b 5322 free_area_init_core(pgdat);
1da177e4
LT
5323}
5324
0ee332c1 5325#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
418508c1
MS
5326
5327#if MAX_NUMNODES > 1
5328/*
5329 * Figure out the number of possible node ids.
5330 */
f9872caf 5331void __init setup_nr_node_ids(void)
418508c1 5332{
904a9553 5333 unsigned int highest;
418508c1 5334
904a9553 5335 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
418508c1
MS
5336 nr_node_ids = highest + 1;
5337}
418508c1
MS
5338#endif
5339
1e01979c
TH
5340/**
5341 * node_map_pfn_alignment - determine the maximum internode alignment
5342 *
5343 * This function should be called after node map is populated and sorted.
5344 * It calculates the maximum power of two alignment which can distinguish
5345 * all the nodes.
5346 *
5347 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5348 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5349 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5350 * shifted, 1GiB is enough and this function will indicate so.
5351 *
5352 * This is used to test whether pfn -> nid mapping of the chosen memory
5353 * model has fine enough granularity to avoid incorrect mapping for the
5354 * populated node map.
5355 *
5356 * Returns the determined alignment in pfn's. 0 if there is no alignment
5357 * requirement (single node).
5358 */
5359unsigned long __init node_map_pfn_alignment(void)
5360{
5361 unsigned long accl_mask = 0, last_end = 0;
c13291a5 5362 unsigned long start, end, mask;
1e01979c 5363 int last_nid = -1;
c13291a5 5364 int i, nid;
1e01979c 5365
c13291a5 5366 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
1e01979c
TH
5367 if (!start || last_nid < 0 || last_nid == nid) {
5368 last_nid = nid;
5369 last_end = end;
5370 continue;
5371 }
5372
5373 /*
5374 * Start with a mask granular enough to pin-point to the
5375 * start pfn and tick off bits one-by-one until it becomes
5376 * too coarse to separate the current node from the last.
5377 */
5378 mask = ~((1 << __ffs(start)) - 1);
5379 while (mask && last_end <= (start & (mask << 1)))
5380 mask <<= 1;
5381
5382 /* accumulate all internode masks */
5383 accl_mask |= mask;
5384 }
5385
5386 /* convert mask to number of pages */
5387 return ~accl_mask + 1;
5388}
5389
a6af2bc3 5390/* Find the lowest pfn for a node */
b69a7288 5391static unsigned long __init find_min_pfn_for_node(int nid)
c713216d 5392{
a6af2bc3 5393 unsigned long min_pfn = ULONG_MAX;
c13291a5
TH
5394 unsigned long start_pfn;
5395 int i;
1abbfb41 5396
c13291a5
TH
5397 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5398 min_pfn = min(min_pfn, start_pfn);
c713216d 5399
a6af2bc3
MG
5400 if (min_pfn == ULONG_MAX) {
5401 printk(KERN_WARNING
2bc0d261 5402 "Could not find start_pfn for node %d\n", nid);
a6af2bc3
MG
5403 return 0;
5404 }
5405
5406 return min_pfn;
c713216d
MG
5407}
5408
5409/**
5410 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5411 *
5412 * It returns the minimum PFN based on information provided via
7d018176 5413 * memblock_set_node().
c713216d
MG
5414 */
5415unsigned long __init find_min_pfn_with_active_regions(void)
5416{
5417 return find_min_pfn_for_node(MAX_NUMNODES);
5418}
5419
37b07e41
LS
5420/*
5421 * early_calculate_totalpages()
5422 * Sum pages in active regions for movable zone.
4b0ef1fe 5423 * Populate N_MEMORY for calculating usable_nodes.
37b07e41 5424 */
484f51f8 5425static unsigned long __init early_calculate_totalpages(void)
7e63efef 5426{
7e63efef 5427 unsigned long totalpages = 0;
c13291a5
TH
5428 unsigned long start_pfn, end_pfn;
5429 int i, nid;
5430
5431 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5432 unsigned long pages = end_pfn - start_pfn;
7e63efef 5433
37b07e41
LS
5434 totalpages += pages;
5435 if (pages)
4b0ef1fe 5436 node_set_state(nid, N_MEMORY);
37b07e41 5437 }
b8af2941 5438 return totalpages;
7e63efef
MG
5439}
5440
2a1e274a
MG
5441/*
5442 * Find the PFN the Movable zone begins in each node. Kernel memory
5443 * is spread evenly between nodes as long as the nodes have enough
5444 * memory. When they don't, some nodes will have more kernelcore than
5445 * others
5446 */
b224ef85 5447static void __init find_zone_movable_pfns_for_nodes(void)
2a1e274a
MG
5448{
5449 int i, nid;
5450 unsigned long usable_startpfn;
5451 unsigned long kernelcore_node, kernelcore_remaining;
66918dcd 5452 /* save the state before borrow the nodemask */
4b0ef1fe 5453 nodemask_t saved_node_state = node_states[N_MEMORY];
37b07e41 5454 unsigned long totalpages = early_calculate_totalpages();
4b0ef1fe 5455 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
136199f0 5456 struct memblock_region *r;
b2f3eebe
TC
5457
5458 /* Need to find movable_zone earlier when movable_node is specified. */
5459 find_usable_zone_for_movable();
5460
5461 /*
5462 * If movable_node is specified, ignore kernelcore and movablecore
5463 * options.
5464 */
5465 if (movable_node_is_enabled()) {
136199f0
EM
5466 for_each_memblock(memory, r) {
5467 if (!memblock_is_hotpluggable(r))
b2f3eebe
TC
5468 continue;
5469
136199f0 5470 nid = r->nid;
b2f3eebe 5471
136199f0 5472 usable_startpfn = PFN_DOWN(r->base);
b2f3eebe
TC
5473 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5474 min(usable_startpfn, zone_movable_pfn[nid]) :
5475 usable_startpfn;
5476 }
5477
5478 goto out2;
5479 }
2a1e274a 5480
7e63efef 5481 /*
b2f3eebe 5482 * If movablecore=nn[KMG] was specified, calculate what size of
7e63efef
MG
5483 * kernelcore that corresponds so that memory usable for
5484 * any allocation type is evenly spread. If both kernelcore
5485 * and movablecore are specified, then the value of kernelcore
5486 * will be used for required_kernelcore if it's greater than
5487 * what movablecore would have allowed.
5488 */
5489 if (required_movablecore) {
7e63efef
MG
5490 unsigned long corepages;
5491
5492 /*
5493 * Round-up so that ZONE_MOVABLE is at least as large as what
5494 * was requested by the user
5495 */
5496 required_movablecore =
5497 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
9fd745d4 5498 required_movablecore = min(totalpages, required_movablecore);
7e63efef
MG
5499 corepages = totalpages - required_movablecore;
5500
5501 required_kernelcore = max(required_kernelcore, corepages);
5502 }
5503
bde304bd
XQ
5504 /*
5505 * If kernelcore was not specified or kernelcore size is larger
5506 * than totalpages, there is no ZONE_MOVABLE.
5507 */
5508 if (!required_kernelcore || required_kernelcore >= totalpages)
66918dcd 5509 goto out;
2a1e274a
MG
5510
5511 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
2a1e274a
MG
5512 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5513
5514restart:
5515 /* Spread kernelcore memory as evenly as possible throughout nodes */
5516 kernelcore_node = required_kernelcore / usable_nodes;
4b0ef1fe 5517 for_each_node_state(nid, N_MEMORY) {
c13291a5
TH
5518 unsigned long start_pfn, end_pfn;
5519
2a1e274a
MG
5520 /*
5521 * Recalculate kernelcore_node if the division per node
5522 * now exceeds what is necessary to satisfy the requested
5523 * amount of memory for the kernel
5524 */
5525 if (required_kernelcore < kernelcore_node)
5526 kernelcore_node = required_kernelcore / usable_nodes;
5527
5528 /*
5529 * As the map is walked, we track how much memory is usable
5530 * by the kernel using kernelcore_remaining. When it is
5531 * 0, the rest of the node is usable by ZONE_MOVABLE
5532 */
5533 kernelcore_remaining = kernelcore_node;
5534
5535 /* Go through each range of PFNs within this node */
c13291a5 5536 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
2a1e274a
MG
5537 unsigned long size_pages;
5538
c13291a5 5539 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
2a1e274a
MG
5540 if (start_pfn >= end_pfn)
5541 continue;
5542
5543 /* Account for what is only usable for kernelcore */
5544 if (start_pfn < usable_startpfn) {
5545 unsigned long kernel_pages;
5546 kernel_pages = min(end_pfn, usable_startpfn)
5547 - start_pfn;
5548
5549 kernelcore_remaining -= min(kernel_pages,
5550 kernelcore_remaining);
5551 required_kernelcore -= min(kernel_pages,
5552 required_kernelcore);
5553
5554 /* Continue if range is now fully accounted */
5555 if (end_pfn <= usable_startpfn) {
5556
5557 /*
5558 * Push zone_movable_pfn to the end so
5559 * that if we have to rebalance
5560 * kernelcore across nodes, we will
5561 * not double account here
5562 */
5563 zone_movable_pfn[nid] = end_pfn;
5564 continue;
5565 }
5566 start_pfn = usable_startpfn;
5567 }
5568
5569 /*
5570 * The usable PFN range for ZONE_MOVABLE is from
5571 * start_pfn->end_pfn. Calculate size_pages as the
5572 * number of pages used as kernelcore
5573 */
5574 size_pages = end_pfn - start_pfn;
5575 if (size_pages > kernelcore_remaining)
5576 size_pages = kernelcore_remaining;
5577 zone_movable_pfn[nid] = start_pfn + size_pages;
5578
5579 /*
5580 * Some kernelcore has been met, update counts and
5581 * break if the kernelcore for this node has been
b8af2941 5582 * satisfied
2a1e274a
MG
5583 */
5584 required_kernelcore -= min(required_kernelcore,
5585 size_pages);
5586 kernelcore_remaining -= size_pages;
5587 if (!kernelcore_remaining)
5588 break;
5589 }
5590 }
5591
5592 /*
5593 * If there is still required_kernelcore, we do another pass with one
5594 * less node in the count. This will push zone_movable_pfn[nid] further
5595 * along on the nodes that still have memory until kernelcore is
b8af2941 5596 * satisfied
2a1e274a
MG
5597 */
5598 usable_nodes--;
5599 if (usable_nodes && required_kernelcore > usable_nodes)
5600 goto restart;
5601
b2f3eebe 5602out2:
2a1e274a
MG
5603 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5604 for (nid = 0; nid < MAX_NUMNODES; nid++)
5605 zone_movable_pfn[nid] =
5606 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
66918dcd 5607
20e6926d 5608out:
66918dcd 5609 /* restore the node_state */
4b0ef1fe 5610 node_states[N_MEMORY] = saved_node_state;
2a1e274a
MG
5611}
5612
4b0ef1fe
LJ
5613/* Any regular or high memory on that node ? */
5614static void check_for_memory(pg_data_t *pgdat, int nid)
37b07e41 5615{
37b07e41
LS
5616 enum zone_type zone_type;
5617
4b0ef1fe
LJ
5618 if (N_MEMORY == N_NORMAL_MEMORY)
5619 return;
5620
5621 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
37b07e41 5622 struct zone *zone = &pgdat->node_zones[zone_type];
b38a8725 5623 if (populated_zone(zone)) {
4b0ef1fe
LJ
5624 node_set_state(nid, N_HIGH_MEMORY);
5625 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5626 zone_type <= ZONE_NORMAL)
5627 node_set_state(nid, N_NORMAL_MEMORY);
d0048b0e
BL
5628 break;
5629 }
37b07e41 5630 }
37b07e41
LS
5631}
5632
c713216d
MG
5633/**
5634 * free_area_init_nodes - Initialise all pg_data_t and zone data
88ca3b94 5635 * @max_zone_pfn: an array of max PFNs for each zone
c713216d
MG
5636 *
5637 * This will call free_area_init_node() for each active node in the system.
7d018176 5638 * Using the page ranges provided by memblock_set_node(), the size of each
c713216d
MG
5639 * zone in each node and their holes is calculated. If the maximum PFN
5640 * between two adjacent zones match, it is assumed that the zone is empty.
5641 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5642 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5643 * starts where the previous one ended. For example, ZONE_DMA32 starts
5644 * at arch_max_dma_pfn.
5645 */
5646void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5647{
c13291a5
TH
5648 unsigned long start_pfn, end_pfn;
5649 int i, nid;
a6af2bc3 5650
c713216d
MG
5651 /* Record where the zone boundaries are */
5652 memset(arch_zone_lowest_possible_pfn, 0,
5653 sizeof(arch_zone_lowest_possible_pfn));
5654 memset(arch_zone_highest_possible_pfn, 0,
5655 sizeof(arch_zone_highest_possible_pfn));
5656 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5657 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5658 for (i = 1; i < MAX_NR_ZONES; i++) {
2a1e274a
MG
5659 if (i == ZONE_MOVABLE)
5660 continue;
c713216d
MG
5661 arch_zone_lowest_possible_pfn[i] =
5662 arch_zone_highest_possible_pfn[i-1];
5663 arch_zone_highest_possible_pfn[i] =
5664 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5665 }
2a1e274a
MG
5666 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5667 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5668
5669 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5670 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
b224ef85 5671 find_zone_movable_pfns_for_nodes();
c713216d 5672
c713216d 5673 /* Print out the zone ranges */
f88dfff5 5674 pr_info("Zone ranges:\n");
2a1e274a
MG
5675 for (i = 0; i < MAX_NR_ZONES; i++) {
5676 if (i == ZONE_MOVABLE)
5677 continue;
f88dfff5 5678 pr_info(" %-8s ", zone_names[i]);
72f0ba02
DR
5679 if (arch_zone_lowest_possible_pfn[i] ==
5680 arch_zone_highest_possible_pfn[i])
f88dfff5 5681 pr_cont("empty\n");
72f0ba02 5682 else
8d29e18a
JG
5683 pr_cont("[mem %#018Lx-%#018Lx]\n",
5684 (u64)arch_zone_lowest_possible_pfn[i]
5685 << PAGE_SHIFT,
5686 ((u64)arch_zone_highest_possible_pfn[i]
a62e2f4f 5687 << PAGE_SHIFT) - 1);
2a1e274a
MG
5688 }
5689
5690 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
f88dfff5 5691 pr_info("Movable zone start for each node\n");
2a1e274a
MG
5692 for (i = 0; i < MAX_NUMNODES; i++) {
5693 if (zone_movable_pfn[i])
8d29e18a
JG
5694 pr_info(" Node %d: %#018Lx\n", i,
5695 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
2a1e274a 5696 }
c713216d 5697
f2d52fe5 5698 /* Print out the early node map */
f88dfff5 5699 pr_info("Early memory node ranges\n");
c13291a5 5700 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
8d29e18a
JG
5701 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5702 (u64)start_pfn << PAGE_SHIFT,
5703 ((u64)end_pfn << PAGE_SHIFT) - 1);
c713216d
MG
5704
5705 /* Initialise every node */
708614e6 5706 mminit_verify_pageflags_layout();
8ef82866 5707 setup_nr_node_ids();
c713216d
MG
5708 for_each_online_node(nid) {
5709 pg_data_t *pgdat = NODE_DATA(nid);
9109fb7b 5710 free_area_init_node(nid, NULL,
c713216d 5711 find_min_pfn_for_node(nid), NULL);
37b07e41
LS
5712
5713 /* Any memory on that node */
5714 if (pgdat->node_present_pages)
4b0ef1fe
LJ
5715 node_set_state(nid, N_MEMORY);
5716 check_for_memory(pgdat, nid);
c713216d
MG
5717 }
5718}
2a1e274a 5719
7e63efef 5720static int __init cmdline_parse_core(char *p, unsigned long *core)
2a1e274a
MG
5721{
5722 unsigned long long coremem;
5723 if (!p)
5724 return -EINVAL;
5725
5726 coremem = memparse(p, &p);
7e63efef 5727 *core = coremem >> PAGE_SHIFT;
2a1e274a 5728
7e63efef 5729 /* Paranoid check that UL is enough for the coremem value */
2a1e274a
MG
5730 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5731
5732 return 0;
5733}
ed7ed365 5734
7e63efef
MG
5735/*
5736 * kernelcore=size sets the amount of memory for use for allocations that
5737 * cannot be reclaimed or migrated.
5738 */
5739static int __init cmdline_parse_kernelcore(char *p)
5740{
5741 return cmdline_parse_core(p, &required_kernelcore);
5742}
5743
5744/*
5745 * movablecore=size sets the amount of memory for use for allocations that
5746 * can be reclaimed or migrated.
5747 */
5748static int __init cmdline_parse_movablecore(char *p)
5749{
5750 return cmdline_parse_core(p, &required_movablecore);
5751}
5752
ed7ed365 5753early_param("kernelcore", cmdline_parse_kernelcore);
7e63efef 5754early_param("movablecore", cmdline_parse_movablecore);
ed7ed365 5755
0ee332c1 5756#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
c713216d 5757
c3d5f5f0
JL
5758void adjust_managed_page_count(struct page *page, long count)
5759{
5760 spin_lock(&managed_page_count_lock);
5761 page_zone(page)->managed_pages += count;
5762 totalram_pages += count;
3dcc0571
JL
5763#ifdef CONFIG_HIGHMEM
5764 if (PageHighMem(page))
5765 totalhigh_pages += count;
5766#endif
c3d5f5f0
JL
5767 spin_unlock(&managed_page_count_lock);
5768}
3dcc0571 5769EXPORT_SYMBOL(adjust_managed_page_count);
c3d5f5f0 5770
11199692 5771unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
69afade7 5772{
11199692
JL
5773 void *pos;
5774 unsigned long pages = 0;
69afade7 5775
11199692
JL
5776 start = (void *)PAGE_ALIGN((unsigned long)start);
5777 end = (void *)((unsigned long)end & PAGE_MASK);
5778 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
dbe67df4 5779 if ((unsigned int)poison <= 0xFF)
11199692
JL
5780 memset(pos, poison, PAGE_SIZE);
5781 free_reserved_page(virt_to_page(pos));
69afade7
JL
5782 }
5783
5784 if (pages && s)
11199692 5785 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
69afade7
JL
5786 s, pages << (PAGE_SHIFT - 10), start, end);
5787
5788 return pages;
5789}
11199692 5790EXPORT_SYMBOL(free_reserved_area);
69afade7 5791
cfa11e08
JL
5792#ifdef CONFIG_HIGHMEM
5793void free_highmem_page(struct page *page)
5794{
5795 __free_reserved_page(page);
5796 totalram_pages++;
7b4b2a0d 5797 page_zone(page)->managed_pages++;
cfa11e08
JL
5798 totalhigh_pages++;
5799}
5800#endif
5801
7ee3d4e8
JL
5802
5803void __init mem_init_print_info(const char *str)
5804{
5805 unsigned long physpages, codesize, datasize, rosize, bss_size;
5806 unsigned long init_code_size, init_data_size;
5807
5808 physpages = get_num_physpages();
5809 codesize = _etext - _stext;
5810 datasize = _edata - _sdata;
5811 rosize = __end_rodata - __start_rodata;
5812 bss_size = __bss_stop - __bss_start;
5813 init_data_size = __init_end - __init_begin;
5814 init_code_size = _einittext - _sinittext;
5815
5816 /*
5817 * Detect special cases and adjust section sizes accordingly:
5818 * 1) .init.* may be embedded into .data sections
5819 * 2) .init.text.* may be out of [__init_begin, __init_end],
5820 * please refer to arch/tile/kernel/vmlinux.lds.S.
5821 * 3) .rodata.* may be embedded into .text or .data sections.
5822 */
5823#define adj_init_size(start, end, size, pos, adj) \
b8af2941
PK
5824 do { \
5825 if (start <= pos && pos < end && size > adj) \
5826 size -= adj; \
5827 } while (0)
7ee3d4e8
JL
5828
5829 adj_init_size(__init_begin, __init_end, init_data_size,
5830 _sinittext, init_code_size);
5831 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5832 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5833 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5834 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5835
5836#undef adj_init_size
5837
f88dfff5 5838 pr_info("Memory: %luK/%luK available "
7ee3d4e8 5839 "(%luK kernel code, %luK rwdata, %luK rodata, "
e48322ab 5840 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
7ee3d4e8
JL
5841#ifdef CONFIG_HIGHMEM
5842 ", %luK highmem"
5843#endif
5844 "%s%s)\n",
5845 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5846 codesize >> 10, datasize >> 10, rosize >> 10,
5847 (init_data_size + init_code_size) >> 10, bss_size >> 10,
e48322ab
PK
5848 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5849 totalcma_pages << (PAGE_SHIFT-10),
7ee3d4e8
JL
5850#ifdef CONFIG_HIGHMEM
5851 totalhigh_pages << (PAGE_SHIFT-10),
5852#endif
5853 str ? ", " : "", str ? str : "");
5854}
5855
0e0b864e 5856/**
88ca3b94
RD
5857 * set_dma_reserve - set the specified number of pages reserved in the first zone
5858 * @new_dma_reserve: The number of pages to mark reserved
0e0b864e 5859 *
013110a7 5860 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
0e0b864e
MG
5861 * In the DMA zone, a significant percentage may be consumed by kernel image
5862 * and other unfreeable allocations which can skew the watermarks badly. This
88ca3b94
RD
5863 * function may optionally be used to account for unfreeable pages in the
5864 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5865 * smaller per-cpu batchsize.
0e0b864e
MG
5866 */
5867void __init set_dma_reserve(unsigned long new_dma_reserve)
5868{
5869 dma_reserve = new_dma_reserve;
5870}
5871
1da177e4
LT
5872void __init free_area_init(unsigned long *zones_size)
5873{
9109fb7b 5874 free_area_init_node(0, zones_size,
1da177e4
LT
5875 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5876}
1da177e4 5877
1da177e4
LT
5878static int page_alloc_cpu_notify(struct notifier_block *self,
5879 unsigned long action, void *hcpu)
5880{
5881 int cpu = (unsigned long)hcpu;
1da177e4 5882
8bb78442 5883 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
f0cb3c76 5884 lru_add_drain_cpu(cpu);
9f8f2172
CL
5885 drain_pages(cpu);
5886
5887 /*
5888 * Spill the event counters of the dead processor
5889 * into the current processors event counters.
5890 * This artificially elevates the count of the current
5891 * processor.
5892 */
f8891e5e 5893 vm_events_fold_cpu(cpu);
9f8f2172
CL
5894
5895 /*
5896 * Zero the differential counters of the dead processor
5897 * so that the vm statistics are consistent.
5898 *
5899 * This is only okay since the processor is dead and cannot
5900 * race with what we are doing.
5901 */
2bb921e5 5902 cpu_vm_stats_fold(cpu);
1da177e4
LT
5903 }
5904 return NOTIFY_OK;
5905}
1da177e4
LT
5906
5907void __init page_alloc_init(void)
5908{
5909 hotcpu_notifier(page_alloc_cpu_notify, 0);
5910}
5911
cb45b0e9 5912/*
34b10060 5913 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
cb45b0e9
HA
5914 * or min_free_kbytes changes.
5915 */
5916static void calculate_totalreserve_pages(void)
5917{
5918 struct pglist_data *pgdat;
5919 unsigned long reserve_pages = 0;
2f6726e5 5920 enum zone_type i, j;
cb45b0e9
HA
5921
5922 for_each_online_pgdat(pgdat) {
5923 for (i = 0; i < MAX_NR_ZONES; i++) {
5924 struct zone *zone = pgdat->node_zones + i;
3484b2de 5925 long max = 0;
cb45b0e9
HA
5926
5927 /* Find valid and maximum lowmem_reserve in the zone */
5928 for (j = i; j < MAX_NR_ZONES; j++) {
5929 if (zone->lowmem_reserve[j] > max)
5930 max = zone->lowmem_reserve[j];
5931 }
5932
41858966
MG
5933 /* we treat the high watermark as reserved pages. */
5934 max += high_wmark_pages(zone);
cb45b0e9 5935
b40da049
JL
5936 if (max > zone->managed_pages)
5937 max = zone->managed_pages;
a8d01437
JW
5938
5939 zone->totalreserve_pages = max;
5940
cb45b0e9
HA
5941 reserve_pages += max;
5942 }
5943 }
5944 totalreserve_pages = reserve_pages;
5945}
5946
1da177e4
LT
5947/*
5948 * setup_per_zone_lowmem_reserve - called whenever
34b10060 5949 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
1da177e4
LT
5950 * has a correct pages reserved value, so an adequate number of
5951 * pages are left in the zone after a successful __alloc_pages().
5952 */
5953static void setup_per_zone_lowmem_reserve(void)
5954{
5955 struct pglist_data *pgdat;
2f6726e5 5956 enum zone_type j, idx;
1da177e4 5957
ec936fc5 5958 for_each_online_pgdat(pgdat) {
1da177e4
LT
5959 for (j = 0; j < MAX_NR_ZONES; j++) {
5960 struct zone *zone = pgdat->node_zones + j;
b40da049 5961 unsigned long managed_pages = zone->managed_pages;
1da177e4
LT
5962
5963 zone->lowmem_reserve[j] = 0;
5964
2f6726e5
CL
5965 idx = j;
5966 while (idx) {
1da177e4
LT
5967 struct zone *lower_zone;
5968
2f6726e5
CL
5969 idx--;
5970
1da177e4
LT
5971 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5972 sysctl_lowmem_reserve_ratio[idx] = 1;
5973
5974 lower_zone = pgdat->node_zones + idx;
b40da049 5975 lower_zone->lowmem_reserve[j] = managed_pages /
1da177e4 5976 sysctl_lowmem_reserve_ratio[idx];
b40da049 5977 managed_pages += lower_zone->managed_pages;
1da177e4
LT
5978 }
5979 }
5980 }
cb45b0e9
HA
5981
5982 /* update totalreserve_pages */
5983 calculate_totalreserve_pages();
1da177e4
LT
5984}
5985
cfd3da1e 5986static void __setup_per_zone_wmarks(void)
1da177e4
LT
5987{
5988 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5989 unsigned long lowmem_pages = 0;
5990 struct zone *zone;
5991 unsigned long flags;
5992
5993 /* Calculate total number of !ZONE_HIGHMEM pages */
5994 for_each_zone(zone) {
5995 if (!is_highmem(zone))
b40da049 5996 lowmem_pages += zone->managed_pages;
1da177e4
LT
5997 }
5998
5999 for_each_zone(zone) {
ac924c60
AM
6000 u64 tmp;
6001
1125b4e3 6002 spin_lock_irqsave(&zone->lock, flags);
b40da049 6003 tmp = (u64)pages_min * zone->managed_pages;
ac924c60 6004 do_div(tmp, lowmem_pages);
1da177e4
LT
6005 if (is_highmem(zone)) {
6006 /*
669ed175
NP
6007 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6008 * need highmem pages, so cap pages_min to a small
6009 * value here.
6010 *
41858966 6011 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
42ff2703 6012 * deltas control asynch page reclaim, and so should
669ed175 6013 * not be capped for highmem.
1da177e4 6014 */
90ae8d67 6015 unsigned long min_pages;
1da177e4 6016
b40da049 6017 min_pages = zone->managed_pages / 1024;
90ae8d67 6018 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
41858966 6019 zone->watermark[WMARK_MIN] = min_pages;
1da177e4 6020 } else {
669ed175
NP
6021 /*
6022 * If it's a lowmem zone, reserve a number of pages
1da177e4
LT
6023 * proportionate to the zone's size.
6024 */
41858966 6025 zone->watermark[WMARK_MIN] = tmp;
1da177e4
LT
6026 }
6027
41858966
MG
6028 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
6029 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
49f223a9 6030
81c0a2bb 6031 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
abe5f972
JW
6032 high_wmark_pages(zone) - low_wmark_pages(zone) -
6033 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
81c0a2bb 6034
1125b4e3 6035 spin_unlock_irqrestore(&zone->lock, flags);
1da177e4 6036 }
cb45b0e9
HA
6037
6038 /* update totalreserve_pages */
6039 calculate_totalreserve_pages();
1da177e4
LT
6040}
6041
cfd3da1e
MG
6042/**
6043 * setup_per_zone_wmarks - called when min_free_kbytes changes
6044 * or when memory is hot-{added|removed}
6045 *
6046 * Ensures that the watermark[min,low,high] values for each zone are set
6047 * correctly with respect to min_free_kbytes.
6048 */
6049void setup_per_zone_wmarks(void)
6050{
6051 mutex_lock(&zonelists_mutex);
6052 __setup_per_zone_wmarks();
6053 mutex_unlock(&zonelists_mutex);
6054}
6055
55a4462a 6056/*
556adecb
RR
6057 * The inactive anon list should be small enough that the VM never has to
6058 * do too much work, but large enough that each inactive page has a chance
6059 * to be referenced again before it is swapped out.
6060 *
6061 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6062 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6063 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6064 * the anonymous pages are kept on the inactive list.
6065 *
6066 * total target max
6067 * memory ratio inactive anon
6068 * -------------------------------------
6069 * 10MB 1 5MB
6070 * 100MB 1 50MB
6071 * 1GB 3 250MB
6072 * 10GB 10 0.9GB
6073 * 100GB 31 3GB
6074 * 1TB 101 10GB
6075 * 10TB 320 32GB
6076 */
1b79acc9 6077static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
556adecb 6078{
96cb4df5 6079 unsigned int gb, ratio;
556adecb 6080
96cb4df5 6081 /* Zone size in gigabytes */
b40da049 6082 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
96cb4df5 6083 if (gb)
556adecb 6084 ratio = int_sqrt(10 * gb);
96cb4df5
MK
6085 else
6086 ratio = 1;
556adecb 6087
96cb4df5
MK
6088 zone->inactive_ratio = ratio;
6089}
556adecb 6090
839a4fcc 6091static void __meminit setup_per_zone_inactive_ratio(void)
96cb4df5
MK
6092{
6093 struct zone *zone;
6094
6095 for_each_zone(zone)
6096 calculate_zone_inactive_ratio(zone);
556adecb
RR
6097}
6098
1da177e4
LT
6099/*
6100 * Initialise min_free_kbytes.
6101 *
6102 * For small machines we want it small (128k min). For large machines
6103 * we want it large (64MB max). But it is not linear, because network
6104 * bandwidth does not increase linearly with machine size. We use
6105 *
b8af2941 6106 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
1da177e4
LT
6107 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6108 *
6109 * which yields
6110 *
6111 * 16MB: 512k
6112 * 32MB: 724k
6113 * 64MB: 1024k
6114 * 128MB: 1448k
6115 * 256MB: 2048k
6116 * 512MB: 2896k
6117 * 1024MB: 4096k
6118 * 2048MB: 5792k
6119 * 4096MB: 8192k
6120 * 8192MB: 11584k
6121 * 16384MB: 16384k
6122 */
1b79acc9 6123int __meminit init_per_zone_wmark_min(void)
1da177e4
LT
6124{
6125 unsigned long lowmem_kbytes;
5f12733e 6126 int new_min_free_kbytes;
1da177e4
LT
6127
6128 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5f12733e
MH
6129 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6130
6131 if (new_min_free_kbytes > user_min_free_kbytes) {
6132 min_free_kbytes = new_min_free_kbytes;
6133 if (min_free_kbytes < 128)
6134 min_free_kbytes = 128;
6135 if (min_free_kbytes > 65536)
6136 min_free_kbytes = 65536;
6137 } else {
6138 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6139 new_min_free_kbytes, user_min_free_kbytes);
6140 }
bc75d33f 6141 setup_per_zone_wmarks();
a6cccdc3 6142 refresh_zone_stat_thresholds();
1da177e4 6143 setup_per_zone_lowmem_reserve();
556adecb 6144 setup_per_zone_inactive_ratio();
1da177e4
LT
6145 return 0;
6146}
bc75d33f 6147module_init(init_per_zone_wmark_min)
1da177e4
LT
6148
6149/*
b8af2941 6150 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
1da177e4
LT
6151 * that we can call two helper functions whenever min_free_kbytes
6152 * changes.
6153 */
cccad5b9 6154int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8d65af78 6155 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 6156{
da8c757b
HP
6157 int rc;
6158
6159 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6160 if (rc)
6161 return rc;
6162
5f12733e
MH
6163 if (write) {
6164 user_min_free_kbytes = min_free_kbytes;
bc75d33f 6165 setup_per_zone_wmarks();
5f12733e 6166 }
1da177e4
LT
6167 return 0;
6168}
6169
9614634f 6170#ifdef CONFIG_NUMA
cccad5b9 6171int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8d65af78 6172 void __user *buffer, size_t *length, loff_t *ppos)
9614634f
CL
6173{
6174 struct zone *zone;
6175 int rc;
6176
8d65af78 6177 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
9614634f
CL
6178 if (rc)
6179 return rc;
6180
6181 for_each_zone(zone)
b40da049 6182 zone->min_unmapped_pages = (zone->managed_pages *
9614634f
CL
6183 sysctl_min_unmapped_ratio) / 100;
6184 return 0;
6185}
0ff38490 6186
cccad5b9 6187int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
8d65af78 6188 void __user *buffer, size_t *length, loff_t *ppos)
0ff38490
CL
6189{
6190 struct zone *zone;
6191 int rc;
6192
8d65af78 6193 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
0ff38490
CL
6194 if (rc)
6195 return rc;
6196
6197 for_each_zone(zone)
b40da049 6198 zone->min_slab_pages = (zone->managed_pages *
0ff38490
CL
6199 sysctl_min_slab_ratio) / 100;
6200 return 0;
6201}
9614634f
CL
6202#endif
6203
1da177e4
LT
6204/*
6205 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6206 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6207 * whenever sysctl_lowmem_reserve_ratio changes.
6208 *
6209 * The reserve ratio obviously has absolutely no relation with the
41858966 6210 * minimum watermarks. The lowmem reserve ratio can only make sense
1da177e4
LT
6211 * if in function of the boot time zone sizes.
6212 */
cccad5b9 6213int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8d65af78 6214 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 6215{
8d65af78 6216 proc_dointvec_minmax(table, write, buffer, length, ppos);
1da177e4
LT
6217 setup_per_zone_lowmem_reserve();
6218 return 0;
6219}
6220
8ad4b1fb
RS
6221/*
6222 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
b8af2941
PK
6223 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6224 * pagelist can have before it gets flushed back to buddy allocator.
8ad4b1fb 6225 */
cccad5b9 6226int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
8d65af78 6227 void __user *buffer, size_t *length, loff_t *ppos)
8ad4b1fb
RS
6228{
6229 struct zone *zone;
7cd2b0a3 6230 int old_percpu_pagelist_fraction;
8ad4b1fb
RS
6231 int ret;
6232
7cd2b0a3
DR
6233 mutex_lock(&pcp_batch_high_lock);
6234 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6235
8d65af78 6236 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7cd2b0a3
DR
6237 if (!write || ret < 0)
6238 goto out;
6239
6240 /* Sanity checking to avoid pcp imbalance */
6241 if (percpu_pagelist_fraction &&
6242 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6243 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6244 ret = -EINVAL;
6245 goto out;
6246 }
6247
6248 /* No change? */
6249 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6250 goto out;
c8e251fa 6251
364df0eb 6252 for_each_populated_zone(zone) {
7cd2b0a3
DR
6253 unsigned int cpu;
6254
22a7f12b 6255 for_each_possible_cpu(cpu)
7cd2b0a3
DR
6256 pageset_set_high_and_batch(zone,
6257 per_cpu_ptr(zone->pageset, cpu));
8ad4b1fb 6258 }
7cd2b0a3 6259out:
c8e251fa 6260 mutex_unlock(&pcp_batch_high_lock);
7cd2b0a3 6261 return ret;
8ad4b1fb
RS
6262}
6263
a9919c79 6264#ifdef CONFIG_NUMA
f034b5d4 6265int hashdist = HASHDIST_DEFAULT;
1da177e4 6266
1da177e4
LT
6267static int __init set_hashdist(char *str)
6268{
6269 if (!str)
6270 return 0;
6271 hashdist = simple_strtoul(str, &str, 0);
6272 return 1;
6273}
6274__setup("hashdist=", set_hashdist);
6275#endif
6276
6277/*
6278 * allocate a large system hash table from bootmem
6279 * - it is assumed that the hash table must contain an exact power-of-2
6280 * quantity of entries
6281 * - limit is the number of hash buckets, not the total allocation size
6282 */
6283void *__init alloc_large_system_hash(const char *tablename,
6284 unsigned long bucketsize,
6285 unsigned long numentries,
6286 int scale,
6287 int flags,
6288 unsigned int *_hash_shift,
6289 unsigned int *_hash_mask,
31fe62b9
TB
6290 unsigned long low_limit,
6291 unsigned long high_limit)
1da177e4 6292{
31fe62b9 6293 unsigned long long max = high_limit;
1da177e4
LT
6294 unsigned long log2qty, size;
6295 void *table = NULL;
6296
6297 /* allow the kernel cmdline to have a say */
6298 if (!numentries) {
6299 /* round applicable memory size up to nearest megabyte */
04903664 6300 numentries = nr_kernel_pages;
a7e83318
JZ
6301
6302 /* It isn't necessary when PAGE_SIZE >= 1MB */
6303 if (PAGE_SHIFT < 20)
6304 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
1da177e4
LT
6305
6306 /* limit to 1 bucket per 2^scale bytes of low memory */
6307 if (scale > PAGE_SHIFT)
6308 numentries >>= (scale - PAGE_SHIFT);
6309 else
6310 numentries <<= (PAGE_SHIFT - scale);
9ab37b8f
PM
6311
6312 /* Make sure we've got at least a 0-order allocation.. */
2c85f51d
JB
6313 if (unlikely(flags & HASH_SMALL)) {
6314 /* Makes no sense without HASH_EARLY */
6315 WARN_ON(!(flags & HASH_EARLY));
6316 if (!(numentries >> *_hash_shift)) {
6317 numentries = 1UL << *_hash_shift;
6318 BUG_ON(!numentries);
6319 }
6320 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
9ab37b8f 6321 numentries = PAGE_SIZE / bucketsize;
1da177e4 6322 }
6e692ed3 6323 numentries = roundup_pow_of_two(numentries);
1da177e4
LT
6324
6325 /* limit allocation size to 1/16 total memory by default */
6326 if (max == 0) {
6327 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6328 do_div(max, bucketsize);
6329 }
074b8517 6330 max = min(max, 0x80000000ULL);
1da177e4 6331
31fe62b9
TB
6332 if (numentries < low_limit)
6333 numentries = low_limit;
1da177e4
LT
6334 if (numentries > max)
6335 numentries = max;
6336
f0d1b0b3 6337 log2qty = ilog2(numentries);
1da177e4
LT
6338
6339 do {
6340 size = bucketsize << log2qty;
6341 if (flags & HASH_EARLY)
6782832e 6342 table = memblock_virt_alloc_nopanic(size, 0);
1da177e4
LT
6343 else if (hashdist)
6344 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6345 else {
1037b83b
ED
6346 /*
6347 * If bucketsize is not a power-of-two, we may free
a1dd268c
MG
6348 * some pages at the end of hash table which
6349 * alloc_pages_exact() automatically does
1037b83b 6350 */
264ef8a9 6351 if (get_order(size) < MAX_ORDER) {
a1dd268c 6352 table = alloc_pages_exact(size, GFP_ATOMIC);
264ef8a9
CM
6353 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6354 }
1da177e4
LT
6355 }
6356 } while (!table && size > PAGE_SIZE && --log2qty);
6357
6358 if (!table)
6359 panic("Failed to allocate %s hash table\n", tablename);
6360
f241e660 6361 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
1da177e4 6362 tablename,
f241e660 6363 (1UL << log2qty),
f0d1b0b3 6364 ilog2(size) - PAGE_SHIFT,
1da177e4
LT
6365 size);
6366
6367 if (_hash_shift)
6368 *_hash_shift = log2qty;
6369 if (_hash_mask)
6370 *_hash_mask = (1 << log2qty) - 1;
6371
6372 return table;
6373}
a117e66e 6374
835c134e
MG
6375/* Return a pointer to the bitmap storing bits affecting a block of pages */
6376static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6377 unsigned long pfn)
6378{
6379#ifdef CONFIG_SPARSEMEM
6380 return __pfn_to_section(pfn)->pageblock_flags;
6381#else
6382 return zone->pageblock_flags;
6383#endif /* CONFIG_SPARSEMEM */
6384}
6385
6386static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6387{
6388#ifdef CONFIG_SPARSEMEM
6389 pfn &= (PAGES_PER_SECTION-1);
d9c23400 6390 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
835c134e 6391#else
c060f943 6392 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
d9c23400 6393 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
835c134e
MG
6394#endif /* CONFIG_SPARSEMEM */
6395}
6396
6397/**
1aab4d77 6398 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
835c134e 6399 * @page: The page within the block of interest
1aab4d77
RD
6400 * @pfn: The target page frame number
6401 * @end_bitidx: The last bit of interest to retrieve
6402 * @mask: mask of bits that the caller is interested in
6403 *
6404 * Return: pageblock_bits flags
835c134e 6405 */
dc4b0caf 6406unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
e58469ba
MG
6407 unsigned long end_bitidx,
6408 unsigned long mask)
835c134e
MG
6409{
6410 struct zone *zone;
6411 unsigned long *bitmap;
dc4b0caf 6412 unsigned long bitidx, word_bitidx;
e58469ba 6413 unsigned long word;
835c134e
MG
6414
6415 zone = page_zone(page);
835c134e
MG
6416 bitmap = get_pageblock_bitmap(zone, pfn);
6417 bitidx = pfn_to_bitidx(zone, pfn);
e58469ba
MG
6418 word_bitidx = bitidx / BITS_PER_LONG;
6419 bitidx &= (BITS_PER_LONG-1);
835c134e 6420
e58469ba
MG
6421 word = bitmap[word_bitidx];
6422 bitidx += end_bitidx;
6423 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
835c134e
MG
6424}
6425
6426/**
dc4b0caf 6427 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
835c134e 6428 * @page: The page within the block of interest
835c134e 6429 * @flags: The flags to set
1aab4d77
RD
6430 * @pfn: The target page frame number
6431 * @end_bitidx: The last bit of interest
6432 * @mask: mask of bits that the caller is interested in
835c134e 6433 */
dc4b0caf
MG
6434void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6435 unsigned long pfn,
e58469ba
MG
6436 unsigned long end_bitidx,
6437 unsigned long mask)
835c134e
MG
6438{
6439 struct zone *zone;
6440 unsigned long *bitmap;
dc4b0caf 6441 unsigned long bitidx, word_bitidx;
e58469ba
MG
6442 unsigned long old_word, word;
6443
6444 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
835c134e
MG
6445
6446 zone = page_zone(page);
835c134e
MG
6447 bitmap = get_pageblock_bitmap(zone, pfn);
6448 bitidx = pfn_to_bitidx(zone, pfn);
e58469ba
MG
6449 word_bitidx = bitidx / BITS_PER_LONG;
6450 bitidx &= (BITS_PER_LONG-1);
6451
309381fe 6452 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
835c134e 6453
e58469ba
MG
6454 bitidx += end_bitidx;
6455 mask <<= (BITS_PER_LONG - bitidx - 1);
6456 flags <<= (BITS_PER_LONG - bitidx - 1);
6457
4db0c3c2 6458 word = READ_ONCE(bitmap[word_bitidx]);
e58469ba
MG
6459 for (;;) {
6460 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6461 if (word == old_word)
6462 break;
6463 word = old_word;
6464 }
835c134e 6465}
a5d76b54
KH
6466
6467/*
80934513
MK
6468 * This function checks whether pageblock includes unmovable pages or not.
6469 * If @count is not zero, it is okay to include less @count unmovable pages
6470 *
b8af2941 6471 * PageLRU check without isolation or lru_lock could race so that
80934513
MK
6472 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6473 * expect this function should be exact.
a5d76b54 6474 */
b023f468
WC
6475bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6476 bool skip_hwpoisoned_pages)
49ac8255
KH
6477{
6478 unsigned long pfn, iter, found;
47118af0
MN
6479 int mt;
6480
49ac8255
KH
6481 /*
6482 * For avoiding noise data, lru_add_drain_all() should be called
80934513 6483 * If ZONE_MOVABLE, the zone never contains unmovable pages
49ac8255
KH
6484 */
6485 if (zone_idx(zone) == ZONE_MOVABLE)
80934513 6486 return false;
47118af0
MN
6487 mt = get_pageblock_migratetype(page);
6488 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
80934513 6489 return false;
49ac8255
KH
6490
6491 pfn = page_to_pfn(page);
6492 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6493 unsigned long check = pfn + iter;
6494
29723fcc 6495 if (!pfn_valid_within(check))
49ac8255 6496 continue;
29723fcc 6497
49ac8255 6498 page = pfn_to_page(check);
c8721bbb
NH
6499
6500 /*
6501 * Hugepages are not in LRU lists, but they're movable.
6502 * We need not scan over tail pages bacause we don't
6503 * handle each tail page individually in migration.
6504 */
6505 if (PageHuge(page)) {
6506 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6507 continue;
6508 }
6509
97d255c8
MK
6510 /*
6511 * We can't use page_count without pin a page
6512 * because another CPU can free compound page.
6513 * This check already skips compound tails of THP
6514 * because their page->_count is zero at all time.
6515 */
6516 if (!atomic_read(&page->_count)) {
49ac8255
KH
6517 if (PageBuddy(page))
6518 iter += (1 << page_order(page)) - 1;
6519 continue;
6520 }
97d255c8 6521
b023f468
WC
6522 /*
6523 * The HWPoisoned page may be not in buddy system, and
6524 * page_count() is not 0.
6525 */
6526 if (skip_hwpoisoned_pages && PageHWPoison(page))
6527 continue;
6528
49ac8255
KH
6529 if (!PageLRU(page))
6530 found++;
6531 /*
6b4f7799
JW
6532 * If there are RECLAIMABLE pages, we need to check
6533 * it. But now, memory offline itself doesn't call
6534 * shrink_node_slabs() and it still to be fixed.
49ac8255
KH
6535 */
6536 /*
6537 * If the page is not RAM, page_count()should be 0.
6538 * we don't need more check. This is an _used_ not-movable page.
6539 *
6540 * The problematic thing here is PG_reserved pages. PG_reserved
6541 * is set to both of a memory hole page and a _used_ kernel
6542 * page at boot.
6543 */
6544 if (found > count)
80934513 6545 return true;
49ac8255 6546 }
80934513 6547 return false;
49ac8255
KH
6548}
6549
6550bool is_pageblock_removable_nolock(struct page *page)
6551{
656a0706
MH
6552 struct zone *zone;
6553 unsigned long pfn;
687875fb
MH
6554
6555 /*
6556 * We have to be careful here because we are iterating over memory
6557 * sections which are not zone aware so we might end up outside of
6558 * the zone but still within the section.
656a0706
MH
6559 * We have to take care about the node as well. If the node is offline
6560 * its NODE_DATA will be NULL - see page_zone.
687875fb 6561 */
656a0706
MH
6562 if (!node_online(page_to_nid(page)))
6563 return false;
6564
6565 zone = page_zone(page);
6566 pfn = page_to_pfn(page);
108bcc96 6567 if (!zone_spans_pfn(zone, pfn))
687875fb
MH
6568 return false;
6569
b023f468 6570 return !has_unmovable_pages(zone, page, 0, true);
a5d76b54 6571}
0c0e6195 6572
041d3a8c
MN
6573#ifdef CONFIG_CMA
6574
6575static unsigned long pfn_max_align_down(unsigned long pfn)
6576{
6577 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6578 pageblock_nr_pages) - 1);
6579}
6580
6581static unsigned long pfn_max_align_up(unsigned long pfn)
6582{
6583 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6584 pageblock_nr_pages));
6585}
6586
041d3a8c 6587/* [start, end) must belong to a single zone. */
bb13ffeb
MG
6588static int __alloc_contig_migrate_range(struct compact_control *cc,
6589 unsigned long start, unsigned long end)
041d3a8c
MN
6590{
6591 /* This function is based on compact_zone() from compaction.c. */
beb51eaa 6592 unsigned long nr_reclaimed;
041d3a8c
MN
6593 unsigned long pfn = start;
6594 unsigned int tries = 0;
6595 int ret = 0;
6596
be49a6e1 6597 migrate_prep();
041d3a8c 6598
bb13ffeb 6599 while (pfn < end || !list_empty(&cc->migratepages)) {
041d3a8c
MN
6600 if (fatal_signal_pending(current)) {
6601 ret = -EINTR;
6602 break;
6603 }
6604
bb13ffeb
MG
6605 if (list_empty(&cc->migratepages)) {
6606 cc->nr_migratepages = 0;
edc2ca61 6607 pfn = isolate_migratepages_range(cc, pfn, end);
041d3a8c
MN
6608 if (!pfn) {
6609 ret = -EINTR;
6610 break;
6611 }
6612 tries = 0;
6613 } else if (++tries == 5) {
6614 ret = ret < 0 ? ret : -EBUSY;
6615 break;
6616 }
6617
beb51eaa
MK
6618 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6619 &cc->migratepages);
6620 cc->nr_migratepages -= nr_reclaimed;
02c6de8d 6621
9c620e2b 6622 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
e0b9daeb 6623 NULL, 0, cc->mode, MR_CMA);
041d3a8c 6624 }
2a6f5124
SP
6625 if (ret < 0) {
6626 putback_movable_pages(&cc->migratepages);
6627 return ret;
6628 }
6629 return 0;
041d3a8c
MN
6630}
6631
6632/**
6633 * alloc_contig_range() -- tries to allocate given range of pages
6634 * @start: start PFN to allocate
6635 * @end: one-past-the-last PFN to allocate
0815f3d8
MN
6636 * @migratetype: migratetype of the underlaying pageblocks (either
6637 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6638 * in range must have the same migratetype and it must
6639 * be either of the two.
041d3a8c
MN
6640 *
6641 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6642 * aligned, however it's the caller's responsibility to guarantee that
6643 * we are the only thread that changes migrate type of pageblocks the
6644 * pages fall in.
6645 *
6646 * The PFN range must belong to a single zone.
6647 *
6648 * Returns zero on success or negative error code. On success all
6649 * pages which PFN is in [start, end) are allocated for the caller and
6650 * need to be freed with free_contig_range().
6651 */
0815f3d8
MN
6652int alloc_contig_range(unsigned long start, unsigned long end,
6653 unsigned migratetype)
041d3a8c 6654{
041d3a8c 6655 unsigned long outer_start, outer_end;
d00181b9
KS
6656 unsigned int order;
6657 int ret = 0;
041d3a8c 6658
bb13ffeb
MG
6659 struct compact_control cc = {
6660 .nr_migratepages = 0,
6661 .order = -1,
6662 .zone = page_zone(pfn_to_page(start)),
e0b9daeb 6663 .mode = MIGRATE_SYNC,
bb13ffeb
MG
6664 .ignore_skip_hint = true,
6665 };
6666 INIT_LIST_HEAD(&cc.migratepages);
6667
041d3a8c
MN
6668 /*
6669 * What we do here is we mark all pageblocks in range as
6670 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6671 * have different sizes, and due to the way page allocator
6672 * work, we align the range to biggest of the two pages so
6673 * that page allocator won't try to merge buddies from
6674 * different pageblocks and change MIGRATE_ISOLATE to some
6675 * other migration type.
6676 *
6677 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6678 * migrate the pages from an unaligned range (ie. pages that
6679 * we are interested in). This will put all the pages in
6680 * range back to page allocator as MIGRATE_ISOLATE.
6681 *
6682 * When this is done, we take the pages in range from page
6683 * allocator removing them from the buddy system. This way
6684 * page allocator will never consider using them.
6685 *
6686 * This lets us mark the pageblocks back as
6687 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6688 * aligned range but not in the unaligned, original range are
6689 * put back to page allocator so that buddy can use them.
6690 */
6691
6692 ret = start_isolate_page_range(pfn_max_align_down(start),
b023f468
WC
6693 pfn_max_align_up(end), migratetype,
6694 false);
041d3a8c 6695 if (ret)
86a595f9 6696 return ret;
041d3a8c 6697
8ef5849f
JK
6698 /*
6699 * In case of -EBUSY, we'd like to know which page causes problem.
6700 * So, just fall through. We will check it in test_pages_isolated().
6701 */
bb13ffeb 6702 ret = __alloc_contig_migrate_range(&cc, start, end);
8ef5849f 6703 if (ret && ret != -EBUSY)
041d3a8c
MN
6704 goto done;
6705
6706 /*
6707 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6708 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6709 * more, all pages in [start, end) are free in page allocator.
6710 * What we are going to do is to allocate all pages from
6711 * [start, end) (that is remove them from page allocator).
6712 *
6713 * The only problem is that pages at the beginning and at the
6714 * end of interesting range may be not aligned with pages that
6715 * page allocator holds, ie. they can be part of higher order
6716 * pages. Because of this, we reserve the bigger range and
6717 * once this is done free the pages we are not interested in.
6718 *
6719 * We don't have to hold zone->lock here because the pages are
6720 * isolated thus they won't get removed from buddy.
6721 */
6722
6723 lru_add_drain_all();
510f5507 6724 drain_all_pages(cc.zone);
041d3a8c
MN
6725
6726 order = 0;
6727 outer_start = start;
6728 while (!PageBuddy(pfn_to_page(outer_start))) {
6729 if (++order >= MAX_ORDER) {
8ef5849f
JK
6730 outer_start = start;
6731 break;
041d3a8c
MN
6732 }
6733 outer_start &= ~0UL << order;
6734 }
6735
8ef5849f
JK
6736 if (outer_start != start) {
6737 order = page_order(pfn_to_page(outer_start));
6738
6739 /*
6740 * outer_start page could be small order buddy page and
6741 * it doesn't include start page. Adjust outer_start
6742 * in this case to report failed page properly
6743 * on tracepoint in test_pages_isolated()
6744 */
6745 if (outer_start + (1UL << order) <= start)
6746 outer_start = start;
6747 }
6748
041d3a8c 6749 /* Make sure the range is really isolated. */
b023f468 6750 if (test_pages_isolated(outer_start, end, false)) {
dae803e1
MN
6751 pr_info("%s: [%lx, %lx) PFNs busy\n",
6752 __func__, outer_start, end);
041d3a8c
MN
6753 ret = -EBUSY;
6754 goto done;
6755 }
6756
49f223a9 6757 /* Grab isolated pages from freelists. */
bb13ffeb 6758 outer_end = isolate_freepages_range(&cc, outer_start, end);
041d3a8c
MN
6759 if (!outer_end) {
6760 ret = -EBUSY;
6761 goto done;
6762 }
6763
6764 /* Free head and tail (if any) */
6765 if (start != outer_start)
6766 free_contig_range(outer_start, start - outer_start);
6767 if (end != outer_end)
6768 free_contig_range(end, outer_end - end);
6769
6770done:
6771 undo_isolate_page_range(pfn_max_align_down(start),
0815f3d8 6772 pfn_max_align_up(end), migratetype);
041d3a8c
MN
6773 return ret;
6774}
6775
6776void free_contig_range(unsigned long pfn, unsigned nr_pages)
6777{
bcc2b02f
MS
6778 unsigned int count = 0;
6779
6780 for (; nr_pages--; pfn++) {
6781 struct page *page = pfn_to_page(pfn);
6782
6783 count += page_count(page) != 1;
6784 __free_page(page);
6785 }
6786 WARN(count != 0, "%d pages are still in use!\n", count);
041d3a8c
MN
6787}
6788#endif
6789
4ed7e022 6790#ifdef CONFIG_MEMORY_HOTPLUG
0a647f38
CS
6791/*
6792 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6793 * page high values need to be recalulated.
6794 */
4ed7e022
JL
6795void __meminit zone_pcp_update(struct zone *zone)
6796{
0a647f38 6797 unsigned cpu;
c8e251fa 6798 mutex_lock(&pcp_batch_high_lock);
0a647f38 6799 for_each_possible_cpu(cpu)
169f6c19
CS
6800 pageset_set_high_and_batch(zone,
6801 per_cpu_ptr(zone->pageset, cpu));
c8e251fa 6802 mutex_unlock(&pcp_batch_high_lock);
4ed7e022
JL
6803}
6804#endif
6805
340175b7
JL
6806void zone_pcp_reset(struct zone *zone)
6807{
6808 unsigned long flags;
5a883813
MK
6809 int cpu;
6810 struct per_cpu_pageset *pset;
340175b7
JL
6811
6812 /* avoid races with drain_pages() */
6813 local_irq_save(flags);
6814 if (zone->pageset != &boot_pageset) {
5a883813
MK
6815 for_each_online_cpu(cpu) {
6816 pset = per_cpu_ptr(zone->pageset, cpu);
6817 drain_zonestat(zone, pset);
6818 }
340175b7
JL
6819 free_percpu(zone->pageset);
6820 zone->pageset = &boot_pageset;
6821 }
6822 local_irq_restore(flags);
6823}
6824
6dcd73d7 6825#ifdef CONFIG_MEMORY_HOTREMOVE
0c0e6195
KH
6826/*
6827 * All pages in the range must be isolated before calling this.
6828 */
6829void
6830__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6831{
6832 struct page *page;
6833 struct zone *zone;
7aeb09f9 6834 unsigned int order, i;
0c0e6195
KH
6835 unsigned long pfn;
6836 unsigned long flags;
6837 /* find the first valid pfn */
6838 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6839 if (pfn_valid(pfn))
6840 break;
6841 if (pfn == end_pfn)
6842 return;
6843 zone = page_zone(pfn_to_page(pfn));
6844 spin_lock_irqsave(&zone->lock, flags);
6845 pfn = start_pfn;
6846 while (pfn < end_pfn) {
6847 if (!pfn_valid(pfn)) {
6848 pfn++;
6849 continue;
6850 }
6851 page = pfn_to_page(pfn);
b023f468
WC
6852 /*
6853 * The HWPoisoned page may be not in buddy system, and
6854 * page_count() is not 0.
6855 */
6856 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6857 pfn++;
6858 SetPageReserved(page);
6859 continue;
6860 }
6861
0c0e6195
KH
6862 BUG_ON(page_count(page));
6863 BUG_ON(!PageBuddy(page));
6864 order = page_order(page);
6865#ifdef CONFIG_DEBUG_VM
6866 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6867 pfn, 1 << order, end_pfn);
6868#endif
6869 list_del(&page->lru);
6870 rmv_page_order(page);
6871 zone->free_area[order].nr_free--;
0c0e6195
KH
6872 for (i = 0; i < (1 << order); i++)
6873 SetPageReserved((page+i));
6874 pfn += (1 << order);
6875 }
6876 spin_unlock_irqrestore(&zone->lock, flags);
6877}
6878#endif
8d22ba1b
WF
6879
6880#ifdef CONFIG_MEMORY_FAILURE
6881bool is_free_buddy_page(struct page *page)
6882{
6883 struct zone *zone = page_zone(page);
6884 unsigned long pfn = page_to_pfn(page);
6885 unsigned long flags;
7aeb09f9 6886 unsigned int order;
8d22ba1b
WF
6887
6888 spin_lock_irqsave(&zone->lock, flags);
6889 for (order = 0; order < MAX_ORDER; order++) {
6890 struct page *page_head = page - (pfn & ((1 << order) - 1));
6891
6892 if (PageBuddy(page_head) && page_order(page_head) >= order)
6893 break;
6894 }
6895 spin_unlock_irqrestore(&zone->lock, flags);
6896
6897 return order < MAX_ORDER;
6898}
6899#endif