]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - mm/page_alloc.c
mm/page_alloc.c: silence build_all_zonelists() section mismatch
[mirror_ubuntu-artful-kernel.git] / mm / page_alloc.c
CommitLineData
1da177e4
LT
1/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
1da177e4
LT
17#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
10ed273f 22#include <linux/jiffies.h>
1da177e4 23#include <linux/bootmem.h>
edbe7d23 24#include <linux/memblock.h>
1da177e4 25#include <linux/compiler.h>
9f158333 26#include <linux/kernel.h>
b1eeab67 27#include <linux/kmemcheck.h>
1da177e4
LT
28#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
5a3135c2 33#include <linux/oom.h>
1da177e4
LT
34#include <linux/notifier.h>
35#include <linux/topology.h>
36#include <linux/sysctl.h>
37#include <linux/cpu.h>
38#include <linux/cpuset.h>
bdc8cb98 39#include <linux/memory_hotplug.h>
1da177e4
LT
40#include <linux/nodemask.h>
41#include <linux/vmalloc.h>
4be38e35 42#include <linux/mempolicy.h>
6811378e 43#include <linux/stop_machine.h>
c713216d
MG
44#include <linux/sort.h>
45#include <linux/pfn.h>
3fcfab16 46#include <linux/backing-dev.h>
933e312e 47#include <linux/fault-inject.h>
a5d76b54 48#include <linux/page-isolation.h>
52d4b9ac 49#include <linux/page_cgroup.h>
3ac7fe5a 50#include <linux/debugobjects.h>
dbb1f81c 51#include <linux/kmemleak.h>
925cc71e 52#include <linux/memory.h>
56de7263 53#include <linux/compaction.h>
0d3d062a 54#include <trace/events/kmem.h>
718a3821 55#include <linux/ftrace_event.h>
f212ad7c 56#include <linux/memcontrol.h>
1da177e4
LT
57
58#include <asm/tlbflush.h>
ac924c60 59#include <asm/div64.h>
1da177e4
LT
60#include "internal.h"
61
72812019
LS
62#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
63DEFINE_PER_CPU(int, numa_node);
64EXPORT_PER_CPU_SYMBOL(numa_node);
65#endif
66
7aac7898
LS
67#ifdef CONFIG_HAVE_MEMORYLESS_NODES
68/*
69 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
70 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
71 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
72 * defined in <linux/topology.h>.
73 */
74DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
75EXPORT_PER_CPU_SYMBOL(_numa_mem_);
76#endif
77
1da177e4 78/*
13808910 79 * Array of node states.
1da177e4 80 */
13808910
CL
81nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
82 [N_POSSIBLE] = NODE_MASK_ALL,
83 [N_ONLINE] = { { [0] = 1UL } },
84#ifndef CONFIG_NUMA
85 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
86#ifdef CONFIG_HIGHMEM
87 [N_HIGH_MEMORY] = { { [0] = 1UL } },
88#endif
89 [N_CPU] = { { [0] = 1UL } },
90#endif /* NUMA */
91};
92EXPORT_SYMBOL(node_states);
93
6c231b7b 94unsigned long totalram_pages __read_mostly;
cb45b0e9 95unsigned long totalreserve_pages __read_mostly;
8ad4b1fb 96int percpu_pagelist_fraction;
dcce284a 97gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
1da177e4 98
452aa699
RW
99#ifdef CONFIG_PM_SLEEP
100/*
101 * The following functions are used by the suspend/hibernate code to temporarily
102 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
103 * while devices are suspended. To avoid races with the suspend/hibernate code,
104 * they should always be called with pm_mutex held (gfp_allowed_mask also should
105 * only be modified with pm_mutex held, unless the suspend/hibernate code is
106 * guaranteed not to run in parallel with that modification).
107 */
c9e664f1
RW
108
109static gfp_t saved_gfp_mask;
110
111void pm_restore_gfp_mask(void)
452aa699
RW
112{
113 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
114 if (saved_gfp_mask) {
115 gfp_allowed_mask = saved_gfp_mask;
116 saved_gfp_mask = 0;
117 }
452aa699
RW
118}
119
c9e664f1 120void pm_restrict_gfp_mask(void)
452aa699 121{
452aa699 122 WARN_ON(!mutex_is_locked(&pm_mutex));
c9e664f1
RW
123 WARN_ON(saved_gfp_mask);
124 saved_gfp_mask = gfp_allowed_mask;
125 gfp_allowed_mask &= ~GFP_IOFS;
452aa699
RW
126}
127#endif /* CONFIG_PM_SLEEP */
128
d9c23400
MG
129#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
130int pageblock_order __read_mostly;
131#endif
132
d98c7a09 133static void __free_pages_ok(struct page *page, unsigned int order);
a226f6c8 134
1da177e4
LT
135/*
136 * results with 256, 32 in the lowmem_reserve sysctl:
137 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
138 * 1G machine -> (16M dma, 784M normal, 224M high)
139 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
140 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
141 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
a2f1b424
AK
142 *
143 * TBD: should special case ZONE_DMA32 machines here - in those we normally
144 * don't need any ZONE_NORMAL reservation
1da177e4 145 */
2f1b6248 146int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
4b51d669 147#ifdef CONFIG_ZONE_DMA
2f1b6248 148 256,
4b51d669 149#endif
fb0e7942 150#ifdef CONFIG_ZONE_DMA32
2f1b6248 151 256,
fb0e7942 152#endif
e53ef38d 153#ifdef CONFIG_HIGHMEM
2a1e274a 154 32,
e53ef38d 155#endif
2a1e274a 156 32,
2f1b6248 157};
1da177e4
LT
158
159EXPORT_SYMBOL(totalram_pages);
1da177e4 160
15ad7cdc 161static char * const zone_names[MAX_NR_ZONES] = {
4b51d669 162#ifdef CONFIG_ZONE_DMA
2f1b6248 163 "DMA",
4b51d669 164#endif
fb0e7942 165#ifdef CONFIG_ZONE_DMA32
2f1b6248 166 "DMA32",
fb0e7942 167#endif
2f1b6248 168 "Normal",
e53ef38d 169#ifdef CONFIG_HIGHMEM
2a1e274a 170 "HighMem",
e53ef38d 171#endif
2a1e274a 172 "Movable",
2f1b6248
CL
173};
174
1da177e4
LT
175int min_free_kbytes = 1024;
176
2c85f51d
JB
177static unsigned long __meminitdata nr_kernel_pages;
178static unsigned long __meminitdata nr_all_pages;
a3142c8e 179static unsigned long __meminitdata dma_reserve;
1da177e4 180
c713216d
MG
181#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
182 /*
183ff22b 183 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
c713216d
MG
184 * ranges of memory (RAM) that may be registered with add_active_range().
185 * Ranges passed to add_active_range() will be merged if possible
186 * so the number of times add_active_range() can be called is
187 * related to the number of nodes and the number of holes
188 */
189 #ifdef CONFIG_MAX_ACTIVE_REGIONS
190 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
191 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
192 #else
193 #if MAX_NUMNODES >= 32
194 /* If there can be many nodes, allow up to 50 holes per node */
195 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
196 #else
197 /* By default, allow up to 256 distinct regions */
198 #define MAX_ACTIVE_REGIONS 256
199 #endif
200 #endif
201
98011f56
JB
202 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
203 static int __meminitdata nr_nodemap_entries;
204 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
205 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
b69a7288 206 static unsigned long __initdata required_kernelcore;
484f51f8 207 static unsigned long __initdata required_movablecore;
b69a7288 208 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
2a1e274a
MG
209
210 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
211 int movable_zone;
212 EXPORT_SYMBOL(movable_zone);
c713216d
MG
213#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
214
418508c1
MS
215#if MAX_NUMNODES > 1
216int nr_node_ids __read_mostly = MAX_NUMNODES;
62bc62a8 217int nr_online_nodes __read_mostly = 1;
418508c1 218EXPORT_SYMBOL(nr_node_ids);
62bc62a8 219EXPORT_SYMBOL(nr_online_nodes);
418508c1
MS
220#endif
221
9ef9acb0
MG
222int page_group_by_mobility_disabled __read_mostly;
223
b2a0ac88
MG
224static void set_pageblock_migratetype(struct page *page, int migratetype)
225{
49255c61
MG
226
227 if (unlikely(page_group_by_mobility_disabled))
228 migratetype = MIGRATE_UNMOVABLE;
229
b2a0ac88
MG
230 set_pageblock_flags_group(page, (unsigned long)migratetype,
231 PB_migrate, PB_migrate_end);
232}
233
7f33d49a
RW
234bool oom_killer_disabled __read_mostly;
235
13e7444b 236#ifdef CONFIG_DEBUG_VM
c6a57e19 237static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
1da177e4 238{
bdc8cb98
DH
239 int ret = 0;
240 unsigned seq;
241 unsigned long pfn = page_to_pfn(page);
c6a57e19 242
bdc8cb98
DH
243 do {
244 seq = zone_span_seqbegin(zone);
245 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
246 ret = 1;
247 else if (pfn < zone->zone_start_pfn)
248 ret = 1;
249 } while (zone_span_seqretry(zone, seq));
250
251 return ret;
c6a57e19
DH
252}
253
254static int page_is_consistent(struct zone *zone, struct page *page)
255{
14e07298 256 if (!pfn_valid_within(page_to_pfn(page)))
c6a57e19 257 return 0;
1da177e4 258 if (zone != page_zone(page))
c6a57e19
DH
259 return 0;
260
261 return 1;
262}
263/*
264 * Temporary debugging check for pages not lying within a given zone.
265 */
266static int bad_range(struct zone *zone, struct page *page)
267{
268 if (page_outside_zone_boundaries(zone, page))
1da177e4 269 return 1;
c6a57e19
DH
270 if (!page_is_consistent(zone, page))
271 return 1;
272
1da177e4
LT
273 return 0;
274}
13e7444b
NP
275#else
276static inline int bad_range(struct zone *zone, struct page *page)
277{
278 return 0;
279}
280#endif
281
224abf92 282static void bad_page(struct page *page)
1da177e4 283{
d936cf9b
HD
284 static unsigned long resume;
285 static unsigned long nr_shown;
286 static unsigned long nr_unshown;
287
2a7684a2
WF
288 /* Don't complain about poisoned pages */
289 if (PageHWPoison(page)) {
ef2b4b95 290 reset_page_mapcount(page); /* remove PageBuddy */
2a7684a2
WF
291 return;
292 }
293
d936cf9b
HD
294 /*
295 * Allow a burst of 60 reports, then keep quiet for that minute;
296 * or allow a steady drip of one report per second.
297 */
298 if (nr_shown == 60) {
299 if (time_before(jiffies, resume)) {
300 nr_unshown++;
301 goto out;
302 }
303 if (nr_unshown) {
1e9e6365
HD
304 printk(KERN_ALERT
305 "BUG: Bad page state: %lu messages suppressed\n",
d936cf9b
HD
306 nr_unshown);
307 nr_unshown = 0;
308 }
309 nr_shown = 0;
310 }
311 if (nr_shown++ == 0)
312 resume = jiffies + 60 * HZ;
313
1e9e6365 314 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
3dc14741 315 current->comm, page_to_pfn(page));
718a3821 316 dump_page(page);
3dc14741 317
1da177e4 318 dump_stack();
d936cf9b 319out:
8cc3b392 320 /* Leave bad fields for debug, except PageBuddy could make trouble */
ef2b4b95 321 reset_page_mapcount(page); /* remove PageBuddy */
9f158333 322 add_taint(TAINT_BAD_PAGE);
1da177e4
LT
323}
324
1da177e4
LT
325/*
326 * Higher-order pages are called "compound pages". They are structured thusly:
327 *
328 * The first PAGE_SIZE page is called the "head page".
329 *
330 * The remaining PAGE_SIZE pages are called "tail pages".
331 *
332 * All pages have PG_compound set. All pages have their ->private pointing at
333 * the head page (even the head page has this).
334 *
41d78ba5
HD
335 * The first tail page's ->lru.next holds the address of the compound page's
336 * put_page() function. Its ->lru.prev holds the order of allocation.
337 * This usage means that zero-order pages may not be compound.
1da177e4 338 */
d98c7a09
HD
339
340static void free_compound_page(struct page *page)
341{
d85f3385 342 __free_pages_ok(page, compound_order(page));
d98c7a09
HD
343}
344
01ad1c08 345void prep_compound_page(struct page *page, unsigned long order)
18229df5
AW
346{
347 int i;
348 int nr_pages = 1 << order;
349
350 set_compound_page_dtor(page, free_compound_page);
351 set_compound_order(page, order);
352 __SetPageHead(page);
353 for (i = 1; i < nr_pages; i++) {
354 struct page *p = page + i;
355
356 __SetPageTail(p);
357 p->first_page = page;
358 }
359}
360
59ff4216 361/* update __split_huge_page_refcount if you change this function */
8cc3b392 362static int destroy_compound_page(struct page *page, unsigned long order)
1da177e4
LT
363{
364 int i;
365 int nr_pages = 1 << order;
8cc3b392 366 int bad = 0;
1da177e4 367
8cc3b392
HD
368 if (unlikely(compound_order(page) != order) ||
369 unlikely(!PageHead(page))) {
224abf92 370 bad_page(page);
8cc3b392
HD
371 bad++;
372 }
1da177e4 373
6d777953 374 __ClearPageHead(page);
8cc3b392 375
18229df5
AW
376 for (i = 1; i < nr_pages; i++) {
377 struct page *p = page + i;
1da177e4 378
e713a21d 379 if (unlikely(!PageTail(p) || (p->first_page != page))) {
224abf92 380 bad_page(page);
8cc3b392
HD
381 bad++;
382 }
d85f3385 383 __ClearPageTail(p);
1da177e4 384 }
8cc3b392
HD
385
386 return bad;
1da177e4 387}
1da177e4 388
17cf4406
NP
389static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
390{
391 int i;
392
6626c5d5
AM
393 /*
394 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
395 * and __GFP_HIGHMEM from hard or soft interrupt context.
396 */
725d704e 397 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
17cf4406
NP
398 for (i = 0; i < (1 << order); i++)
399 clear_highpage(page + i);
400}
401
6aa3001b
AM
402static inline void set_page_order(struct page *page, int order)
403{
4c21e2f2 404 set_page_private(page, order);
676165a8 405 __SetPageBuddy(page);
1da177e4
LT
406}
407
408static inline void rmv_page_order(struct page *page)
409{
676165a8 410 __ClearPageBuddy(page);
4c21e2f2 411 set_page_private(page, 0);
1da177e4
LT
412}
413
414/*
415 * Locate the struct page for both the matching buddy in our
416 * pair (buddy1) and the combined O(n+1) page they form (page).
417 *
418 * 1) Any buddy B1 will have an order O twin B2 which satisfies
419 * the following equation:
420 * B2 = B1 ^ (1 << O)
421 * For example, if the starting buddy (buddy2) is #8 its order
422 * 1 buddy is #10:
423 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
424 *
425 * 2) Any buddy B will have an order O+1 parent P which
426 * satisfies the following equation:
427 * P = B & ~(1 << O)
428 *
d6e05edc 429 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
1da177e4 430 */
1da177e4 431static inline unsigned long
43506fad 432__find_buddy_index(unsigned long page_idx, unsigned int order)
1da177e4 433{
43506fad 434 return page_idx ^ (1 << order);
1da177e4
LT
435}
436
437/*
438 * This function checks whether a page is free && is the buddy
439 * we can do coalesce a page and its buddy if
13e7444b 440 * (a) the buddy is not in a hole &&
676165a8 441 * (b) the buddy is in the buddy system &&
cb2b95e1
AW
442 * (c) a page and its buddy have the same order &&
443 * (d) a page and its buddy are in the same zone.
676165a8 444 *
5f24ce5f
AA
445 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
446 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
1da177e4 447 *
676165a8 448 * For recording page's order, we use page_private(page).
1da177e4 449 */
cb2b95e1
AW
450static inline int page_is_buddy(struct page *page, struct page *buddy,
451 int order)
1da177e4 452{
14e07298 453 if (!pfn_valid_within(page_to_pfn(buddy)))
13e7444b 454 return 0;
13e7444b 455
cb2b95e1
AW
456 if (page_zone_id(page) != page_zone_id(buddy))
457 return 0;
458
459 if (PageBuddy(buddy) && page_order(buddy) == order) {
a3af9c38 460 VM_BUG_ON(page_count(buddy) != 0);
6aa3001b 461 return 1;
676165a8 462 }
6aa3001b 463 return 0;
1da177e4
LT
464}
465
466/*
467 * Freeing function for a buddy system allocator.
468 *
469 * The concept of a buddy system is to maintain direct-mapped table
470 * (containing bit values) for memory blocks of various "orders".
471 * The bottom level table contains the map for the smallest allocatable
472 * units of memory (here, pages), and each level above it describes
473 * pairs of units from the levels below, hence, "buddies".
474 * At a high level, all that happens here is marking the table entry
475 * at the bottom level available, and propagating the changes upward
476 * as necessary, plus some accounting needed to play nicely with other
477 * parts of the VM system.
478 * At each level, we keep a list of pages, which are heads of continuous
5f24ce5f 479 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
4c21e2f2 480 * order is recorded in page_private(page) field.
1da177e4
LT
481 * So when we are allocating or freeing one, we can derive the state of the
482 * other. That is, if we allocate a small block, and both were
483 * free, the remainder of the region must be split into blocks.
484 * If a block is freed, and its buddy is also free, then this
485 * triggers coalescing into a block of larger size.
486 *
487 * -- wli
488 */
489
48db57f8 490static inline void __free_one_page(struct page *page,
ed0ae21d
MG
491 struct zone *zone, unsigned int order,
492 int migratetype)
1da177e4
LT
493{
494 unsigned long page_idx;
6dda9d55 495 unsigned long combined_idx;
43506fad 496 unsigned long uninitialized_var(buddy_idx);
6dda9d55 497 struct page *buddy;
1da177e4 498
224abf92 499 if (unlikely(PageCompound(page)))
8cc3b392
HD
500 if (unlikely(destroy_compound_page(page, order)))
501 return;
1da177e4 502
ed0ae21d
MG
503 VM_BUG_ON(migratetype == -1);
504
1da177e4
LT
505 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
506
f2260e6b 507 VM_BUG_ON(page_idx & ((1 << order) - 1));
725d704e 508 VM_BUG_ON(bad_range(zone, page));
1da177e4 509
1da177e4 510 while (order < MAX_ORDER-1) {
43506fad
KC
511 buddy_idx = __find_buddy_index(page_idx, order);
512 buddy = page + (buddy_idx - page_idx);
cb2b95e1 513 if (!page_is_buddy(page, buddy, order))
3c82d0ce 514 break;
13e7444b 515
3c82d0ce 516 /* Our buddy is free, merge with it and move up one order. */
1da177e4 517 list_del(&buddy->lru);
b2a0ac88 518 zone->free_area[order].nr_free--;
1da177e4 519 rmv_page_order(buddy);
43506fad 520 combined_idx = buddy_idx & page_idx;
1da177e4
LT
521 page = page + (combined_idx - page_idx);
522 page_idx = combined_idx;
523 order++;
524 }
525 set_page_order(page, order);
6dda9d55
CZ
526
527 /*
528 * If this is not the largest possible page, check if the buddy
529 * of the next-highest order is free. If it is, it's possible
530 * that pages are being freed that will coalesce soon. In case,
531 * that is happening, add the free page to the tail of the list
532 * so it's less likely to be used soon and more likely to be merged
533 * as a higher order page
534 */
b7f50cfa 535 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
6dda9d55 536 struct page *higher_page, *higher_buddy;
43506fad
KC
537 combined_idx = buddy_idx & page_idx;
538 higher_page = page + (combined_idx - page_idx);
539 buddy_idx = __find_buddy_index(combined_idx, order + 1);
540 higher_buddy = page + (buddy_idx - combined_idx);
6dda9d55
CZ
541 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
542 list_add_tail(&page->lru,
543 &zone->free_area[order].free_list[migratetype]);
544 goto out;
545 }
546 }
547
548 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
549out:
1da177e4
LT
550 zone->free_area[order].nr_free++;
551}
552
092cead6
KM
553/*
554 * free_page_mlock() -- clean up attempts to free and mlocked() page.
555 * Page should not be on lru, so no need to fix that up.
556 * free_pages_check() will verify...
557 */
558static inline void free_page_mlock(struct page *page)
559{
092cead6
KM
560 __dec_zone_page_state(page, NR_MLOCK);
561 __count_vm_event(UNEVICTABLE_MLOCKFREED);
562}
092cead6 563
224abf92 564static inline int free_pages_check(struct page *page)
1da177e4 565{
92be2e33
NP
566 if (unlikely(page_mapcount(page) |
567 (page->mapping != NULL) |
a3af9c38 568 (atomic_read(&page->_count) != 0) |
f212ad7c
DN
569 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
570 (mem_cgroup_bad_page_check(page)))) {
224abf92 571 bad_page(page);
79f4b7bf 572 return 1;
8cc3b392 573 }
79f4b7bf
HD
574 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
575 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
576 return 0;
1da177e4
LT
577}
578
579/*
5f8dcc21 580 * Frees a number of pages from the PCP lists
1da177e4 581 * Assumes all pages on list are in same zone, and of same order.
207f36ee 582 * count is the number of pages to free.
1da177e4
LT
583 *
584 * If the zone was previously in an "all pages pinned" state then look to
585 * see if this freeing clears that state.
586 *
587 * And clear the zone's pages_scanned counter, to hold off the "all pages are
588 * pinned" detection logic.
589 */
5f8dcc21
MG
590static void free_pcppages_bulk(struct zone *zone, int count,
591 struct per_cpu_pages *pcp)
1da177e4 592{
5f8dcc21 593 int migratetype = 0;
a6f9edd6 594 int batch_free = 0;
72853e29 595 int to_free = count;
5f8dcc21 596
c54ad30c 597 spin_lock(&zone->lock);
93e4a89a 598 zone->all_unreclaimable = 0;
1da177e4 599 zone->pages_scanned = 0;
f2260e6b 600
72853e29 601 while (to_free) {
48db57f8 602 struct page *page;
5f8dcc21
MG
603 struct list_head *list;
604
605 /*
a6f9edd6
MG
606 * Remove pages from lists in a round-robin fashion. A
607 * batch_free count is maintained that is incremented when an
608 * empty list is encountered. This is so more pages are freed
609 * off fuller lists instead of spinning excessively around empty
610 * lists
5f8dcc21
MG
611 */
612 do {
a6f9edd6 613 batch_free++;
5f8dcc21
MG
614 if (++migratetype == MIGRATE_PCPTYPES)
615 migratetype = 0;
616 list = &pcp->lists[migratetype];
617 } while (list_empty(list));
48db57f8 618
1d16871d
NK
619 /* This is the only non-empty list. Free them all. */
620 if (batch_free == MIGRATE_PCPTYPES)
621 batch_free = to_free;
622
a6f9edd6
MG
623 do {
624 page = list_entry(list->prev, struct page, lru);
625 /* must delete as __free_one_page list manipulates */
626 list_del(&page->lru);
a7016235
HD
627 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
628 __free_one_page(page, zone, 0, page_private(page));
629 trace_mm_page_pcpu_drain(page, 0, page_private(page));
72853e29 630 } while (--to_free && --batch_free && !list_empty(list));
1da177e4 631 }
72853e29 632 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
c54ad30c 633 spin_unlock(&zone->lock);
1da177e4
LT
634}
635
ed0ae21d
MG
636static void free_one_page(struct zone *zone, struct page *page, int order,
637 int migratetype)
1da177e4 638{
006d22d9 639 spin_lock(&zone->lock);
93e4a89a 640 zone->all_unreclaimable = 0;
006d22d9 641 zone->pages_scanned = 0;
f2260e6b 642
ed0ae21d 643 __free_one_page(page, zone, order, migratetype);
72853e29 644 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
006d22d9 645 spin_unlock(&zone->lock);
48db57f8
NP
646}
647
ec95f53a 648static bool free_pages_prepare(struct page *page, unsigned int order)
48db57f8 649{
1da177e4 650 int i;
8cc3b392 651 int bad = 0;
1da177e4 652
f650316c 653 trace_mm_page_free_direct(page, order);
b1eeab67
VN
654 kmemcheck_free_shadow(page, order);
655
8dd60a3a
AA
656 if (PageAnon(page))
657 page->mapping = NULL;
658 for (i = 0; i < (1 << order); i++)
659 bad += free_pages_check(page + i);
8cc3b392 660 if (bad)
ec95f53a 661 return false;
689bcebf 662
3ac7fe5a 663 if (!PageHighMem(page)) {
9858db50 664 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
3ac7fe5a
TG
665 debug_check_no_obj_freed(page_address(page),
666 PAGE_SIZE << order);
667 }
dafb1367 668 arch_free_page(page, order);
48db57f8 669 kernel_map_pages(page, 1 << order, 0);
dafb1367 670
ec95f53a
KM
671 return true;
672}
673
674static void __free_pages_ok(struct page *page, unsigned int order)
675{
676 unsigned long flags;
677 int wasMlocked = __TestClearPageMlocked(page);
678
679 if (!free_pages_prepare(page, order))
680 return;
681
c54ad30c 682 local_irq_save(flags);
c277331d 683 if (unlikely(wasMlocked))
da456f14 684 free_page_mlock(page);
f8891e5e 685 __count_vm_events(PGFREE, 1 << order);
ed0ae21d
MG
686 free_one_page(page_zone(page), page, order,
687 get_pageblock_migratetype(page));
c54ad30c 688 local_irq_restore(flags);
1da177e4
LT
689}
690
a226f6c8
DH
691/*
692 * permit the bootmem allocator to evade page validation on high-order frees
693 */
af370fb8 694void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
a226f6c8
DH
695{
696 if (order == 0) {
697 __ClearPageReserved(page);
698 set_page_count(page, 0);
7835e98b 699 set_page_refcounted(page);
545b1ea9 700 __free_page(page);
a226f6c8 701 } else {
a226f6c8
DH
702 int loop;
703
545b1ea9 704 prefetchw(page);
a226f6c8
DH
705 for (loop = 0; loop < BITS_PER_LONG; loop++) {
706 struct page *p = &page[loop];
707
545b1ea9
NP
708 if (loop + 1 < BITS_PER_LONG)
709 prefetchw(p + 1);
a226f6c8
DH
710 __ClearPageReserved(p);
711 set_page_count(p, 0);
712 }
713
7835e98b 714 set_page_refcounted(page);
545b1ea9 715 __free_pages(page, order);
a226f6c8
DH
716 }
717}
718
1da177e4
LT
719
720/*
721 * The order of subdivision here is critical for the IO subsystem.
722 * Please do not alter this order without good reasons and regression
723 * testing. Specifically, as large blocks of memory are subdivided,
724 * the order in which smaller blocks are delivered depends on the order
725 * they're subdivided in this function. This is the primary factor
726 * influencing the order in which pages are delivered to the IO
727 * subsystem according to empirical testing, and this is also justified
728 * by considering the behavior of a buddy system containing a single
729 * large block of memory acted on by a series of small allocations.
730 * This behavior is a critical factor in sglist merging's success.
731 *
732 * -- wli
733 */
085cc7d5 734static inline void expand(struct zone *zone, struct page *page,
b2a0ac88
MG
735 int low, int high, struct free_area *area,
736 int migratetype)
1da177e4
LT
737{
738 unsigned long size = 1 << high;
739
740 while (high > low) {
741 area--;
742 high--;
743 size >>= 1;
725d704e 744 VM_BUG_ON(bad_range(zone, &page[size]));
b2a0ac88 745 list_add(&page[size].lru, &area->free_list[migratetype]);
1da177e4
LT
746 area->nr_free++;
747 set_page_order(&page[size], high);
748 }
1da177e4
LT
749}
750
1da177e4
LT
751/*
752 * This page is about to be returned from the page allocator
753 */
2a7684a2 754static inline int check_new_page(struct page *page)
1da177e4 755{
92be2e33
NP
756 if (unlikely(page_mapcount(page) |
757 (page->mapping != NULL) |
a3af9c38 758 (atomic_read(&page->_count) != 0) |
f212ad7c
DN
759 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
760 (mem_cgroup_bad_page_check(page)))) {
224abf92 761 bad_page(page);
689bcebf 762 return 1;
8cc3b392 763 }
2a7684a2
WF
764 return 0;
765}
766
767static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
768{
769 int i;
770
771 for (i = 0; i < (1 << order); i++) {
772 struct page *p = page + i;
773 if (unlikely(check_new_page(p)))
774 return 1;
775 }
689bcebf 776
4c21e2f2 777 set_page_private(page, 0);
7835e98b 778 set_page_refcounted(page);
cc102509
NP
779
780 arch_alloc_page(page, order);
1da177e4 781 kernel_map_pages(page, 1 << order, 1);
17cf4406
NP
782
783 if (gfp_flags & __GFP_ZERO)
784 prep_zero_page(page, order, gfp_flags);
785
786 if (order && (gfp_flags & __GFP_COMP))
787 prep_compound_page(page, order);
788
689bcebf 789 return 0;
1da177e4
LT
790}
791
56fd56b8
MG
792/*
793 * Go through the free lists for the given migratetype and remove
794 * the smallest available page from the freelists
795 */
728ec980
MG
796static inline
797struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
56fd56b8
MG
798 int migratetype)
799{
800 unsigned int current_order;
801 struct free_area * area;
802 struct page *page;
803
804 /* Find a page of the appropriate size in the preferred list */
805 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
806 area = &(zone->free_area[current_order]);
807 if (list_empty(&area->free_list[migratetype]))
808 continue;
809
810 page = list_entry(area->free_list[migratetype].next,
811 struct page, lru);
812 list_del(&page->lru);
813 rmv_page_order(page);
814 area->nr_free--;
56fd56b8
MG
815 expand(zone, page, order, current_order, area, migratetype);
816 return page;
817 }
818
819 return NULL;
820}
821
822
b2a0ac88
MG
823/*
824 * This array describes the order lists are fallen back to when
825 * the free lists for the desirable migrate type are depleted
826 */
827static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
64c5e135
MG
828 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
829 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
830 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
831 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
b2a0ac88
MG
832};
833
c361be55
MG
834/*
835 * Move the free pages in a range to the free lists of the requested type.
d9c23400 836 * Note that start_page and end_pages are not aligned on a pageblock
c361be55
MG
837 * boundary. If alignment is required, use move_freepages_block()
838 */
b69a7288
AB
839static int move_freepages(struct zone *zone,
840 struct page *start_page, struct page *end_page,
841 int migratetype)
c361be55
MG
842{
843 struct page *page;
844 unsigned long order;
d100313f 845 int pages_moved = 0;
c361be55
MG
846
847#ifndef CONFIG_HOLES_IN_ZONE
848 /*
849 * page_zone is not safe to call in this context when
850 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
851 * anyway as we check zone boundaries in move_freepages_block().
852 * Remove at a later date when no bug reports exist related to
ac0e5b7a 853 * grouping pages by mobility
c361be55
MG
854 */
855 BUG_ON(page_zone(start_page) != page_zone(end_page));
856#endif
857
858 for (page = start_page; page <= end_page;) {
344c790e
AL
859 /* Make sure we are not inadvertently changing nodes */
860 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
861
c361be55
MG
862 if (!pfn_valid_within(page_to_pfn(page))) {
863 page++;
864 continue;
865 }
866
867 if (!PageBuddy(page)) {
868 page++;
869 continue;
870 }
871
872 order = page_order(page);
84be48d8
KS
873 list_move(&page->lru,
874 &zone->free_area[order].free_list[migratetype]);
c361be55 875 page += 1 << order;
d100313f 876 pages_moved += 1 << order;
c361be55
MG
877 }
878
d100313f 879 return pages_moved;
c361be55
MG
880}
881
b69a7288
AB
882static int move_freepages_block(struct zone *zone, struct page *page,
883 int migratetype)
c361be55
MG
884{
885 unsigned long start_pfn, end_pfn;
886 struct page *start_page, *end_page;
887
888 start_pfn = page_to_pfn(page);
d9c23400 889 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
c361be55 890 start_page = pfn_to_page(start_pfn);
d9c23400
MG
891 end_page = start_page + pageblock_nr_pages - 1;
892 end_pfn = start_pfn + pageblock_nr_pages - 1;
c361be55
MG
893
894 /* Do not cross zone boundaries */
895 if (start_pfn < zone->zone_start_pfn)
896 start_page = page;
897 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
898 return 0;
899
900 return move_freepages(zone, start_page, end_page, migratetype);
901}
902
2f66a68f
MG
903static void change_pageblock_range(struct page *pageblock_page,
904 int start_order, int migratetype)
905{
906 int nr_pageblocks = 1 << (start_order - pageblock_order);
907
908 while (nr_pageblocks--) {
909 set_pageblock_migratetype(pageblock_page, migratetype);
910 pageblock_page += pageblock_nr_pages;
911 }
912}
913
b2a0ac88 914/* Remove an element from the buddy allocator from the fallback list */
0ac3a409
MG
915static inline struct page *
916__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
b2a0ac88
MG
917{
918 struct free_area * area;
919 int current_order;
920 struct page *page;
921 int migratetype, i;
922
923 /* Find the largest possible block of pages in the other list */
924 for (current_order = MAX_ORDER-1; current_order >= order;
925 --current_order) {
926 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
927 migratetype = fallbacks[start_migratetype][i];
928
56fd56b8
MG
929 /* MIGRATE_RESERVE handled later if necessary */
930 if (migratetype == MIGRATE_RESERVE)
931 continue;
e010487d 932
b2a0ac88
MG
933 area = &(zone->free_area[current_order]);
934 if (list_empty(&area->free_list[migratetype]))
935 continue;
936
937 page = list_entry(area->free_list[migratetype].next,
938 struct page, lru);
939 area->nr_free--;
940
941 /*
c361be55 942 * If breaking a large block of pages, move all free
46dafbca
MG
943 * pages to the preferred allocation list. If falling
944 * back for a reclaimable kernel allocation, be more
25985edc 945 * aggressive about taking ownership of free pages
b2a0ac88 946 */
d9c23400 947 if (unlikely(current_order >= (pageblock_order >> 1)) ||
dd5d241e
MG
948 start_migratetype == MIGRATE_RECLAIMABLE ||
949 page_group_by_mobility_disabled) {
46dafbca
MG
950 unsigned long pages;
951 pages = move_freepages_block(zone, page,
952 start_migratetype);
953
954 /* Claim the whole block if over half of it is free */
dd5d241e
MG
955 if (pages >= (1 << (pageblock_order-1)) ||
956 page_group_by_mobility_disabled)
46dafbca
MG
957 set_pageblock_migratetype(page,
958 start_migratetype);
959
b2a0ac88 960 migratetype = start_migratetype;
c361be55 961 }
b2a0ac88
MG
962
963 /* Remove the page from the freelists */
964 list_del(&page->lru);
965 rmv_page_order(page);
b2a0ac88 966
2f66a68f
MG
967 /* Take ownership for orders >= pageblock_order */
968 if (current_order >= pageblock_order)
969 change_pageblock_range(page, current_order,
b2a0ac88
MG
970 start_migratetype);
971
972 expand(zone, page, order, current_order, area, migratetype);
e0fff1bd
MG
973
974 trace_mm_page_alloc_extfrag(page, order, current_order,
975 start_migratetype, migratetype);
976
b2a0ac88
MG
977 return page;
978 }
979 }
980
728ec980 981 return NULL;
b2a0ac88
MG
982}
983
56fd56b8 984/*
1da177e4
LT
985 * Do the hard work of removing an element from the buddy allocator.
986 * Call me with the zone->lock already held.
987 */
b2a0ac88
MG
988static struct page *__rmqueue(struct zone *zone, unsigned int order,
989 int migratetype)
1da177e4 990{
1da177e4
LT
991 struct page *page;
992
728ec980 993retry_reserve:
56fd56b8 994 page = __rmqueue_smallest(zone, order, migratetype);
b2a0ac88 995
728ec980 996 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
56fd56b8 997 page = __rmqueue_fallback(zone, order, migratetype);
b2a0ac88 998
728ec980
MG
999 /*
1000 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1001 * is used because __rmqueue_smallest is an inline function
1002 * and we want just one call site
1003 */
1004 if (!page) {
1005 migratetype = MIGRATE_RESERVE;
1006 goto retry_reserve;
1007 }
1008 }
1009
0d3d062a 1010 trace_mm_page_alloc_zone_locked(page, order, migratetype);
b2a0ac88 1011 return page;
1da177e4
LT
1012}
1013
1014/*
1015 * Obtain a specified number of elements from the buddy allocator, all under
1016 * a single hold of the lock, for efficiency. Add them to the supplied list.
1017 * Returns the number of new pages which were placed at *list.
1018 */
1019static int rmqueue_bulk(struct zone *zone, unsigned int order,
b2a0ac88 1020 unsigned long count, struct list_head *list,
e084b2d9 1021 int migratetype, int cold)
1da177e4 1022{
1da177e4 1023 int i;
1da177e4 1024
c54ad30c 1025 spin_lock(&zone->lock);
1da177e4 1026 for (i = 0; i < count; ++i) {
b2a0ac88 1027 struct page *page = __rmqueue(zone, order, migratetype);
085cc7d5 1028 if (unlikely(page == NULL))
1da177e4 1029 break;
81eabcbe
MG
1030
1031 /*
1032 * Split buddy pages returned by expand() are received here
1033 * in physical page order. The page is added to the callers and
1034 * list and the list head then moves forward. From the callers
1035 * perspective, the linked list is ordered by page number in
1036 * some conditions. This is useful for IO devices that can
1037 * merge IO requests if the physical pages are ordered
1038 * properly.
1039 */
e084b2d9
MG
1040 if (likely(cold == 0))
1041 list_add(&page->lru, list);
1042 else
1043 list_add_tail(&page->lru, list);
535131e6 1044 set_page_private(page, migratetype);
81eabcbe 1045 list = &page->lru;
1da177e4 1046 }
f2260e6b 1047 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
c54ad30c 1048 spin_unlock(&zone->lock);
085cc7d5 1049 return i;
1da177e4
LT
1050}
1051
4ae7c039 1052#ifdef CONFIG_NUMA
8fce4d8e 1053/*
4037d452
CL
1054 * Called from the vmstat counter updater to drain pagesets of this
1055 * currently executing processor on remote nodes after they have
1056 * expired.
1057 *
879336c3
CL
1058 * Note that this function must be called with the thread pinned to
1059 * a single processor.
8fce4d8e 1060 */
4037d452 1061void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
4ae7c039 1062{
4ae7c039 1063 unsigned long flags;
4037d452 1064 int to_drain;
4ae7c039 1065
4037d452
CL
1066 local_irq_save(flags);
1067 if (pcp->count >= pcp->batch)
1068 to_drain = pcp->batch;
1069 else
1070 to_drain = pcp->count;
5f8dcc21 1071 free_pcppages_bulk(zone, to_drain, pcp);
4037d452
CL
1072 pcp->count -= to_drain;
1073 local_irq_restore(flags);
4ae7c039
CL
1074}
1075#endif
1076
9f8f2172
CL
1077/*
1078 * Drain pages of the indicated processor.
1079 *
1080 * The processor must either be the current processor and the
1081 * thread pinned to the current processor or a processor that
1082 * is not online.
1083 */
1084static void drain_pages(unsigned int cpu)
1da177e4 1085{
c54ad30c 1086 unsigned long flags;
1da177e4 1087 struct zone *zone;
1da177e4 1088
ee99c71c 1089 for_each_populated_zone(zone) {
1da177e4 1090 struct per_cpu_pageset *pset;
3dfa5721 1091 struct per_cpu_pages *pcp;
1da177e4 1092
99dcc3e5
CL
1093 local_irq_save(flags);
1094 pset = per_cpu_ptr(zone->pageset, cpu);
3dfa5721
CL
1095
1096 pcp = &pset->pcp;
2ff754fa
DR
1097 if (pcp->count) {
1098 free_pcppages_bulk(zone, pcp->count, pcp);
1099 pcp->count = 0;
1100 }
3dfa5721 1101 local_irq_restore(flags);
1da177e4
LT
1102 }
1103}
1da177e4 1104
9f8f2172
CL
1105/*
1106 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1107 */
1108void drain_local_pages(void *arg)
1109{
1110 drain_pages(smp_processor_id());
1111}
1112
1113/*
1114 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1115 */
1116void drain_all_pages(void)
1117{
15c8b6c1 1118 on_each_cpu(drain_local_pages, NULL, 1);
9f8f2172
CL
1119}
1120
296699de 1121#ifdef CONFIG_HIBERNATION
1da177e4
LT
1122
1123void mark_free_pages(struct zone *zone)
1124{
f623f0db
RW
1125 unsigned long pfn, max_zone_pfn;
1126 unsigned long flags;
b2a0ac88 1127 int order, t;
1da177e4
LT
1128 struct list_head *curr;
1129
1130 if (!zone->spanned_pages)
1131 return;
1132
1133 spin_lock_irqsave(&zone->lock, flags);
f623f0db
RW
1134
1135 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1136 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1137 if (pfn_valid(pfn)) {
1138 struct page *page = pfn_to_page(pfn);
1139
7be98234
RW
1140 if (!swsusp_page_is_forbidden(page))
1141 swsusp_unset_page_free(page);
f623f0db 1142 }
1da177e4 1143
b2a0ac88
MG
1144 for_each_migratetype_order(order, t) {
1145 list_for_each(curr, &zone->free_area[order].free_list[t]) {
f623f0db 1146 unsigned long i;
1da177e4 1147
f623f0db
RW
1148 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1149 for (i = 0; i < (1UL << order); i++)
7be98234 1150 swsusp_set_page_free(pfn_to_page(pfn + i));
f623f0db 1151 }
b2a0ac88 1152 }
1da177e4
LT
1153 spin_unlock_irqrestore(&zone->lock, flags);
1154}
e2c55dc8 1155#endif /* CONFIG_PM */
1da177e4 1156
1da177e4
LT
1157/*
1158 * Free a 0-order page
fc91668e 1159 * cold == 1 ? free a cold page : free a hot page
1da177e4 1160 */
fc91668e 1161void free_hot_cold_page(struct page *page, int cold)
1da177e4
LT
1162{
1163 struct zone *zone = page_zone(page);
1164 struct per_cpu_pages *pcp;
1165 unsigned long flags;
5f8dcc21 1166 int migratetype;
451ea25d 1167 int wasMlocked = __TestClearPageMlocked(page);
1da177e4 1168
ec95f53a 1169 if (!free_pages_prepare(page, 0))
689bcebf
HD
1170 return;
1171
5f8dcc21
MG
1172 migratetype = get_pageblock_migratetype(page);
1173 set_page_private(page, migratetype);
1da177e4 1174 local_irq_save(flags);
c277331d 1175 if (unlikely(wasMlocked))
da456f14 1176 free_page_mlock(page);
f8891e5e 1177 __count_vm_event(PGFREE);
da456f14 1178
5f8dcc21
MG
1179 /*
1180 * We only track unmovable, reclaimable and movable on pcp lists.
1181 * Free ISOLATE pages back to the allocator because they are being
1182 * offlined but treat RESERVE as movable pages so we can get those
1183 * areas back if necessary. Otherwise, we may have to free
1184 * excessively into the page allocator
1185 */
1186 if (migratetype >= MIGRATE_PCPTYPES) {
1187 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1188 free_one_page(zone, page, 0, migratetype);
1189 goto out;
1190 }
1191 migratetype = MIGRATE_MOVABLE;
1192 }
1193
99dcc3e5 1194 pcp = &this_cpu_ptr(zone->pageset)->pcp;
3dfa5721 1195 if (cold)
5f8dcc21 1196 list_add_tail(&page->lru, &pcp->lists[migratetype]);
3dfa5721 1197 else
5f8dcc21 1198 list_add(&page->lru, &pcp->lists[migratetype]);
1da177e4 1199 pcp->count++;
48db57f8 1200 if (pcp->count >= pcp->high) {
5f8dcc21 1201 free_pcppages_bulk(zone, pcp->batch, pcp);
48db57f8
NP
1202 pcp->count -= pcp->batch;
1203 }
5f8dcc21
MG
1204
1205out:
1da177e4 1206 local_irq_restore(flags);
1da177e4
LT
1207}
1208
8dfcc9ba
NP
1209/*
1210 * split_page takes a non-compound higher-order page, and splits it into
1211 * n (1<<order) sub-pages: page[0..n]
1212 * Each sub-page must be freed individually.
1213 *
1214 * Note: this is probably too low level an operation for use in drivers.
1215 * Please consult with lkml before using this in your driver.
1216 */
1217void split_page(struct page *page, unsigned int order)
1218{
1219 int i;
1220
725d704e
NP
1221 VM_BUG_ON(PageCompound(page));
1222 VM_BUG_ON(!page_count(page));
b1eeab67
VN
1223
1224#ifdef CONFIG_KMEMCHECK
1225 /*
1226 * Split shadow pages too, because free(page[0]) would
1227 * otherwise free the whole shadow.
1228 */
1229 if (kmemcheck_page_is_tracked(page))
1230 split_page(virt_to_page(page[0].shadow), order);
1231#endif
1232
7835e98b
NP
1233 for (i = 1; i < (1 << order); i++)
1234 set_page_refcounted(page + i);
8dfcc9ba 1235}
8dfcc9ba 1236
748446bb
MG
1237/*
1238 * Similar to split_page except the page is already free. As this is only
1239 * being used for migration, the migratetype of the block also changes.
1240 * As this is called with interrupts disabled, the caller is responsible
1241 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1242 * are enabled.
1243 *
1244 * Note: this is probably too low level an operation for use in drivers.
1245 * Please consult with lkml before using this in your driver.
1246 */
1247int split_free_page(struct page *page)
1248{
1249 unsigned int order;
1250 unsigned long watermark;
1251 struct zone *zone;
1252
1253 BUG_ON(!PageBuddy(page));
1254
1255 zone = page_zone(page);
1256 order = page_order(page);
1257
1258 /* Obey watermarks as if the page was being allocated */
1259 watermark = low_wmark_pages(zone) + (1 << order);
1260 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1261 return 0;
1262
1263 /* Remove page from free list */
1264 list_del(&page->lru);
1265 zone->free_area[order].nr_free--;
1266 rmv_page_order(page);
1267 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1268
1269 /* Split into individual pages */
1270 set_page_refcounted(page);
1271 split_page(page, order);
1272
1273 if (order >= pageblock_order - 1) {
1274 struct page *endpage = page + (1 << order) - 1;
1275 for (; page < endpage; page += pageblock_nr_pages)
1276 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1277 }
1278
1279 return 1 << order;
1280}
1281
1da177e4
LT
1282/*
1283 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1284 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1285 * or two.
1286 */
0a15c3e9
MG
1287static inline
1288struct page *buffered_rmqueue(struct zone *preferred_zone,
3dd28266
MG
1289 struct zone *zone, int order, gfp_t gfp_flags,
1290 int migratetype)
1da177e4
LT
1291{
1292 unsigned long flags;
689bcebf 1293 struct page *page;
1da177e4
LT
1294 int cold = !!(gfp_flags & __GFP_COLD);
1295
689bcebf 1296again:
48db57f8 1297 if (likely(order == 0)) {
1da177e4 1298 struct per_cpu_pages *pcp;
5f8dcc21 1299 struct list_head *list;
1da177e4 1300
1da177e4 1301 local_irq_save(flags);
99dcc3e5
CL
1302 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1303 list = &pcp->lists[migratetype];
5f8dcc21 1304 if (list_empty(list)) {
535131e6 1305 pcp->count += rmqueue_bulk(zone, 0,
5f8dcc21 1306 pcp->batch, list,
e084b2d9 1307 migratetype, cold);
5f8dcc21 1308 if (unlikely(list_empty(list)))
6fb332fa 1309 goto failed;
535131e6 1310 }
b92a6edd 1311
5f8dcc21
MG
1312 if (cold)
1313 page = list_entry(list->prev, struct page, lru);
1314 else
1315 page = list_entry(list->next, struct page, lru);
1316
b92a6edd
MG
1317 list_del(&page->lru);
1318 pcp->count--;
7fb1d9fc 1319 } else {
dab48dab
AM
1320 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1321 /*
1322 * __GFP_NOFAIL is not to be used in new code.
1323 *
1324 * All __GFP_NOFAIL callers should be fixed so that they
1325 * properly detect and handle allocation failures.
1326 *
1327 * We most definitely don't want callers attempting to
4923abf9 1328 * allocate greater than order-1 page units with
dab48dab
AM
1329 * __GFP_NOFAIL.
1330 */
4923abf9 1331 WARN_ON_ONCE(order > 1);
dab48dab 1332 }
1da177e4 1333 spin_lock_irqsave(&zone->lock, flags);
b2a0ac88 1334 page = __rmqueue(zone, order, migratetype);
a74609fa
NP
1335 spin_unlock(&zone->lock);
1336 if (!page)
1337 goto failed;
6ccf80eb 1338 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1da177e4
LT
1339 }
1340
f8891e5e 1341 __count_zone_vm_events(PGALLOC, zone, 1 << order);
78afd561 1342 zone_statistics(preferred_zone, zone, gfp_flags);
a74609fa 1343 local_irq_restore(flags);
1da177e4 1344
725d704e 1345 VM_BUG_ON(bad_range(zone, page));
17cf4406 1346 if (prep_new_page(page, order, gfp_flags))
a74609fa 1347 goto again;
1da177e4 1348 return page;
a74609fa
NP
1349
1350failed:
1351 local_irq_restore(flags);
a74609fa 1352 return NULL;
1da177e4
LT
1353}
1354
41858966
MG
1355/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1356#define ALLOC_WMARK_MIN WMARK_MIN
1357#define ALLOC_WMARK_LOW WMARK_LOW
1358#define ALLOC_WMARK_HIGH WMARK_HIGH
1359#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1360
1361/* Mask to get the watermark bits */
1362#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1363
3148890b
NP
1364#define ALLOC_HARDER 0x10 /* try to alloc harder */
1365#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1366#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
7fb1d9fc 1367
933e312e
AM
1368#ifdef CONFIG_FAIL_PAGE_ALLOC
1369
1370static struct fail_page_alloc_attr {
1371 struct fault_attr attr;
1372
1373 u32 ignore_gfp_highmem;
1374 u32 ignore_gfp_wait;
54114994 1375 u32 min_order;
933e312e
AM
1376
1377#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1378
1379 struct dentry *ignore_gfp_highmem_file;
1380 struct dentry *ignore_gfp_wait_file;
54114994 1381 struct dentry *min_order_file;
933e312e
AM
1382
1383#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1384
1385} fail_page_alloc = {
1386 .attr = FAULT_ATTR_INITIALIZER,
6b1b60f4
DM
1387 .ignore_gfp_wait = 1,
1388 .ignore_gfp_highmem = 1,
54114994 1389 .min_order = 1,
933e312e
AM
1390};
1391
1392static int __init setup_fail_page_alloc(char *str)
1393{
1394 return setup_fault_attr(&fail_page_alloc.attr, str);
1395}
1396__setup("fail_page_alloc=", setup_fail_page_alloc);
1397
1398static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1399{
54114994
AM
1400 if (order < fail_page_alloc.min_order)
1401 return 0;
933e312e
AM
1402 if (gfp_mask & __GFP_NOFAIL)
1403 return 0;
1404 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1405 return 0;
1406 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1407 return 0;
1408
1409 return should_fail(&fail_page_alloc.attr, 1 << order);
1410}
1411
1412#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1413
1414static int __init fail_page_alloc_debugfs(void)
1415{
1416 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1417 struct dentry *dir;
1418 int err;
1419
1420 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1421 "fail_page_alloc");
1422 if (err)
1423 return err;
1424 dir = fail_page_alloc.attr.dentries.dir;
1425
1426 fail_page_alloc.ignore_gfp_wait_file =
1427 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1428 &fail_page_alloc.ignore_gfp_wait);
1429
1430 fail_page_alloc.ignore_gfp_highmem_file =
1431 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1432 &fail_page_alloc.ignore_gfp_highmem);
54114994
AM
1433 fail_page_alloc.min_order_file =
1434 debugfs_create_u32("min-order", mode, dir,
1435 &fail_page_alloc.min_order);
933e312e
AM
1436
1437 if (!fail_page_alloc.ignore_gfp_wait_file ||
54114994
AM
1438 !fail_page_alloc.ignore_gfp_highmem_file ||
1439 !fail_page_alloc.min_order_file) {
933e312e
AM
1440 err = -ENOMEM;
1441 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1442 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
54114994 1443 debugfs_remove(fail_page_alloc.min_order_file);
933e312e
AM
1444 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1445 }
1446
1447 return err;
1448}
1449
1450late_initcall(fail_page_alloc_debugfs);
1451
1452#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1453
1454#else /* CONFIG_FAIL_PAGE_ALLOC */
1455
1456static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1457{
1458 return 0;
1459}
1460
1461#endif /* CONFIG_FAIL_PAGE_ALLOC */
1462
1da177e4 1463/*
88f5acf8 1464 * Return true if free pages are above 'mark'. This takes into account the order
1da177e4
LT
1465 * of the allocation.
1466 */
88f5acf8
MG
1467static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1468 int classzone_idx, int alloc_flags, long free_pages)
1da177e4
LT
1469{
1470 /* free_pages my go negative - that's OK */
d23ad423 1471 long min = mark;
1da177e4
LT
1472 int o;
1473
88f5acf8 1474 free_pages -= (1 << order) + 1;
7fb1d9fc 1475 if (alloc_flags & ALLOC_HIGH)
1da177e4 1476 min -= min / 2;
7fb1d9fc 1477 if (alloc_flags & ALLOC_HARDER)
1da177e4
LT
1478 min -= min / 4;
1479
1480 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
88f5acf8 1481 return false;
1da177e4
LT
1482 for (o = 0; o < order; o++) {
1483 /* At the next order, this order's pages become unavailable */
1484 free_pages -= z->free_area[o].nr_free << o;
1485
1486 /* Require fewer higher order pages to be free */
1487 min >>= 1;
1488
1489 if (free_pages <= min)
88f5acf8 1490 return false;
1da177e4 1491 }
88f5acf8
MG
1492 return true;
1493}
1494
1495bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1496 int classzone_idx, int alloc_flags)
1497{
1498 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1499 zone_page_state(z, NR_FREE_PAGES));
1500}
1501
1502bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1503 int classzone_idx, int alloc_flags)
1504{
1505 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1506
1507 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1508 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1509
1510 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1511 free_pages);
1da177e4
LT
1512}
1513
9276b1bc
PJ
1514#ifdef CONFIG_NUMA
1515/*
1516 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1517 * skip over zones that are not allowed by the cpuset, or that have
1518 * been recently (in last second) found to be nearly full. See further
1519 * comments in mmzone.h. Reduces cache footprint of zonelist scans
183ff22b 1520 * that have to skip over a lot of full or unallowed zones.
9276b1bc
PJ
1521 *
1522 * If the zonelist cache is present in the passed in zonelist, then
1523 * returns a pointer to the allowed node mask (either the current
37b07e41 1524 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
9276b1bc
PJ
1525 *
1526 * If the zonelist cache is not available for this zonelist, does
1527 * nothing and returns NULL.
1528 *
1529 * If the fullzones BITMAP in the zonelist cache is stale (more than
1530 * a second since last zap'd) then we zap it out (clear its bits.)
1531 *
1532 * We hold off even calling zlc_setup, until after we've checked the
1533 * first zone in the zonelist, on the theory that most allocations will
1534 * be satisfied from that first zone, so best to examine that zone as
1535 * quickly as we can.
1536 */
1537static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1538{
1539 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1540 nodemask_t *allowednodes; /* zonelist_cache approximation */
1541
1542 zlc = zonelist->zlcache_ptr;
1543 if (!zlc)
1544 return NULL;
1545
f05111f5 1546 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
9276b1bc
PJ
1547 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1548 zlc->last_full_zap = jiffies;
1549 }
1550
1551 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1552 &cpuset_current_mems_allowed :
37b07e41 1553 &node_states[N_HIGH_MEMORY];
9276b1bc
PJ
1554 return allowednodes;
1555}
1556
1557/*
1558 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1559 * if it is worth looking at further for free memory:
1560 * 1) Check that the zone isn't thought to be full (doesn't have its
1561 * bit set in the zonelist_cache fullzones BITMAP).
1562 * 2) Check that the zones node (obtained from the zonelist_cache
1563 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1564 * Return true (non-zero) if zone is worth looking at further, or
1565 * else return false (zero) if it is not.
1566 *
1567 * This check -ignores- the distinction between various watermarks,
1568 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1569 * found to be full for any variation of these watermarks, it will
1570 * be considered full for up to one second by all requests, unless
1571 * we are so low on memory on all allowed nodes that we are forced
1572 * into the second scan of the zonelist.
1573 *
1574 * In the second scan we ignore this zonelist cache and exactly
1575 * apply the watermarks to all zones, even it is slower to do so.
1576 * We are low on memory in the second scan, and should leave no stone
1577 * unturned looking for a free page.
1578 */
dd1a239f 1579static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
9276b1bc
PJ
1580 nodemask_t *allowednodes)
1581{
1582 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1583 int i; /* index of *z in zonelist zones */
1584 int n; /* node that zone *z is on */
1585
1586 zlc = zonelist->zlcache_ptr;
1587 if (!zlc)
1588 return 1;
1589
dd1a239f 1590 i = z - zonelist->_zonerefs;
9276b1bc
PJ
1591 n = zlc->z_to_n[i];
1592
1593 /* This zone is worth trying if it is allowed but not full */
1594 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1595}
1596
1597/*
1598 * Given 'z' scanning a zonelist, set the corresponding bit in
1599 * zlc->fullzones, so that subsequent attempts to allocate a page
1600 * from that zone don't waste time re-examining it.
1601 */
dd1a239f 1602static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
9276b1bc
PJ
1603{
1604 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1605 int i; /* index of *z in zonelist zones */
1606
1607 zlc = zonelist->zlcache_ptr;
1608 if (!zlc)
1609 return;
1610
dd1a239f 1611 i = z - zonelist->_zonerefs;
9276b1bc
PJ
1612
1613 set_bit(i, zlc->fullzones);
1614}
1615
1616#else /* CONFIG_NUMA */
1617
1618static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1619{
1620 return NULL;
1621}
1622
dd1a239f 1623static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
9276b1bc
PJ
1624 nodemask_t *allowednodes)
1625{
1626 return 1;
1627}
1628
dd1a239f 1629static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
9276b1bc
PJ
1630{
1631}
1632#endif /* CONFIG_NUMA */
1633
7fb1d9fc 1634/*
0798e519 1635 * get_page_from_freelist goes through the zonelist trying to allocate
7fb1d9fc
RS
1636 * a page.
1637 */
1638static struct page *
19770b32 1639get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
5117f45d 1640 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
3dd28266 1641 struct zone *preferred_zone, int migratetype)
753ee728 1642{
dd1a239f 1643 struct zoneref *z;
7fb1d9fc 1644 struct page *page = NULL;
54a6eb5c 1645 int classzone_idx;
5117f45d 1646 struct zone *zone;
9276b1bc
PJ
1647 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1648 int zlc_active = 0; /* set if using zonelist_cache */
1649 int did_zlc_setup = 0; /* just call zlc_setup() one time */
54a6eb5c 1650
19770b32 1651 classzone_idx = zone_idx(preferred_zone);
9276b1bc 1652zonelist_scan:
7fb1d9fc 1653 /*
9276b1bc 1654 * Scan zonelist, looking for a zone with enough free.
7fb1d9fc
RS
1655 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1656 */
19770b32
MG
1657 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1658 high_zoneidx, nodemask) {
9276b1bc
PJ
1659 if (NUMA_BUILD && zlc_active &&
1660 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1661 continue;
7fb1d9fc 1662 if ((alloc_flags & ALLOC_CPUSET) &&
02a0e53d 1663 !cpuset_zone_allowed_softwall(zone, gfp_mask))
9276b1bc 1664 goto try_next_zone;
7fb1d9fc 1665
41858966 1666 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
7fb1d9fc 1667 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
3148890b 1668 unsigned long mark;
fa5e084e
MG
1669 int ret;
1670
41858966 1671 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
fa5e084e
MG
1672 if (zone_watermark_ok(zone, order, mark,
1673 classzone_idx, alloc_flags))
1674 goto try_this_zone;
1675
1676 if (zone_reclaim_mode == 0)
1677 goto this_zone_full;
1678
1679 ret = zone_reclaim(zone, gfp_mask, order);
1680 switch (ret) {
1681 case ZONE_RECLAIM_NOSCAN:
1682 /* did not scan */
1683 goto try_next_zone;
1684 case ZONE_RECLAIM_FULL:
1685 /* scanned but unreclaimable */
1686 goto this_zone_full;
1687 default:
1688 /* did we reclaim enough */
1689 if (!zone_watermark_ok(zone, order, mark,
1690 classzone_idx, alloc_flags))
9276b1bc 1691 goto this_zone_full;
0798e519 1692 }
7fb1d9fc
RS
1693 }
1694
fa5e084e 1695try_this_zone:
3dd28266
MG
1696 page = buffered_rmqueue(preferred_zone, zone, order,
1697 gfp_mask, migratetype);
0798e519 1698 if (page)
7fb1d9fc 1699 break;
9276b1bc
PJ
1700this_zone_full:
1701 if (NUMA_BUILD)
1702 zlc_mark_zone_full(zonelist, z);
1703try_next_zone:
62bc62a8 1704 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
d395b734
MG
1705 /*
1706 * we do zlc_setup after the first zone is tried but only
1707 * if there are multiple nodes make it worthwhile
1708 */
9276b1bc
PJ
1709 allowednodes = zlc_setup(zonelist, alloc_flags);
1710 zlc_active = 1;
1711 did_zlc_setup = 1;
1712 }
54a6eb5c 1713 }
9276b1bc
PJ
1714
1715 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1716 /* Disable zlc cache for second zonelist scan */
1717 zlc_active = 0;
1718 goto zonelist_scan;
1719 }
7fb1d9fc 1720 return page;
753ee728
MH
1721}
1722
29423e77
DR
1723/*
1724 * Large machines with many possible nodes should not always dump per-node
1725 * meminfo in irq context.
1726 */
1727static inline bool should_suppress_show_mem(void)
1728{
1729 bool ret = false;
1730
1731#if NODES_SHIFT > 8
1732 ret = in_interrupt();
1733#endif
1734 return ret;
1735}
1736
11e33f6a
MG
1737static inline int
1738should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1739 unsigned long pages_reclaimed)
1da177e4 1740{
11e33f6a
MG
1741 /* Do not loop if specifically requested */
1742 if (gfp_mask & __GFP_NORETRY)
1743 return 0;
1da177e4 1744
11e33f6a
MG
1745 /*
1746 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1747 * means __GFP_NOFAIL, but that may not be true in other
1748 * implementations.
1749 */
1750 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1751 return 1;
1752
1753 /*
1754 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1755 * specified, then we retry until we no longer reclaim any pages
1756 * (above), or we've reclaimed an order of pages at least as
1757 * large as the allocation's order. In both cases, if the
1758 * allocation still fails, we stop retrying.
1759 */
1760 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1761 return 1;
cf40bd16 1762
11e33f6a
MG
1763 /*
1764 * Don't let big-order allocations loop unless the caller
1765 * explicitly requests that.
1766 */
1767 if (gfp_mask & __GFP_NOFAIL)
1768 return 1;
1da177e4 1769
11e33f6a
MG
1770 return 0;
1771}
933e312e 1772
11e33f6a
MG
1773static inline struct page *
1774__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1775 struct zonelist *zonelist, enum zone_type high_zoneidx,
3dd28266
MG
1776 nodemask_t *nodemask, struct zone *preferred_zone,
1777 int migratetype)
11e33f6a
MG
1778{
1779 struct page *page;
1780
1781 /* Acquire the OOM killer lock for the zones in zonelist */
ff321fea 1782 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
11e33f6a 1783 schedule_timeout_uninterruptible(1);
1da177e4
LT
1784 return NULL;
1785 }
6b1de916 1786
11e33f6a
MG
1787 /*
1788 * Go through the zonelist yet one more time, keep very high watermark
1789 * here, this is only to catch a parallel oom killing, we must fail if
1790 * we're still under heavy pressure.
1791 */
1792 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1793 order, zonelist, high_zoneidx,
5117f45d 1794 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
3dd28266 1795 preferred_zone, migratetype);
7fb1d9fc 1796 if (page)
11e33f6a
MG
1797 goto out;
1798
4365a567
KH
1799 if (!(gfp_mask & __GFP_NOFAIL)) {
1800 /* The OOM killer will not help higher order allocs */
1801 if (order > PAGE_ALLOC_COSTLY_ORDER)
1802 goto out;
03668b3c
DR
1803 /* The OOM killer does not needlessly kill tasks for lowmem */
1804 if (high_zoneidx < ZONE_NORMAL)
1805 goto out;
4365a567
KH
1806 /*
1807 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1808 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1809 * The caller should handle page allocation failure by itself if
1810 * it specifies __GFP_THISNODE.
1811 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1812 */
1813 if (gfp_mask & __GFP_THISNODE)
1814 goto out;
1815 }
11e33f6a 1816 /* Exhausted what can be done so it's blamo time */
4365a567 1817 out_of_memory(zonelist, gfp_mask, order, nodemask);
11e33f6a
MG
1818
1819out:
1820 clear_zonelist_oom(zonelist, gfp_mask);
1821 return page;
1822}
1823
56de7263
MG
1824#ifdef CONFIG_COMPACTION
1825/* Try memory compaction for high-order allocations before reclaim */
1826static struct page *
1827__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1828 struct zonelist *zonelist, enum zone_type high_zoneidx,
1829 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
77f1fe6b
MG
1830 int migratetype, unsigned long *did_some_progress,
1831 bool sync_migration)
56de7263
MG
1832{
1833 struct page *page;
1834
4f92e258 1835 if (!order || compaction_deferred(preferred_zone))
56de7263
MG
1836 return NULL;
1837
c06b1fca 1838 current->flags |= PF_MEMALLOC;
56de7263 1839 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
77f1fe6b 1840 nodemask, sync_migration);
c06b1fca 1841 current->flags &= ~PF_MEMALLOC;
56de7263
MG
1842 if (*did_some_progress != COMPACT_SKIPPED) {
1843
1844 /* Page migration frees to the PCP lists but we want merging */
1845 drain_pages(get_cpu());
1846 put_cpu();
1847
1848 page = get_page_from_freelist(gfp_mask, nodemask,
1849 order, zonelist, high_zoneidx,
1850 alloc_flags, preferred_zone,
1851 migratetype);
1852 if (page) {
4f92e258
MG
1853 preferred_zone->compact_considered = 0;
1854 preferred_zone->compact_defer_shift = 0;
56de7263
MG
1855 count_vm_event(COMPACTSUCCESS);
1856 return page;
1857 }
1858
1859 /*
1860 * It's bad if compaction run occurs and fails.
1861 * The most likely reason is that pages exist,
1862 * but not enough to satisfy watermarks.
1863 */
1864 count_vm_event(COMPACTFAIL);
4f92e258 1865 defer_compaction(preferred_zone);
56de7263
MG
1866
1867 cond_resched();
1868 }
1869
1870 return NULL;
1871}
1872#else
1873static inline struct page *
1874__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1875 struct zonelist *zonelist, enum zone_type high_zoneidx,
1876 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
77f1fe6b
MG
1877 int migratetype, unsigned long *did_some_progress,
1878 bool sync_migration)
56de7263
MG
1879{
1880 return NULL;
1881}
1882#endif /* CONFIG_COMPACTION */
1883
11e33f6a
MG
1884/* The really slow allocator path where we enter direct reclaim */
1885static inline struct page *
1886__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1887 struct zonelist *zonelist, enum zone_type high_zoneidx,
5117f45d 1888 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
3dd28266 1889 int migratetype, unsigned long *did_some_progress)
11e33f6a
MG
1890{
1891 struct page *page = NULL;
1892 struct reclaim_state reclaim_state;
9ee493ce 1893 bool drained = false;
11e33f6a
MG
1894
1895 cond_resched();
1896
1897 /* We now go into synchronous reclaim */
1898 cpuset_memory_pressure_bump();
c06b1fca 1899 current->flags |= PF_MEMALLOC;
11e33f6a
MG
1900 lockdep_set_current_reclaim_state(gfp_mask);
1901 reclaim_state.reclaimed_slab = 0;
c06b1fca 1902 current->reclaim_state = &reclaim_state;
11e33f6a
MG
1903
1904 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1905
c06b1fca 1906 current->reclaim_state = NULL;
11e33f6a 1907 lockdep_clear_current_reclaim_state();
c06b1fca 1908 current->flags &= ~PF_MEMALLOC;
11e33f6a
MG
1909
1910 cond_resched();
1911
9ee493ce
MG
1912 if (unlikely(!(*did_some_progress)))
1913 return NULL;
11e33f6a 1914
9ee493ce
MG
1915retry:
1916 page = get_page_from_freelist(gfp_mask, nodemask, order,
5117f45d 1917 zonelist, high_zoneidx,
3dd28266
MG
1918 alloc_flags, preferred_zone,
1919 migratetype);
9ee493ce
MG
1920
1921 /*
1922 * If an allocation failed after direct reclaim, it could be because
1923 * pages are pinned on the per-cpu lists. Drain them and try again
1924 */
1925 if (!page && !drained) {
1926 drain_all_pages();
1927 drained = true;
1928 goto retry;
1929 }
1930
11e33f6a
MG
1931 return page;
1932}
1933
1da177e4 1934/*
11e33f6a
MG
1935 * This is called in the allocator slow-path if the allocation request is of
1936 * sufficient urgency to ignore watermarks and take other desperate measures
1da177e4 1937 */
11e33f6a
MG
1938static inline struct page *
1939__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1940 struct zonelist *zonelist, enum zone_type high_zoneidx,
3dd28266
MG
1941 nodemask_t *nodemask, struct zone *preferred_zone,
1942 int migratetype)
11e33f6a
MG
1943{
1944 struct page *page;
1945
1946 do {
1947 page = get_page_from_freelist(gfp_mask, nodemask, order,
5117f45d 1948 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
3dd28266 1949 preferred_zone, migratetype);
11e33f6a
MG
1950
1951 if (!page && gfp_mask & __GFP_NOFAIL)
0e093d99 1952 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
11e33f6a
MG
1953 } while (!page && (gfp_mask & __GFP_NOFAIL));
1954
1955 return page;
1956}
1957
1958static inline
1959void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
99504748
MG
1960 enum zone_type high_zoneidx,
1961 enum zone_type classzone_idx)
1da177e4 1962{
dd1a239f
MG
1963 struct zoneref *z;
1964 struct zone *zone;
1da177e4 1965
11e33f6a 1966 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
99504748 1967 wakeup_kswapd(zone, order, classzone_idx);
11e33f6a 1968}
cf40bd16 1969
341ce06f
PZ
1970static inline int
1971gfp_to_alloc_flags(gfp_t gfp_mask)
1972{
341ce06f
PZ
1973 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1974 const gfp_t wait = gfp_mask & __GFP_WAIT;
1da177e4 1975
a56f57ff 1976 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
e6223a3b 1977 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
933e312e 1978
341ce06f
PZ
1979 /*
1980 * The caller may dip into page reserves a bit more if the caller
1981 * cannot run direct reclaim, or if the caller has realtime scheduling
1982 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1983 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1984 */
e6223a3b 1985 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
1da177e4 1986
341ce06f 1987 if (!wait) {
5c3240d9
AA
1988 /*
1989 * Not worth trying to allocate harder for
1990 * __GFP_NOMEMALLOC even if it can't schedule.
1991 */
1992 if (!(gfp_mask & __GFP_NOMEMALLOC))
1993 alloc_flags |= ALLOC_HARDER;
523b9458 1994 /*
341ce06f
PZ
1995 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1996 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
523b9458 1997 */
341ce06f 1998 alloc_flags &= ~ALLOC_CPUSET;
c06b1fca 1999 } else if (unlikely(rt_task(current)) && !in_interrupt())
341ce06f
PZ
2000 alloc_flags |= ALLOC_HARDER;
2001
2002 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2003 if (!in_interrupt() &&
c06b1fca 2004 ((current->flags & PF_MEMALLOC) ||
341ce06f
PZ
2005 unlikely(test_thread_flag(TIF_MEMDIE))))
2006 alloc_flags |= ALLOC_NO_WATERMARKS;
1da177e4 2007 }
6b1de916 2008
341ce06f
PZ
2009 return alloc_flags;
2010}
2011
11e33f6a
MG
2012static inline struct page *
2013__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2014 struct zonelist *zonelist, enum zone_type high_zoneidx,
3dd28266
MG
2015 nodemask_t *nodemask, struct zone *preferred_zone,
2016 int migratetype)
11e33f6a
MG
2017{
2018 const gfp_t wait = gfp_mask & __GFP_WAIT;
2019 struct page *page = NULL;
2020 int alloc_flags;
2021 unsigned long pages_reclaimed = 0;
2022 unsigned long did_some_progress;
77f1fe6b 2023 bool sync_migration = false;
1da177e4 2024
72807a74
MG
2025 /*
2026 * In the slowpath, we sanity check order to avoid ever trying to
2027 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2028 * be using allocators in order of preference for an area that is
2029 * too large.
2030 */
1fc28b70
MG
2031 if (order >= MAX_ORDER) {
2032 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
72807a74 2033 return NULL;
1fc28b70 2034 }
1da177e4 2035
952f3b51
CL
2036 /*
2037 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2038 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2039 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2040 * using a larger set of nodes after it has established that the
2041 * allowed per node queues are empty and that nodes are
2042 * over allocated.
2043 */
2044 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2045 goto nopage;
2046
cc4a6851 2047restart:
32dba98e
AA
2048 if (!(gfp_mask & __GFP_NO_KSWAPD))
2049 wake_all_kswapd(order, zonelist, high_zoneidx,
99504748 2050 zone_idx(preferred_zone));
1da177e4 2051
9bf2229f 2052 /*
7fb1d9fc
RS
2053 * OK, we're below the kswapd watermark and have kicked background
2054 * reclaim. Now things get more complex, so set up alloc_flags according
2055 * to how we want to proceed.
9bf2229f 2056 */
341ce06f 2057 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1da177e4 2058
f33261d7
DR
2059 /*
2060 * Find the true preferred zone if the allocation is unconstrained by
2061 * cpusets.
2062 */
2063 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2064 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2065 &preferred_zone);
2066
341ce06f 2067 /* This is the last chance, in general, before the goto nopage. */
19770b32 2068 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
341ce06f
PZ
2069 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2070 preferred_zone, migratetype);
7fb1d9fc
RS
2071 if (page)
2072 goto got_pg;
1da177e4 2073
b43a57bb 2074rebalance:
11e33f6a 2075 /* Allocate without watermarks if the context allows */
341ce06f
PZ
2076 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2077 page = __alloc_pages_high_priority(gfp_mask, order,
2078 zonelist, high_zoneidx, nodemask,
2079 preferred_zone, migratetype);
2080 if (page)
2081 goto got_pg;
1da177e4
LT
2082 }
2083
2084 /* Atomic allocations - we can't balance anything */
2085 if (!wait)
2086 goto nopage;
2087
341ce06f 2088 /* Avoid recursion of direct reclaim */
c06b1fca 2089 if (current->flags & PF_MEMALLOC)
341ce06f
PZ
2090 goto nopage;
2091
6583bb64
DR
2092 /* Avoid allocations with no watermarks from looping endlessly */
2093 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2094 goto nopage;
2095
77f1fe6b
MG
2096 /*
2097 * Try direct compaction. The first pass is asynchronous. Subsequent
2098 * attempts after direct reclaim are synchronous
2099 */
56de7263
MG
2100 page = __alloc_pages_direct_compact(gfp_mask, order,
2101 zonelist, high_zoneidx,
2102 nodemask,
2103 alloc_flags, preferred_zone,
77f1fe6b
MG
2104 migratetype, &did_some_progress,
2105 sync_migration);
56de7263
MG
2106 if (page)
2107 goto got_pg;
11bc82d6 2108 sync_migration = !(gfp_mask & __GFP_NO_KSWAPD);
56de7263 2109
11e33f6a
MG
2110 /* Try direct reclaim and then allocating */
2111 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2112 zonelist, high_zoneidx,
2113 nodemask,
5117f45d 2114 alloc_flags, preferred_zone,
3dd28266 2115 migratetype, &did_some_progress);
11e33f6a
MG
2116 if (page)
2117 goto got_pg;
1da177e4 2118
e33c3b5e 2119 /*
11e33f6a
MG
2120 * If we failed to make any progress reclaiming, then we are
2121 * running out of options and have to consider going OOM
e33c3b5e 2122 */
11e33f6a
MG
2123 if (!did_some_progress) {
2124 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
7f33d49a
RW
2125 if (oom_killer_disabled)
2126 goto nopage;
11e33f6a
MG
2127 page = __alloc_pages_may_oom(gfp_mask, order,
2128 zonelist, high_zoneidx,
3dd28266
MG
2129 nodemask, preferred_zone,
2130 migratetype);
11e33f6a
MG
2131 if (page)
2132 goto got_pg;
1da177e4 2133
03668b3c
DR
2134 if (!(gfp_mask & __GFP_NOFAIL)) {
2135 /*
2136 * The oom killer is not called for high-order
2137 * allocations that may fail, so if no progress
2138 * is being made, there are no other options and
2139 * retrying is unlikely to help.
2140 */
2141 if (order > PAGE_ALLOC_COSTLY_ORDER)
2142 goto nopage;
2143 /*
2144 * The oom killer is not called for lowmem
2145 * allocations to prevent needlessly killing
2146 * innocent tasks.
2147 */
2148 if (high_zoneidx < ZONE_NORMAL)
2149 goto nopage;
2150 }
e2c55dc8 2151
ff0ceb9d
DR
2152 goto restart;
2153 }
1da177e4
LT
2154 }
2155
11e33f6a 2156 /* Check if we should retry the allocation */
a41f24ea 2157 pages_reclaimed += did_some_progress;
11e33f6a
MG
2158 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2159 /* Wait for some write requests to complete then retry */
0e093d99 2160 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
1da177e4 2161 goto rebalance;
3e7d3449
MG
2162 } else {
2163 /*
2164 * High-order allocations do not necessarily loop after
2165 * direct reclaim and reclaim/compaction depends on compaction
2166 * being called after reclaim so call directly if necessary
2167 */
2168 page = __alloc_pages_direct_compact(gfp_mask, order,
2169 zonelist, high_zoneidx,
2170 nodemask,
2171 alloc_flags, preferred_zone,
77f1fe6b
MG
2172 migratetype, &did_some_progress,
2173 sync_migration);
3e7d3449
MG
2174 if (page)
2175 goto got_pg;
1da177e4
LT
2176 }
2177
2178nopage:
2179 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
cbf978bf
DR
2180 unsigned int filter = SHOW_MEM_FILTER_NODES;
2181
2182 /*
2183 * This documents exceptions given to allocations in certain
2184 * contexts that are allowed to allocate outside current's set
2185 * of allowed nodes.
2186 */
2187 if (!(gfp_mask & __GFP_NOMEMALLOC))
2188 if (test_thread_flag(TIF_MEMDIE) ||
2189 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2190 filter &= ~SHOW_MEM_FILTER_NODES;
2191 if (in_interrupt() || !wait)
2192 filter &= ~SHOW_MEM_FILTER_NODES;
2193
2194 pr_warning("%s: page allocation failure. order:%d, mode:0x%x\n",
c06b1fca 2195 current->comm, order, gfp_mask);
1da177e4 2196 dump_stack();
29423e77 2197 if (!should_suppress_show_mem())
b2b755b5 2198 show_mem(filter);
1da177e4 2199 }
b1eeab67 2200 return page;
1da177e4 2201got_pg:
b1eeab67
VN
2202 if (kmemcheck_enabled)
2203 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
1da177e4 2204 return page;
11e33f6a 2205
1da177e4 2206}
11e33f6a
MG
2207
2208/*
2209 * This is the 'heart' of the zoned buddy allocator.
2210 */
2211struct page *
2212__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2213 struct zonelist *zonelist, nodemask_t *nodemask)
2214{
2215 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
5117f45d 2216 struct zone *preferred_zone;
11e33f6a 2217 struct page *page;
3dd28266 2218 int migratetype = allocflags_to_migratetype(gfp_mask);
11e33f6a 2219
dcce284a
BH
2220 gfp_mask &= gfp_allowed_mask;
2221
11e33f6a
MG
2222 lockdep_trace_alloc(gfp_mask);
2223
2224 might_sleep_if(gfp_mask & __GFP_WAIT);
2225
2226 if (should_fail_alloc_page(gfp_mask, order))
2227 return NULL;
2228
2229 /*
2230 * Check the zones suitable for the gfp_mask contain at least one
2231 * valid zone. It's possible to have an empty zonelist as a result
2232 * of GFP_THISNODE and a memoryless node
2233 */
2234 if (unlikely(!zonelist->_zonerefs->zone))
2235 return NULL;
2236
c0ff7453 2237 get_mems_allowed();
5117f45d 2238 /* The preferred zone is used for statistics later */
f33261d7
DR
2239 first_zones_zonelist(zonelist, high_zoneidx,
2240 nodemask ? : &cpuset_current_mems_allowed,
2241 &preferred_zone);
c0ff7453
MX
2242 if (!preferred_zone) {
2243 put_mems_allowed();
5117f45d 2244 return NULL;
c0ff7453 2245 }
5117f45d
MG
2246
2247 /* First allocation attempt */
11e33f6a 2248 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
5117f45d 2249 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
3dd28266 2250 preferred_zone, migratetype);
11e33f6a
MG
2251 if (unlikely(!page))
2252 page = __alloc_pages_slowpath(gfp_mask, order,
5117f45d 2253 zonelist, high_zoneidx, nodemask,
3dd28266 2254 preferred_zone, migratetype);
c0ff7453 2255 put_mems_allowed();
11e33f6a 2256
4b4f278c 2257 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
11e33f6a 2258 return page;
1da177e4 2259}
d239171e 2260EXPORT_SYMBOL(__alloc_pages_nodemask);
1da177e4
LT
2261
2262/*
2263 * Common helper functions.
2264 */
920c7a5d 2265unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1da177e4 2266{
945a1113
AM
2267 struct page *page;
2268
2269 /*
2270 * __get_free_pages() returns a 32-bit address, which cannot represent
2271 * a highmem page
2272 */
2273 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2274
1da177e4
LT
2275 page = alloc_pages(gfp_mask, order);
2276 if (!page)
2277 return 0;
2278 return (unsigned long) page_address(page);
2279}
1da177e4
LT
2280EXPORT_SYMBOL(__get_free_pages);
2281
920c7a5d 2282unsigned long get_zeroed_page(gfp_t gfp_mask)
1da177e4 2283{
945a1113 2284 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
1da177e4 2285}
1da177e4
LT
2286EXPORT_SYMBOL(get_zeroed_page);
2287
2288void __pagevec_free(struct pagevec *pvec)
2289{
2290 int i = pagevec_count(pvec);
2291
4b4f278c
MG
2292 while (--i >= 0) {
2293 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
1da177e4 2294 free_hot_cold_page(pvec->pages[i], pvec->cold);
4b4f278c 2295 }
1da177e4
LT
2296}
2297
920c7a5d 2298void __free_pages(struct page *page, unsigned int order)
1da177e4 2299{
b5810039 2300 if (put_page_testzero(page)) {
1da177e4 2301 if (order == 0)
fc91668e 2302 free_hot_cold_page(page, 0);
1da177e4
LT
2303 else
2304 __free_pages_ok(page, order);
2305 }
2306}
2307
2308EXPORT_SYMBOL(__free_pages);
2309
920c7a5d 2310void free_pages(unsigned long addr, unsigned int order)
1da177e4
LT
2311{
2312 if (addr != 0) {
725d704e 2313 VM_BUG_ON(!virt_addr_valid((void *)addr));
1da177e4
LT
2314 __free_pages(virt_to_page((void *)addr), order);
2315 }
2316}
2317
2318EXPORT_SYMBOL(free_pages);
2319
2be0ffe2
TT
2320/**
2321 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2322 * @size: the number of bytes to allocate
2323 * @gfp_mask: GFP flags for the allocation
2324 *
2325 * This function is similar to alloc_pages(), except that it allocates the
2326 * minimum number of pages to satisfy the request. alloc_pages() can only
2327 * allocate memory in power-of-two pages.
2328 *
2329 * This function is also limited by MAX_ORDER.
2330 *
2331 * Memory allocated by this function must be released by free_pages_exact().
2332 */
2333void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2334{
2335 unsigned int order = get_order(size);
2336 unsigned long addr;
2337
2338 addr = __get_free_pages(gfp_mask, order);
2339 if (addr) {
2340 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2341 unsigned long used = addr + PAGE_ALIGN(size);
2342
5bfd7560 2343 split_page(virt_to_page((void *)addr), order);
2be0ffe2
TT
2344 while (used < alloc_end) {
2345 free_page(used);
2346 used += PAGE_SIZE;
2347 }
2348 }
2349
2350 return (void *)addr;
2351}
2352EXPORT_SYMBOL(alloc_pages_exact);
2353
2354/**
2355 * free_pages_exact - release memory allocated via alloc_pages_exact()
2356 * @virt: the value returned by alloc_pages_exact.
2357 * @size: size of allocation, same value as passed to alloc_pages_exact().
2358 *
2359 * Release the memory allocated by a previous call to alloc_pages_exact.
2360 */
2361void free_pages_exact(void *virt, size_t size)
2362{
2363 unsigned long addr = (unsigned long)virt;
2364 unsigned long end = addr + PAGE_ALIGN(size);
2365
2366 while (addr < end) {
2367 free_page(addr);
2368 addr += PAGE_SIZE;
2369 }
2370}
2371EXPORT_SYMBOL(free_pages_exact);
2372
1da177e4
LT
2373static unsigned int nr_free_zone_pages(int offset)
2374{
dd1a239f 2375 struct zoneref *z;
54a6eb5c
MG
2376 struct zone *zone;
2377
e310fd43 2378 /* Just pick one node, since fallback list is circular */
1da177e4
LT
2379 unsigned int sum = 0;
2380
0e88460d 2381 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1da177e4 2382
54a6eb5c 2383 for_each_zone_zonelist(zone, z, zonelist, offset) {
e310fd43 2384 unsigned long size = zone->present_pages;
41858966 2385 unsigned long high = high_wmark_pages(zone);
e310fd43
MB
2386 if (size > high)
2387 sum += size - high;
1da177e4
LT
2388 }
2389
2390 return sum;
2391}
2392
2393/*
2394 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2395 */
2396unsigned int nr_free_buffer_pages(void)
2397{
af4ca457 2398 return nr_free_zone_pages(gfp_zone(GFP_USER));
1da177e4 2399}
c2f1a551 2400EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1da177e4
LT
2401
2402/*
2403 * Amount of free RAM allocatable within all zones
2404 */
2405unsigned int nr_free_pagecache_pages(void)
2406{
2a1e274a 2407 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1da177e4 2408}
08e0f6a9
CL
2409
2410static inline void show_node(struct zone *zone)
1da177e4 2411{
08e0f6a9 2412 if (NUMA_BUILD)
25ba77c1 2413 printk("Node %d ", zone_to_nid(zone));
1da177e4 2414}
1da177e4 2415
1da177e4
LT
2416void si_meminfo(struct sysinfo *val)
2417{
2418 val->totalram = totalram_pages;
2419 val->sharedram = 0;
d23ad423 2420 val->freeram = global_page_state(NR_FREE_PAGES);
1da177e4 2421 val->bufferram = nr_blockdev_pages();
1da177e4
LT
2422 val->totalhigh = totalhigh_pages;
2423 val->freehigh = nr_free_highpages();
1da177e4
LT
2424 val->mem_unit = PAGE_SIZE;
2425}
2426
2427EXPORT_SYMBOL(si_meminfo);
2428
2429#ifdef CONFIG_NUMA
2430void si_meminfo_node(struct sysinfo *val, int nid)
2431{
2432 pg_data_t *pgdat = NODE_DATA(nid);
2433
2434 val->totalram = pgdat->node_present_pages;
d23ad423 2435 val->freeram = node_page_state(nid, NR_FREE_PAGES);
98d2b0eb 2436#ifdef CONFIG_HIGHMEM
1da177e4 2437 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
d23ad423
CL
2438 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2439 NR_FREE_PAGES);
98d2b0eb
CL
2440#else
2441 val->totalhigh = 0;
2442 val->freehigh = 0;
2443#endif
1da177e4
LT
2444 val->mem_unit = PAGE_SIZE;
2445}
2446#endif
2447
ddd588b5
DR
2448/*
2449 * Determine whether the zone's node should be displayed or not, depending on
2450 * whether SHOW_MEM_FILTER_NODES was passed to __show_free_areas().
2451 */
2452static bool skip_free_areas_zone(unsigned int flags, const struct zone *zone)
2453{
2454 bool ret = false;
2455
2456 if (!(flags & SHOW_MEM_FILTER_NODES))
2457 goto out;
2458
2459 get_mems_allowed();
2460 ret = !node_isset(zone->zone_pgdat->node_id,
2461 cpuset_current_mems_allowed);
2462 put_mems_allowed();
2463out:
2464 return ret;
2465}
2466
1da177e4
LT
2467#define K(x) ((x) << (PAGE_SHIFT-10))
2468
2469/*
2470 * Show free area list (used inside shift_scroll-lock stuff)
2471 * We also calculate the percentage fragmentation. We do this by counting the
2472 * memory on each free list with the exception of the first item on the list.
ddd588b5
DR
2473 * Suppresses nodes that are not allowed by current's cpuset if
2474 * SHOW_MEM_FILTER_NODES is passed.
1da177e4 2475 */
ddd588b5 2476void __show_free_areas(unsigned int filter)
1da177e4 2477{
c7241913 2478 int cpu;
1da177e4
LT
2479 struct zone *zone;
2480
ee99c71c 2481 for_each_populated_zone(zone) {
ddd588b5
DR
2482 if (skip_free_areas_zone(filter, zone))
2483 continue;
c7241913
JS
2484 show_node(zone);
2485 printk("%s per-cpu:\n", zone->name);
1da177e4 2486
6b482c67 2487 for_each_online_cpu(cpu) {
1da177e4
LT
2488 struct per_cpu_pageset *pageset;
2489
99dcc3e5 2490 pageset = per_cpu_ptr(zone->pageset, cpu);
1da177e4 2491
3dfa5721
CL
2492 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2493 cpu, pageset->pcp.high,
2494 pageset->pcp.batch, pageset->pcp.count);
1da177e4
LT
2495 }
2496 }
2497
a731286d
KM
2498 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2499 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
7b854121 2500 " unevictable:%lu"
b76146ed 2501 " dirty:%lu writeback:%lu unstable:%lu\n"
3701b033 2502 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4b02108a 2503 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
4f98a2fe 2504 global_page_state(NR_ACTIVE_ANON),
4f98a2fe 2505 global_page_state(NR_INACTIVE_ANON),
a731286d
KM
2506 global_page_state(NR_ISOLATED_ANON),
2507 global_page_state(NR_ACTIVE_FILE),
4f98a2fe 2508 global_page_state(NR_INACTIVE_FILE),
a731286d 2509 global_page_state(NR_ISOLATED_FILE),
7b854121 2510 global_page_state(NR_UNEVICTABLE),
b1e7a8fd 2511 global_page_state(NR_FILE_DIRTY),
ce866b34 2512 global_page_state(NR_WRITEBACK),
fd39fc85 2513 global_page_state(NR_UNSTABLE_NFS),
d23ad423 2514 global_page_state(NR_FREE_PAGES),
3701b033
KM
2515 global_page_state(NR_SLAB_RECLAIMABLE),
2516 global_page_state(NR_SLAB_UNRECLAIMABLE),
65ba55f5 2517 global_page_state(NR_FILE_MAPPED),
4b02108a 2518 global_page_state(NR_SHMEM),
a25700a5
AM
2519 global_page_state(NR_PAGETABLE),
2520 global_page_state(NR_BOUNCE));
1da177e4 2521
ee99c71c 2522 for_each_populated_zone(zone) {
1da177e4
LT
2523 int i;
2524
ddd588b5
DR
2525 if (skip_free_areas_zone(filter, zone))
2526 continue;
1da177e4
LT
2527 show_node(zone);
2528 printk("%s"
2529 " free:%lukB"
2530 " min:%lukB"
2531 " low:%lukB"
2532 " high:%lukB"
4f98a2fe
RR
2533 " active_anon:%lukB"
2534 " inactive_anon:%lukB"
2535 " active_file:%lukB"
2536 " inactive_file:%lukB"
7b854121 2537 " unevictable:%lukB"
a731286d
KM
2538 " isolated(anon):%lukB"
2539 " isolated(file):%lukB"
1da177e4 2540 " present:%lukB"
4a0aa73f
KM
2541 " mlocked:%lukB"
2542 " dirty:%lukB"
2543 " writeback:%lukB"
2544 " mapped:%lukB"
4b02108a 2545 " shmem:%lukB"
4a0aa73f
KM
2546 " slab_reclaimable:%lukB"
2547 " slab_unreclaimable:%lukB"
c6a7f572 2548 " kernel_stack:%lukB"
4a0aa73f
KM
2549 " pagetables:%lukB"
2550 " unstable:%lukB"
2551 " bounce:%lukB"
2552 " writeback_tmp:%lukB"
1da177e4
LT
2553 " pages_scanned:%lu"
2554 " all_unreclaimable? %s"
2555 "\n",
2556 zone->name,
88f5acf8 2557 K(zone_page_state(zone, NR_FREE_PAGES)),
41858966
MG
2558 K(min_wmark_pages(zone)),
2559 K(low_wmark_pages(zone)),
2560 K(high_wmark_pages(zone)),
4f98a2fe
RR
2561 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2562 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2563 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2564 K(zone_page_state(zone, NR_INACTIVE_FILE)),
7b854121 2565 K(zone_page_state(zone, NR_UNEVICTABLE)),
a731286d
KM
2566 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2567 K(zone_page_state(zone, NR_ISOLATED_FILE)),
1da177e4 2568 K(zone->present_pages),
4a0aa73f
KM
2569 K(zone_page_state(zone, NR_MLOCK)),
2570 K(zone_page_state(zone, NR_FILE_DIRTY)),
2571 K(zone_page_state(zone, NR_WRITEBACK)),
2572 K(zone_page_state(zone, NR_FILE_MAPPED)),
4b02108a 2573 K(zone_page_state(zone, NR_SHMEM)),
4a0aa73f
KM
2574 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2575 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
c6a7f572
KM
2576 zone_page_state(zone, NR_KERNEL_STACK) *
2577 THREAD_SIZE / 1024,
4a0aa73f
KM
2578 K(zone_page_state(zone, NR_PAGETABLE)),
2579 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2580 K(zone_page_state(zone, NR_BOUNCE)),
2581 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
1da177e4 2582 zone->pages_scanned,
93e4a89a 2583 (zone->all_unreclaimable ? "yes" : "no")
1da177e4
LT
2584 );
2585 printk("lowmem_reserve[]:");
2586 for (i = 0; i < MAX_NR_ZONES; i++)
2587 printk(" %lu", zone->lowmem_reserve[i]);
2588 printk("\n");
2589 }
2590
ee99c71c 2591 for_each_populated_zone(zone) {
8f9de51a 2592 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1da177e4 2593
ddd588b5
DR
2594 if (skip_free_areas_zone(filter, zone))
2595 continue;
1da177e4
LT
2596 show_node(zone);
2597 printk("%s: ", zone->name);
1da177e4
LT
2598
2599 spin_lock_irqsave(&zone->lock, flags);
2600 for (order = 0; order < MAX_ORDER; order++) {
8f9de51a
KK
2601 nr[order] = zone->free_area[order].nr_free;
2602 total += nr[order] << order;
1da177e4
LT
2603 }
2604 spin_unlock_irqrestore(&zone->lock, flags);
8f9de51a
KK
2605 for (order = 0; order < MAX_ORDER; order++)
2606 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1da177e4
LT
2607 printk("= %lukB\n", K(total));
2608 }
2609
e6f3602d
LW
2610 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2611
1da177e4
LT
2612 show_swap_cache_info();
2613}
2614
ddd588b5
DR
2615void show_free_areas(void)
2616{
2617 __show_free_areas(0);
2618}
2619
19770b32
MG
2620static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2621{
2622 zoneref->zone = zone;
2623 zoneref->zone_idx = zone_idx(zone);
2624}
2625
1da177e4
LT
2626/*
2627 * Builds allocation fallback zone lists.
1a93205b
CL
2628 *
2629 * Add all populated zones of a node to the zonelist.
1da177e4 2630 */
f0c0b2b8
KH
2631static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2632 int nr_zones, enum zone_type zone_type)
1da177e4 2633{
1a93205b
CL
2634 struct zone *zone;
2635
98d2b0eb 2636 BUG_ON(zone_type >= MAX_NR_ZONES);
2f6726e5 2637 zone_type++;
02a68a5e
CL
2638
2639 do {
2f6726e5 2640 zone_type--;
070f8032 2641 zone = pgdat->node_zones + zone_type;
1a93205b 2642 if (populated_zone(zone)) {
dd1a239f
MG
2643 zoneref_set_zone(zone,
2644 &zonelist->_zonerefs[nr_zones++]);
070f8032 2645 check_highest_zone(zone_type);
1da177e4 2646 }
02a68a5e 2647
2f6726e5 2648 } while (zone_type);
070f8032 2649 return nr_zones;
1da177e4
LT
2650}
2651
f0c0b2b8
KH
2652
2653/*
2654 * zonelist_order:
2655 * 0 = automatic detection of better ordering.
2656 * 1 = order by ([node] distance, -zonetype)
2657 * 2 = order by (-zonetype, [node] distance)
2658 *
2659 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2660 * the same zonelist. So only NUMA can configure this param.
2661 */
2662#define ZONELIST_ORDER_DEFAULT 0
2663#define ZONELIST_ORDER_NODE 1
2664#define ZONELIST_ORDER_ZONE 2
2665
2666/* zonelist order in the kernel.
2667 * set_zonelist_order() will set this to NODE or ZONE.
2668 */
2669static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2670static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2671
2672
1da177e4 2673#ifdef CONFIG_NUMA
f0c0b2b8
KH
2674/* The value user specified ....changed by config */
2675static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2676/* string for sysctl */
2677#define NUMA_ZONELIST_ORDER_LEN 16
2678char numa_zonelist_order[16] = "default";
2679
2680/*
2681 * interface for configure zonelist ordering.
2682 * command line option "numa_zonelist_order"
2683 * = "[dD]efault - default, automatic configuration.
2684 * = "[nN]ode - order by node locality, then by zone within node
2685 * = "[zZ]one - order by zone, then by locality within zone
2686 */
2687
2688static int __parse_numa_zonelist_order(char *s)
2689{
2690 if (*s == 'd' || *s == 'D') {
2691 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2692 } else if (*s == 'n' || *s == 'N') {
2693 user_zonelist_order = ZONELIST_ORDER_NODE;
2694 } else if (*s == 'z' || *s == 'Z') {
2695 user_zonelist_order = ZONELIST_ORDER_ZONE;
2696 } else {
2697 printk(KERN_WARNING
2698 "Ignoring invalid numa_zonelist_order value: "
2699 "%s\n", s);
2700 return -EINVAL;
2701 }
2702 return 0;
2703}
2704
2705static __init int setup_numa_zonelist_order(char *s)
2706{
ecb256f8
VL
2707 int ret;
2708
2709 if (!s)
2710 return 0;
2711
2712 ret = __parse_numa_zonelist_order(s);
2713 if (ret == 0)
2714 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2715
2716 return ret;
f0c0b2b8
KH
2717}
2718early_param("numa_zonelist_order", setup_numa_zonelist_order);
2719
2720/*
2721 * sysctl handler for numa_zonelist_order
2722 */
2723int numa_zonelist_order_handler(ctl_table *table, int write,
8d65af78 2724 void __user *buffer, size_t *length,
f0c0b2b8
KH
2725 loff_t *ppos)
2726{
2727 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2728 int ret;
443c6f14 2729 static DEFINE_MUTEX(zl_order_mutex);
f0c0b2b8 2730
443c6f14 2731 mutex_lock(&zl_order_mutex);
f0c0b2b8 2732 if (write)
443c6f14 2733 strcpy(saved_string, (char*)table->data);
8d65af78 2734 ret = proc_dostring(table, write, buffer, length, ppos);
f0c0b2b8 2735 if (ret)
443c6f14 2736 goto out;
f0c0b2b8
KH
2737 if (write) {
2738 int oldval = user_zonelist_order;
2739 if (__parse_numa_zonelist_order((char*)table->data)) {
2740 /*
2741 * bogus value. restore saved string
2742 */
2743 strncpy((char*)table->data, saved_string,
2744 NUMA_ZONELIST_ORDER_LEN);
2745 user_zonelist_order = oldval;
4eaf3f64
HL
2746 } else if (oldval != user_zonelist_order) {
2747 mutex_lock(&zonelists_mutex);
1f522509 2748 build_all_zonelists(NULL);
4eaf3f64
HL
2749 mutex_unlock(&zonelists_mutex);
2750 }
f0c0b2b8 2751 }
443c6f14
AK
2752out:
2753 mutex_unlock(&zl_order_mutex);
2754 return ret;
f0c0b2b8
KH
2755}
2756
2757
62bc62a8 2758#define MAX_NODE_LOAD (nr_online_nodes)
f0c0b2b8
KH
2759static int node_load[MAX_NUMNODES];
2760
1da177e4 2761/**
4dc3b16b 2762 * find_next_best_node - find the next node that should appear in a given node's fallback list
1da177e4
LT
2763 * @node: node whose fallback list we're appending
2764 * @used_node_mask: nodemask_t of already used nodes
2765 *
2766 * We use a number of factors to determine which is the next node that should
2767 * appear on a given node's fallback list. The node should not have appeared
2768 * already in @node's fallback list, and it should be the next closest node
2769 * according to the distance array (which contains arbitrary distance values
2770 * from each node to each node in the system), and should also prefer nodes
2771 * with no CPUs, since presumably they'll have very little allocation pressure
2772 * on them otherwise.
2773 * It returns -1 if no node is found.
2774 */
f0c0b2b8 2775static int find_next_best_node(int node, nodemask_t *used_node_mask)
1da177e4 2776{
4cf808eb 2777 int n, val;
1da177e4
LT
2778 int min_val = INT_MAX;
2779 int best_node = -1;
a70f7302 2780 const struct cpumask *tmp = cpumask_of_node(0);
1da177e4 2781
4cf808eb
LT
2782 /* Use the local node if we haven't already */
2783 if (!node_isset(node, *used_node_mask)) {
2784 node_set(node, *used_node_mask);
2785 return node;
2786 }
1da177e4 2787
37b07e41 2788 for_each_node_state(n, N_HIGH_MEMORY) {
1da177e4
LT
2789
2790 /* Don't want a node to appear more than once */
2791 if (node_isset(n, *used_node_mask))
2792 continue;
2793
1da177e4
LT
2794 /* Use the distance array to find the distance */
2795 val = node_distance(node, n);
2796
4cf808eb
LT
2797 /* Penalize nodes under us ("prefer the next node") */
2798 val += (n < node);
2799
1da177e4 2800 /* Give preference to headless and unused nodes */
a70f7302
RR
2801 tmp = cpumask_of_node(n);
2802 if (!cpumask_empty(tmp))
1da177e4
LT
2803 val += PENALTY_FOR_NODE_WITH_CPUS;
2804
2805 /* Slight preference for less loaded node */
2806 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2807 val += node_load[n];
2808
2809 if (val < min_val) {
2810 min_val = val;
2811 best_node = n;
2812 }
2813 }
2814
2815 if (best_node >= 0)
2816 node_set(best_node, *used_node_mask);
2817
2818 return best_node;
2819}
2820
f0c0b2b8
KH
2821
2822/*
2823 * Build zonelists ordered by node and zones within node.
2824 * This results in maximum locality--normal zone overflows into local
2825 * DMA zone, if any--but risks exhausting DMA zone.
2826 */
2827static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
1da177e4 2828{
f0c0b2b8 2829 int j;
1da177e4 2830 struct zonelist *zonelist;
f0c0b2b8 2831
54a6eb5c 2832 zonelist = &pgdat->node_zonelists[0];
dd1a239f 2833 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
54a6eb5c
MG
2834 ;
2835 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2836 MAX_NR_ZONES - 1);
dd1a239f
MG
2837 zonelist->_zonerefs[j].zone = NULL;
2838 zonelist->_zonerefs[j].zone_idx = 0;
f0c0b2b8
KH
2839}
2840
523b9458
CL
2841/*
2842 * Build gfp_thisnode zonelists
2843 */
2844static void build_thisnode_zonelists(pg_data_t *pgdat)
2845{
523b9458
CL
2846 int j;
2847 struct zonelist *zonelist;
2848
54a6eb5c
MG
2849 zonelist = &pgdat->node_zonelists[1];
2850 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
dd1a239f
MG
2851 zonelist->_zonerefs[j].zone = NULL;
2852 zonelist->_zonerefs[j].zone_idx = 0;
523b9458
CL
2853}
2854
f0c0b2b8
KH
2855/*
2856 * Build zonelists ordered by zone and nodes within zones.
2857 * This results in conserving DMA zone[s] until all Normal memory is
2858 * exhausted, but results in overflowing to remote node while memory
2859 * may still exist in local DMA zone.
2860 */
2861static int node_order[MAX_NUMNODES];
2862
2863static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2864{
f0c0b2b8
KH
2865 int pos, j, node;
2866 int zone_type; /* needs to be signed */
2867 struct zone *z;
2868 struct zonelist *zonelist;
2869
54a6eb5c
MG
2870 zonelist = &pgdat->node_zonelists[0];
2871 pos = 0;
2872 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2873 for (j = 0; j < nr_nodes; j++) {
2874 node = node_order[j];
2875 z = &NODE_DATA(node)->node_zones[zone_type];
2876 if (populated_zone(z)) {
dd1a239f
MG
2877 zoneref_set_zone(z,
2878 &zonelist->_zonerefs[pos++]);
54a6eb5c 2879 check_highest_zone(zone_type);
f0c0b2b8
KH
2880 }
2881 }
f0c0b2b8 2882 }
dd1a239f
MG
2883 zonelist->_zonerefs[pos].zone = NULL;
2884 zonelist->_zonerefs[pos].zone_idx = 0;
f0c0b2b8
KH
2885}
2886
2887static int default_zonelist_order(void)
2888{
2889 int nid, zone_type;
2890 unsigned long low_kmem_size,total_size;
2891 struct zone *z;
2892 int average_size;
2893 /*
88393161 2894 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
f0c0b2b8
KH
2895 * If they are really small and used heavily, the system can fall
2896 * into OOM very easily.
e325c90f 2897 * This function detect ZONE_DMA/DMA32 size and configures zone order.
f0c0b2b8
KH
2898 */
2899 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2900 low_kmem_size = 0;
2901 total_size = 0;
2902 for_each_online_node(nid) {
2903 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2904 z = &NODE_DATA(nid)->node_zones[zone_type];
2905 if (populated_zone(z)) {
2906 if (zone_type < ZONE_NORMAL)
2907 low_kmem_size += z->present_pages;
2908 total_size += z->present_pages;
e325c90f
DR
2909 } else if (zone_type == ZONE_NORMAL) {
2910 /*
2911 * If any node has only lowmem, then node order
2912 * is preferred to allow kernel allocations
2913 * locally; otherwise, they can easily infringe
2914 * on other nodes when there is an abundance of
2915 * lowmem available to allocate from.
2916 */
2917 return ZONELIST_ORDER_NODE;
f0c0b2b8
KH
2918 }
2919 }
2920 }
2921 if (!low_kmem_size || /* there are no DMA area. */
2922 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2923 return ZONELIST_ORDER_NODE;
2924 /*
2925 * look into each node's config.
2926 * If there is a node whose DMA/DMA32 memory is very big area on
2927 * local memory, NODE_ORDER may be suitable.
2928 */
37b07e41
LS
2929 average_size = total_size /
2930 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
f0c0b2b8
KH
2931 for_each_online_node(nid) {
2932 low_kmem_size = 0;
2933 total_size = 0;
2934 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2935 z = &NODE_DATA(nid)->node_zones[zone_type];
2936 if (populated_zone(z)) {
2937 if (zone_type < ZONE_NORMAL)
2938 low_kmem_size += z->present_pages;
2939 total_size += z->present_pages;
2940 }
2941 }
2942 if (low_kmem_size &&
2943 total_size > average_size && /* ignore small node */
2944 low_kmem_size > total_size * 70/100)
2945 return ZONELIST_ORDER_NODE;
2946 }
2947 return ZONELIST_ORDER_ZONE;
2948}
2949
2950static void set_zonelist_order(void)
2951{
2952 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2953 current_zonelist_order = default_zonelist_order();
2954 else
2955 current_zonelist_order = user_zonelist_order;
2956}
2957
2958static void build_zonelists(pg_data_t *pgdat)
2959{
2960 int j, node, load;
2961 enum zone_type i;
1da177e4 2962 nodemask_t used_mask;
f0c0b2b8
KH
2963 int local_node, prev_node;
2964 struct zonelist *zonelist;
2965 int order = current_zonelist_order;
1da177e4
LT
2966
2967 /* initialize zonelists */
523b9458 2968 for (i = 0; i < MAX_ZONELISTS; i++) {
1da177e4 2969 zonelist = pgdat->node_zonelists + i;
dd1a239f
MG
2970 zonelist->_zonerefs[0].zone = NULL;
2971 zonelist->_zonerefs[0].zone_idx = 0;
1da177e4
LT
2972 }
2973
2974 /* NUMA-aware ordering of nodes */
2975 local_node = pgdat->node_id;
62bc62a8 2976 load = nr_online_nodes;
1da177e4
LT
2977 prev_node = local_node;
2978 nodes_clear(used_mask);
f0c0b2b8 2979
f0c0b2b8
KH
2980 memset(node_order, 0, sizeof(node_order));
2981 j = 0;
2982
1da177e4 2983 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
9eeff239
CL
2984 int distance = node_distance(local_node, node);
2985
2986 /*
2987 * If another node is sufficiently far away then it is better
2988 * to reclaim pages in a zone before going off node.
2989 */
2990 if (distance > RECLAIM_DISTANCE)
2991 zone_reclaim_mode = 1;
2992
1da177e4
LT
2993 /*
2994 * We don't want to pressure a particular node.
2995 * So adding penalty to the first node in same
2996 * distance group to make it round-robin.
2997 */
9eeff239 2998 if (distance != node_distance(local_node, prev_node))
f0c0b2b8
KH
2999 node_load[node] = load;
3000
1da177e4
LT
3001 prev_node = node;
3002 load--;
f0c0b2b8
KH
3003 if (order == ZONELIST_ORDER_NODE)
3004 build_zonelists_in_node_order(pgdat, node);
3005 else
3006 node_order[j++] = node; /* remember order */
3007 }
1da177e4 3008
f0c0b2b8
KH
3009 if (order == ZONELIST_ORDER_ZONE) {
3010 /* calculate node order -- i.e., DMA last! */
3011 build_zonelists_in_zone_order(pgdat, j);
1da177e4 3012 }
523b9458
CL
3013
3014 build_thisnode_zonelists(pgdat);
1da177e4
LT
3015}
3016
9276b1bc 3017/* Construct the zonelist performance cache - see further mmzone.h */
f0c0b2b8 3018static void build_zonelist_cache(pg_data_t *pgdat)
9276b1bc 3019{
54a6eb5c
MG
3020 struct zonelist *zonelist;
3021 struct zonelist_cache *zlc;
dd1a239f 3022 struct zoneref *z;
9276b1bc 3023
54a6eb5c
MG
3024 zonelist = &pgdat->node_zonelists[0];
3025 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3026 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
dd1a239f
MG
3027 for (z = zonelist->_zonerefs; z->zone; z++)
3028 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
9276b1bc
PJ
3029}
3030
7aac7898
LS
3031#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3032/*
3033 * Return node id of node used for "local" allocations.
3034 * I.e., first node id of first zone in arg node's generic zonelist.
3035 * Used for initializing percpu 'numa_mem', which is used primarily
3036 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3037 */
3038int local_memory_node(int node)
3039{
3040 struct zone *zone;
3041
3042 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3043 gfp_zone(GFP_KERNEL),
3044 NULL,
3045 &zone);
3046 return zone->node;
3047}
3048#endif
f0c0b2b8 3049
1da177e4
LT
3050#else /* CONFIG_NUMA */
3051
f0c0b2b8
KH
3052static void set_zonelist_order(void)
3053{
3054 current_zonelist_order = ZONELIST_ORDER_ZONE;
3055}
3056
3057static void build_zonelists(pg_data_t *pgdat)
1da177e4 3058{
19655d34 3059 int node, local_node;
54a6eb5c
MG
3060 enum zone_type j;
3061 struct zonelist *zonelist;
1da177e4
LT
3062
3063 local_node = pgdat->node_id;
1da177e4 3064
54a6eb5c
MG
3065 zonelist = &pgdat->node_zonelists[0];
3066 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
1da177e4 3067
54a6eb5c
MG
3068 /*
3069 * Now we build the zonelist so that it contains the zones
3070 * of all the other nodes.
3071 * We don't want to pressure a particular node, so when
3072 * building the zones for node N, we make sure that the
3073 * zones coming right after the local ones are those from
3074 * node N+1 (modulo N)
3075 */
3076 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3077 if (!node_online(node))
3078 continue;
3079 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3080 MAX_NR_ZONES - 1);
1da177e4 3081 }
54a6eb5c
MG
3082 for (node = 0; node < local_node; node++) {
3083 if (!node_online(node))
3084 continue;
3085 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3086 MAX_NR_ZONES - 1);
3087 }
3088
dd1a239f
MG
3089 zonelist->_zonerefs[j].zone = NULL;
3090 zonelist->_zonerefs[j].zone_idx = 0;
1da177e4
LT
3091}
3092
9276b1bc 3093/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
f0c0b2b8 3094static void build_zonelist_cache(pg_data_t *pgdat)
9276b1bc 3095{
54a6eb5c 3096 pgdat->node_zonelists[0].zlcache_ptr = NULL;
9276b1bc
PJ
3097}
3098
1da177e4
LT
3099#endif /* CONFIG_NUMA */
3100
99dcc3e5
CL
3101/*
3102 * Boot pageset table. One per cpu which is going to be used for all
3103 * zones and all nodes. The parameters will be set in such a way
3104 * that an item put on a list will immediately be handed over to
3105 * the buddy list. This is safe since pageset manipulation is done
3106 * with interrupts disabled.
3107 *
3108 * The boot_pagesets must be kept even after bootup is complete for
3109 * unused processors and/or zones. They do play a role for bootstrapping
3110 * hotplugged processors.
3111 *
3112 * zoneinfo_show() and maybe other functions do
3113 * not check if the processor is online before following the pageset pointer.
3114 * Other parts of the kernel may not check if the zone is available.
3115 */
3116static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3117static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
1f522509 3118static void setup_zone_pageset(struct zone *zone);
99dcc3e5 3119
4eaf3f64
HL
3120/*
3121 * Global mutex to protect against size modification of zonelists
3122 * as well as to serialize pageset setup for the new populated zone.
3123 */
3124DEFINE_MUTEX(zonelists_mutex);
3125
9b1a4d38 3126/* return values int ....just for stop_machine() */
1f522509 3127static __init_refok int __build_all_zonelists(void *data)
1da177e4 3128{
6811378e 3129 int nid;
99dcc3e5 3130 int cpu;
9276b1bc 3131
7f9cfb31
BL
3132#ifdef CONFIG_NUMA
3133 memset(node_load, 0, sizeof(node_load));
3134#endif
9276b1bc 3135 for_each_online_node(nid) {
7ea1530a
CL
3136 pg_data_t *pgdat = NODE_DATA(nid);
3137
3138 build_zonelists(pgdat);
3139 build_zonelist_cache(pgdat);
9276b1bc 3140 }
99dcc3e5
CL
3141
3142 /*
3143 * Initialize the boot_pagesets that are going to be used
3144 * for bootstrapping processors. The real pagesets for
3145 * each zone will be allocated later when the per cpu
3146 * allocator is available.
3147 *
3148 * boot_pagesets are used also for bootstrapping offline
3149 * cpus if the system is already booted because the pagesets
3150 * are needed to initialize allocators on a specific cpu too.
3151 * F.e. the percpu allocator needs the page allocator which
3152 * needs the percpu allocator in order to allocate its pagesets
3153 * (a chicken-egg dilemma).
3154 */
7aac7898 3155 for_each_possible_cpu(cpu) {
99dcc3e5
CL
3156 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3157
7aac7898
LS
3158#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3159 /*
3160 * We now know the "local memory node" for each node--
3161 * i.e., the node of the first zone in the generic zonelist.
3162 * Set up numa_mem percpu variable for on-line cpus. During
3163 * boot, only the boot cpu should be on-line; we'll init the
3164 * secondary cpus' numa_mem as they come on-line. During
3165 * node/memory hotplug, we'll fixup all on-line cpus.
3166 */
3167 if (cpu_online(cpu))
3168 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3169#endif
3170 }
3171
6811378e
YG
3172 return 0;
3173}
3174
4eaf3f64
HL
3175/*
3176 * Called with zonelists_mutex held always
3177 * unless system_state == SYSTEM_BOOTING.
3178 */
9f6ae448 3179void __ref build_all_zonelists(void *data)
6811378e 3180{
f0c0b2b8
KH
3181 set_zonelist_order();
3182
6811378e 3183 if (system_state == SYSTEM_BOOTING) {
423b41d7 3184 __build_all_zonelists(NULL);
68ad8df4 3185 mminit_verify_zonelist();
6811378e
YG
3186 cpuset_init_current_mems_allowed();
3187 } else {
183ff22b 3188 /* we have to stop all cpus to guarantee there is no user
6811378e 3189 of zonelist */
e9959f0f
KH
3190#ifdef CONFIG_MEMORY_HOTPLUG
3191 if (data)
3192 setup_zone_pageset((struct zone *)data);
3193#endif
3194 stop_machine(__build_all_zonelists, NULL, NULL);
6811378e
YG
3195 /* cpuset refresh routine should be here */
3196 }
bd1e22b8 3197 vm_total_pages = nr_free_pagecache_pages();
9ef9acb0
MG
3198 /*
3199 * Disable grouping by mobility if the number of pages in the
3200 * system is too low to allow the mechanism to work. It would be
3201 * more accurate, but expensive to check per-zone. This check is
3202 * made on memory-hotadd so a system can start with mobility
3203 * disabled and enable it later
3204 */
d9c23400 3205 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
9ef9acb0
MG
3206 page_group_by_mobility_disabled = 1;
3207 else
3208 page_group_by_mobility_disabled = 0;
3209
3210 printk("Built %i zonelists in %s order, mobility grouping %s. "
3211 "Total pages: %ld\n",
62bc62a8 3212 nr_online_nodes,
f0c0b2b8 3213 zonelist_order_name[current_zonelist_order],
9ef9acb0 3214 page_group_by_mobility_disabled ? "off" : "on",
f0c0b2b8
KH
3215 vm_total_pages);
3216#ifdef CONFIG_NUMA
3217 printk("Policy zone: %s\n", zone_names[policy_zone]);
3218#endif
1da177e4
LT
3219}
3220
3221/*
3222 * Helper functions to size the waitqueue hash table.
3223 * Essentially these want to choose hash table sizes sufficiently
3224 * large so that collisions trying to wait on pages are rare.
3225 * But in fact, the number of active page waitqueues on typical
3226 * systems is ridiculously low, less than 200. So this is even
3227 * conservative, even though it seems large.
3228 *
3229 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3230 * waitqueues, i.e. the size of the waitq table given the number of pages.
3231 */
3232#define PAGES_PER_WAITQUEUE 256
3233
cca448fe 3234#ifndef CONFIG_MEMORY_HOTPLUG
02b694de 3235static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1da177e4
LT
3236{
3237 unsigned long size = 1;
3238
3239 pages /= PAGES_PER_WAITQUEUE;
3240
3241 while (size < pages)
3242 size <<= 1;
3243
3244 /*
3245 * Once we have dozens or even hundreds of threads sleeping
3246 * on IO we've got bigger problems than wait queue collision.
3247 * Limit the size of the wait table to a reasonable size.
3248 */
3249 size = min(size, 4096UL);
3250
3251 return max(size, 4UL);
3252}
cca448fe
YG
3253#else
3254/*
3255 * A zone's size might be changed by hot-add, so it is not possible to determine
3256 * a suitable size for its wait_table. So we use the maximum size now.
3257 *
3258 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3259 *
3260 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3261 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3262 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3263 *
3264 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3265 * or more by the traditional way. (See above). It equals:
3266 *
3267 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3268 * ia64(16K page size) : = ( 8G + 4M)byte.
3269 * powerpc (64K page size) : = (32G +16M)byte.
3270 */
3271static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3272{
3273 return 4096UL;
3274}
3275#endif
1da177e4
LT
3276
3277/*
3278 * This is an integer logarithm so that shifts can be used later
3279 * to extract the more random high bits from the multiplicative
3280 * hash function before the remainder is taken.
3281 */
3282static inline unsigned long wait_table_bits(unsigned long size)
3283{
3284 return ffz(~size);
3285}
3286
3287#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3288
56fd56b8 3289/*
d9c23400 3290 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
41858966
MG
3291 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3292 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
56fd56b8
MG
3293 * higher will lead to a bigger reserve which will get freed as contiguous
3294 * blocks as reclaim kicks in
3295 */
3296static void setup_zone_migrate_reserve(struct zone *zone)
3297{
3298 unsigned long start_pfn, pfn, end_pfn;
3299 struct page *page;
78986a67
MG
3300 unsigned long block_migratetype;
3301 int reserve;
56fd56b8
MG
3302
3303 /* Get the start pfn, end pfn and the number of blocks to reserve */
3304 start_pfn = zone->zone_start_pfn;
3305 end_pfn = start_pfn + zone->spanned_pages;
41858966 3306 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
d9c23400 3307 pageblock_order;
56fd56b8 3308
78986a67
MG
3309 /*
3310 * Reserve blocks are generally in place to help high-order atomic
3311 * allocations that are short-lived. A min_free_kbytes value that
3312 * would result in more than 2 reserve blocks for atomic allocations
3313 * is assumed to be in place to help anti-fragmentation for the
3314 * future allocation of hugepages at runtime.
3315 */
3316 reserve = min(2, reserve);
3317
d9c23400 3318 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
56fd56b8
MG
3319 if (!pfn_valid(pfn))
3320 continue;
3321 page = pfn_to_page(pfn);
3322
344c790e
AL
3323 /* Watch out for overlapping nodes */
3324 if (page_to_nid(page) != zone_to_nid(zone))
3325 continue;
3326
56fd56b8
MG
3327 /* Blocks with reserved pages will never free, skip them. */
3328 if (PageReserved(page))
3329 continue;
3330
3331 block_migratetype = get_pageblock_migratetype(page);
3332
3333 /* If this block is reserved, account for it */
3334 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3335 reserve--;
3336 continue;
3337 }
3338
3339 /* Suitable for reserving if this block is movable */
3340 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3341 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3342 move_freepages_block(zone, page, MIGRATE_RESERVE);
3343 reserve--;
3344 continue;
3345 }
3346
3347 /*
3348 * If the reserve is met and this is a previous reserved block,
3349 * take it back
3350 */
3351 if (block_migratetype == MIGRATE_RESERVE) {
3352 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3353 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3354 }
3355 }
3356}
ac0e5b7a 3357
1da177e4
LT
3358/*
3359 * Initially all pages are reserved - free ones are freed
3360 * up by free_all_bootmem() once the early boot process is
3361 * done. Non-atomic initialization, single-pass.
3362 */
c09b4240 3363void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
a2f3aa02 3364 unsigned long start_pfn, enum memmap_context context)
1da177e4 3365{
1da177e4 3366 struct page *page;
29751f69
AW
3367 unsigned long end_pfn = start_pfn + size;
3368 unsigned long pfn;
86051ca5 3369 struct zone *z;
1da177e4 3370
22b31eec
HD
3371 if (highest_memmap_pfn < end_pfn - 1)
3372 highest_memmap_pfn = end_pfn - 1;
3373
86051ca5 3374 z = &NODE_DATA(nid)->node_zones[zone];
cbe8dd4a 3375 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
a2f3aa02
DH
3376 /*
3377 * There can be holes in boot-time mem_map[]s
3378 * handed to this function. They do not
3379 * exist on hotplugged memory.
3380 */
3381 if (context == MEMMAP_EARLY) {
3382 if (!early_pfn_valid(pfn))
3383 continue;
3384 if (!early_pfn_in_nid(pfn, nid))
3385 continue;
3386 }
d41dee36
AW
3387 page = pfn_to_page(pfn);
3388 set_page_links(page, zone, nid, pfn);
708614e6 3389 mminit_verify_page_links(page, zone, nid, pfn);
7835e98b 3390 init_page_count(page);
1da177e4
LT
3391 reset_page_mapcount(page);
3392 SetPageReserved(page);
b2a0ac88
MG
3393 /*
3394 * Mark the block movable so that blocks are reserved for
3395 * movable at startup. This will force kernel allocations
3396 * to reserve their blocks rather than leaking throughout
3397 * the address space during boot when many long-lived
56fd56b8
MG
3398 * kernel allocations are made. Later some blocks near
3399 * the start are marked MIGRATE_RESERVE by
3400 * setup_zone_migrate_reserve()
86051ca5
KH
3401 *
3402 * bitmap is created for zone's valid pfn range. but memmap
3403 * can be created for invalid pages (for alignment)
3404 * check here not to call set_pageblock_migratetype() against
3405 * pfn out of zone.
b2a0ac88 3406 */
86051ca5
KH
3407 if ((z->zone_start_pfn <= pfn)
3408 && (pfn < z->zone_start_pfn + z->spanned_pages)
3409 && !(pfn & (pageblock_nr_pages - 1)))
56fd56b8 3410 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
b2a0ac88 3411
1da177e4
LT
3412 INIT_LIST_HEAD(&page->lru);
3413#ifdef WANT_PAGE_VIRTUAL
3414 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3415 if (!is_highmem_idx(zone))
3212c6be 3416 set_page_address(page, __va(pfn << PAGE_SHIFT));
1da177e4 3417#endif
1da177e4
LT
3418 }
3419}
3420
1e548deb 3421static void __meminit zone_init_free_lists(struct zone *zone)
1da177e4 3422{
b2a0ac88
MG
3423 int order, t;
3424 for_each_migratetype_order(order, t) {
3425 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1da177e4
LT
3426 zone->free_area[order].nr_free = 0;
3427 }
3428}
3429
3430#ifndef __HAVE_ARCH_MEMMAP_INIT
3431#define memmap_init(size, nid, zone, start_pfn) \
a2f3aa02 3432 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
1da177e4
LT
3433#endif
3434
1d6f4e60 3435static int zone_batchsize(struct zone *zone)
e7c8d5c9 3436{
3a6be87f 3437#ifdef CONFIG_MMU
e7c8d5c9
CL
3438 int batch;
3439
3440 /*
3441 * The per-cpu-pages pools are set to around 1000th of the
ba56e91c 3442 * size of the zone. But no more than 1/2 of a meg.
e7c8d5c9
CL
3443 *
3444 * OK, so we don't know how big the cache is. So guess.
3445 */
3446 batch = zone->present_pages / 1024;
ba56e91c
SR
3447 if (batch * PAGE_SIZE > 512 * 1024)
3448 batch = (512 * 1024) / PAGE_SIZE;
e7c8d5c9
CL
3449 batch /= 4; /* We effectively *= 4 below */
3450 if (batch < 1)
3451 batch = 1;
3452
3453 /*
0ceaacc9
NP
3454 * Clamp the batch to a 2^n - 1 value. Having a power
3455 * of 2 value was found to be more likely to have
3456 * suboptimal cache aliasing properties in some cases.
e7c8d5c9 3457 *
0ceaacc9
NP
3458 * For example if 2 tasks are alternately allocating
3459 * batches of pages, one task can end up with a lot
3460 * of pages of one half of the possible page colors
3461 * and the other with pages of the other colors.
e7c8d5c9 3462 */
9155203a 3463 batch = rounddown_pow_of_two(batch + batch/2) - 1;
ba56e91c 3464
e7c8d5c9 3465 return batch;
3a6be87f
DH
3466
3467#else
3468 /* The deferral and batching of frees should be suppressed under NOMMU
3469 * conditions.
3470 *
3471 * The problem is that NOMMU needs to be able to allocate large chunks
3472 * of contiguous memory as there's no hardware page translation to
3473 * assemble apparent contiguous memory from discontiguous pages.
3474 *
3475 * Queueing large contiguous runs of pages for batching, however,
3476 * causes the pages to actually be freed in smaller chunks. As there
3477 * can be a significant delay between the individual batches being
3478 * recycled, this leads to the once large chunks of space being
3479 * fragmented and becoming unavailable for high-order allocations.
3480 */
3481 return 0;
3482#endif
e7c8d5c9
CL
3483}
3484
b69a7288 3485static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2caaad41
CL
3486{
3487 struct per_cpu_pages *pcp;
5f8dcc21 3488 int migratetype;
2caaad41 3489
1c6fe946
MD
3490 memset(p, 0, sizeof(*p));
3491
3dfa5721 3492 pcp = &p->pcp;
2caaad41 3493 pcp->count = 0;
2caaad41
CL
3494 pcp->high = 6 * batch;
3495 pcp->batch = max(1UL, 1 * batch);
5f8dcc21
MG
3496 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3497 INIT_LIST_HEAD(&pcp->lists[migratetype]);
2caaad41
CL
3498}
3499
8ad4b1fb
RS
3500/*
3501 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3502 * to the value high for the pageset p.
3503 */
3504
3505static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3506 unsigned long high)
3507{
3508 struct per_cpu_pages *pcp;
3509
3dfa5721 3510 pcp = &p->pcp;
8ad4b1fb
RS
3511 pcp->high = high;
3512 pcp->batch = max(1UL, high/4);
3513 if ((high/4) > (PAGE_SHIFT * 8))
3514 pcp->batch = PAGE_SHIFT * 8;
3515}
3516
319774e2
WF
3517static __meminit void setup_zone_pageset(struct zone *zone)
3518{
3519 int cpu;
3520
3521 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3522
3523 for_each_possible_cpu(cpu) {
3524 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3525
3526 setup_pageset(pcp, zone_batchsize(zone));
3527
3528 if (percpu_pagelist_fraction)
3529 setup_pagelist_highmark(pcp,
3530 (zone->present_pages /
3531 percpu_pagelist_fraction));
3532 }
3533}
3534
2caaad41 3535/*
99dcc3e5
CL
3536 * Allocate per cpu pagesets and initialize them.
3537 * Before this call only boot pagesets were available.
e7c8d5c9 3538 */
99dcc3e5 3539void __init setup_per_cpu_pageset(void)
e7c8d5c9 3540{
99dcc3e5 3541 struct zone *zone;
e7c8d5c9 3542
319774e2
WF
3543 for_each_populated_zone(zone)
3544 setup_zone_pageset(zone);
e7c8d5c9
CL
3545}
3546
577a32f6 3547static noinline __init_refok
cca448fe 3548int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
ed8ece2e
DH
3549{
3550 int i;
3551 struct pglist_data *pgdat = zone->zone_pgdat;
cca448fe 3552 size_t alloc_size;
ed8ece2e
DH
3553
3554 /*
3555 * The per-page waitqueue mechanism uses hashed waitqueues
3556 * per zone.
3557 */
02b694de
YG
3558 zone->wait_table_hash_nr_entries =
3559 wait_table_hash_nr_entries(zone_size_pages);
3560 zone->wait_table_bits =
3561 wait_table_bits(zone->wait_table_hash_nr_entries);
cca448fe
YG
3562 alloc_size = zone->wait_table_hash_nr_entries
3563 * sizeof(wait_queue_head_t);
3564
cd94b9db 3565 if (!slab_is_available()) {
cca448fe
YG
3566 zone->wait_table = (wait_queue_head_t *)
3567 alloc_bootmem_node(pgdat, alloc_size);
3568 } else {
3569 /*
3570 * This case means that a zone whose size was 0 gets new memory
3571 * via memory hot-add.
3572 * But it may be the case that a new node was hot-added. In
3573 * this case vmalloc() will not be able to use this new node's
3574 * memory - this wait_table must be initialized to use this new
3575 * node itself as well.
3576 * To use this new node's memory, further consideration will be
3577 * necessary.
3578 */
8691f3a7 3579 zone->wait_table = vmalloc(alloc_size);
cca448fe
YG
3580 }
3581 if (!zone->wait_table)
3582 return -ENOMEM;
ed8ece2e 3583
02b694de 3584 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
ed8ece2e 3585 init_waitqueue_head(zone->wait_table + i);
cca448fe
YG
3586
3587 return 0;
ed8ece2e
DH
3588}
3589
112067f0
SL
3590static int __zone_pcp_update(void *data)
3591{
3592 struct zone *zone = data;
3593 int cpu;
3594 unsigned long batch = zone_batchsize(zone), flags;
3595
2d30a1f6 3596 for_each_possible_cpu(cpu) {
112067f0
SL
3597 struct per_cpu_pageset *pset;
3598 struct per_cpu_pages *pcp;
3599
99dcc3e5 3600 pset = per_cpu_ptr(zone->pageset, cpu);
112067f0
SL
3601 pcp = &pset->pcp;
3602
3603 local_irq_save(flags);
5f8dcc21 3604 free_pcppages_bulk(zone, pcp->count, pcp);
112067f0
SL
3605 setup_pageset(pset, batch);
3606 local_irq_restore(flags);
3607 }
3608 return 0;
3609}
3610
3611void zone_pcp_update(struct zone *zone)
3612{
3613 stop_machine(__zone_pcp_update, zone, NULL);
3614}
3615
c09b4240 3616static __meminit void zone_pcp_init(struct zone *zone)
ed8ece2e 3617{
99dcc3e5
CL
3618 /*
3619 * per cpu subsystem is not up at this point. The following code
3620 * relies on the ability of the linker to provide the
3621 * offset of a (static) per cpu variable into the per cpu area.
3622 */
3623 zone->pageset = &boot_pageset;
ed8ece2e 3624
f5335c0f 3625 if (zone->present_pages)
99dcc3e5
CL
3626 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3627 zone->name, zone->present_pages,
3628 zone_batchsize(zone));
ed8ece2e
DH
3629}
3630
718127cc
YG
3631__meminit int init_currently_empty_zone(struct zone *zone,
3632 unsigned long zone_start_pfn,
a2f3aa02
DH
3633 unsigned long size,
3634 enum memmap_context context)
ed8ece2e
DH
3635{
3636 struct pglist_data *pgdat = zone->zone_pgdat;
cca448fe
YG
3637 int ret;
3638 ret = zone_wait_table_init(zone, size);
3639 if (ret)
3640 return ret;
ed8ece2e
DH
3641 pgdat->nr_zones = zone_idx(zone) + 1;
3642
ed8ece2e
DH
3643 zone->zone_start_pfn = zone_start_pfn;
3644
708614e6
MG
3645 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3646 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3647 pgdat->node_id,
3648 (unsigned long)zone_idx(zone),
3649 zone_start_pfn, (zone_start_pfn + size));
3650
1e548deb 3651 zone_init_free_lists(zone);
718127cc
YG
3652
3653 return 0;
ed8ece2e
DH
3654}
3655
c713216d
MG
3656#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3657/*
3658 * Basic iterator support. Return the first range of PFNs for a node
3659 * Note: nid == MAX_NUMNODES returns first region regardless of node
3660 */
a3142c8e 3661static int __meminit first_active_region_index_in_nid(int nid)
c713216d
MG
3662{
3663 int i;
3664
3665 for (i = 0; i < nr_nodemap_entries; i++)
3666 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3667 return i;
3668
3669 return -1;
3670}
3671
3672/*
3673 * Basic iterator support. Return the next active range of PFNs for a node
183ff22b 3674 * Note: nid == MAX_NUMNODES returns next region regardless of node
c713216d 3675 */
a3142c8e 3676static int __meminit next_active_region_index_in_nid(int index, int nid)
c713216d
MG
3677{
3678 for (index = index + 1; index < nr_nodemap_entries; index++)
3679 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3680 return index;
3681
3682 return -1;
3683}
3684
3685#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3686/*
3687 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3688 * Architectures may implement their own version but if add_active_range()
3689 * was used and there are no special requirements, this is a convenient
3690 * alternative
3691 */
f2dbcfa7 3692int __meminit __early_pfn_to_nid(unsigned long pfn)
c713216d
MG
3693{
3694 int i;
3695
3696 for (i = 0; i < nr_nodemap_entries; i++) {
3697 unsigned long start_pfn = early_node_map[i].start_pfn;
3698 unsigned long end_pfn = early_node_map[i].end_pfn;
3699
3700 if (start_pfn <= pfn && pfn < end_pfn)
3701 return early_node_map[i].nid;
3702 }
cc2559bc
KH
3703 /* This is a memory hole */
3704 return -1;
c713216d
MG
3705}
3706#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3707
f2dbcfa7
KH
3708int __meminit early_pfn_to_nid(unsigned long pfn)
3709{
cc2559bc
KH
3710 int nid;
3711
3712 nid = __early_pfn_to_nid(pfn);
3713 if (nid >= 0)
3714 return nid;
3715 /* just returns 0 */
3716 return 0;
f2dbcfa7
KH
3717}
3718
cc2559bc
KH
3719#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3720bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3721{
3722 int nid;
3723
3724 nid = __early_pfn_to_nid(pfn);
3725 if (nid >= 0 && nid != node)
3726 return false;
3727 return true;
3728}
3729#endif
f2dbcfa7 3730
c713216d
MG
3731/* Basic iterator support to walk early_node_map[] */
3732#define for_each_active_range_index_in_nid(i, nid) \
3733 for (i = first_active_region_index_in_nid(nid); i != -1; \
3734 i = next_active_region_index_in_nid(i, nid))
3735
3736/**
3737 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
88ca3b94
RD
3738 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3739 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
c713216d
MG
3740 *
3741 * If an architecture guarantees that all ranges registered with
3742 * add_active_ranges() contain no holes and may be freed, this
3743 * this function may be used instead of calling free_bootmem() manually.
3744 */
3745void __init free_bootmem_with_active_regions(int nid,
3746 unsigned long max_low_pfn)
3747{
3748 int i;
3749
3750 for_each_active_range_index_in_nid(i, nid) {
3751 unsigned long size_pages = 0;
3752 unsigned long end_pfn = early_node_map[i].end_pfn;
3753
3754 if (early_node_map[i].start_pfn >= max_low_pfn)
3755 continue;
3756
3757 if (end_pfn > max_low_pfn)
3758 end_pfn = max_low_pfn;
3759
3760 size_pages = end_pfn - early_node_map[i].start_pfn;
3761 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3762 PFN_PHYS(early_node_map[i].start_pfn),
3763 size_pages << PAGE_SHIFT);
3764 }
3765}
3766
edbe7d23 3767#ifdef CONFIG_HAVE_MEMBLOCK
cc289894
YL
3768/*
3769 * Basic iterator support. Return the last range of PFNs for a node
3770 * Note: nid == MAX_NUMNODES returns last region regardless of node
3771 */
3772static int __meminit last_active_region_index_in_nid(int nid)
3773{
3774 int i;
3775
3776 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3777 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3778 return i;
3779
3780 return -1;
3781}
3782
3783/*
3784 * Basic iterator support. Return the previous active range of PFNs for a node
3785 * Note: nid == MAX_NUMNODES returns next region regardless of node
3786 */
3787static int __meminit previous_active_region_index_in_nid(int index, int nid)
3788{
3789 for (index = index - 1; index >= 0; index--)
3790 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3791 return index;
3792
3793 return -1;
3794}
3795
3796#define for_each_active_range_index_in_nid_reverse(i, nid) \
3797 for (i = last_active_region_index_in_nid(nid); i != -1; \
3798 i = previous_active_region_index_in_nid(i, nid))
3799
edbe7d23
YL
3800u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3801 u64 goal, u64 limit)
3802{
3803 int i;
3804
3805 /* Need to go over early_node_map to find out good range for node */
1a4a678b 3806 for_each_active_range_index_in_nid_reverse(i, nid) {
edbe7d23
YL
3807 u64 addr;
3808 u64 ei_start, ei_last;
3809 u64 final_start, final_end;
3810
3811 ei_last = early_node_map[i].end_pfn;
3812 ei_last <<= PAGE_SHIFT;
3813 ei_start = early_node_map[i].start_pfn;
3814 ei_start <<= PAGE_SHIFT;
3815
3816 final_start = max(ei_start, goal);
3817 final_end = min(ei_last, limit);
3818
3819 if (final_start >= final_end)
3820 continue;
3821
3822 addr = memblock_find_in_range(final_start, final_end, size, align);
3823
3824 if (addr == MEMBLOCK_ERROR)
3825 continue;
3826
3827 return addr;
3828 }
3829
3830 return MEMBLOCK_ERROR;
3831}
3832#endif
3833
08677214
YL
3834int __init add_from_early_node_map(struct range *range, int az,
3835 int nr_range, int nid)
3836{
3837 int i;
3838 u64 start, end;
3839
3840 /* need to go over early_node_map to find out good range for node */
3841 for_each_active_range_index_in_nid(i, nid) {
3842 start = early_node_map[i].start_pfn;
3843 end = early_node_map[i].end_pfn;
3844 nr_range = add_range(range, az, nr_range, start, end);
3845 }
3846 return nr_range;
3847}
3848
b5bc6c0e
YL
3849void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3850{
3851 int i;
d52d53b8 3852 int ret;
b5bc6c0e 3853
d52d53b8
YL
3854 for_each_active_range_index_in_nid(i, nid) {
3855 ret = work_fn(early_node_map[i].start_pfn,
3856 early_node_map[i].end_pfn, data);
3857 if (ret)
3858 break;
3859 }
b5bc6c0e 3860}
c713216d
MG
3861/**
3862 * sparse_memory_present_with_active_regions - Call memory_present for each active range
88ca3b94 3863 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
c713216d
MG
3864 *
3865 * If an architecture guarantees that all ranges registered with
3866 * add_active_ranges() contain no holes and may be freed, this
88ca3b94 3867 * function may be used instead of calling memory_present() manually.
c713216d
MG
3868 */
3869void __init sparse_memory_present_with_active_regions(int nid)
3870{
3871 int i;
3872
3873 for_each_active_range_index_in_nid(i, nid)
3874 memory_present(early_node_map[i].nid,
3875 early_node_map[i].start_pfn,
3876 early_node_map[i].end_pfn);
3877}
3878
3879/**
3880 * get_pfn_range_for_nid - Return the start and end page frames for a node
88ca3b94
RD
3881 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3882 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3883 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
c713216d
MG
3884 *
3885 * It returns the start and end page frame of a node based on information
3886 * provided by an arch calling add_active_range(). If called for a node
3887 * with no available memory, a warning is printed and the start and end
88ca3b94 3888 * PFNs will be 0.
c713216d 3889 */
a3142c8e 3890void __meminit get_pfn_range_for_nid(unsigned int nid,
c713216d
MG
3891 unsigned long *start_pfn, unsigned long *end_pfn)
3892{
3893 int i;
3894 *start_pfn = -1UL;
3895 *end_pfn = 0;
3896
3897 for_each_active_range_index_in_nid(i, nid) {
3898 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3899 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3900 }
3901
633c0666 3902 if (*start_pfn == -1UL)
c713216d 3903 *start_pfn = 0;
c713216d
MG
3904}
3905
2a1e274a
MG
3906/*
3907 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3908 * assumption is made that zones within a node are ordered in monotonic
3909 * increasing memory addresses so that the "highest" populated zone is used
3910 */
b69a7288 3911static void __init find_usable_zone_for_movable(void)
2a1e274a
MG
3912{
3913 int zone_index;
3914 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3915 if (zone_index == ZONE_MOVABLE)
3916 continue;
3917
3918 if (arch_zone_highest_possible_pfn[zone_index] >
3919 arch_zone_lowest_possible_pfn[zone_index])
3920 break;
3921 }
3922
3923 VM_BUG_ON(zone_index == -1);
3924 movable_zone = zone_index;
3925}
3926
3927/*
3928 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
25985edc 3929 * because it is sized independent of architecture. Unlike the other zones,
2a1e274a
MG
3930 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3931 * in each node depending on the size of each node and how evenly kernelcore
3932 * is distributed. This helper function adjusts the zone ranges
3933 * provided by the architecture for a given node by using the end of the
3934 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3935 * zones within a node are in order of monotonic increases memory addresses
3936 */
b69a7288 3937static void __meminit adjust_zone_range_for_zone_movable(int nid,
2a1e274a
MG
3938 unsigned long zone_type,
3939 unsigned long node_start_pfn,
3940 unsigned long node_end_pfn,
3941 unsigned long *zone_start_pfn,
3942 unsigned long *zone_end_pfn)
3943{
3944 /* Only adjust if ZONE_MOVABLE is on this node */
3945 if (zone_movable_pfn[nid]) {
3946 /* Size ZONE_MOVABLE */
3947 if (zone_type == ZONE_MOVABLE) {
3948 *zone_start_pfn = zone_movable_pfn[nid];
3949 *zone_end_pfn = min(node_end_pfn,
3950 arch_zone_highest_possible_pfn[movable_zone]);
3951
3952 /* Adjust for ZONE_MOVABLE starting within this range */
3953 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3954 *zone_end_pfn > zone_movable_pfn[nid]) {
3955 *zone_end_pfn = zone_movable_pfn[nid];
3956
3957 /* Check if this whole range is within ZONE_MOVABLE */
3958 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3959 *zone_start_pfn = *zone_end_pfn;
3960 }
3961}
3962
c713216d
MG
3963/*
3964 * Return the number of pages a zone spans in a node, including holes
3965 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3966 */
6ea6e688 3967static unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d
MG
3968 unsigned long zone_type,
3969 unsigned long *ignored)
3970{
3971 unsigned long node_start_pfn, node_end_pfn;
3972 unsigned long zone_start_pfn, zone_end_pfn;
3973
3974 /* Get the start and end of the node and zone */
3975 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3976 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3977 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2a1e274a
MG
3978 adjust_zone_range_for_zone_movable(nid, zone_type,
3979 node_start_pfn, node_end_pfn,
3980 &zone_start_pfn, &zone_end_pfn);
c713216d
MG
3981
3982 /* Check that this node has pages within the zone's required range */
3983 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3984 return 0;
3985
3986 /* Move the zone boundaries inside the node if necessary */
3987 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3988 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3989
3990 /* Return the spanned pages */
3991 return zone_end_pfn - zone_start_pfn;
3992}
3993
3994/*
3995 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
88ca3b94 3996 * then all holes in the requested range will be accounted for.
c713216d 3997 */
32996250 3998unsigned long __meminit __absent_pages_in_range(int nid,
c713216d
MG
3999 unsigned long range_start_pfn,
4000 unsigned long range_end_pfn)
4001{
4002 int i = 0;
4003 unsigned long prev_end_pfn = 0, hole_pages = 0;
4004 unsigned long start_pfn;
4005
4006 /* Find the end_pfn of the first active range of pfns in the node */
4007 i = first_active_region_index_in_nid(nid);
4008 if (i == -1)
4009 return 0;
4010
b5445f95
MG
4011 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4012
9c7cd687
MG
4013 /* Account for ranges before physical memory on this node */
4014 if (early_node_map[i].start_pfn > range_start_pfn)
b5445f95 4015 hole_pages = prev_end_pfn - range_start_pfn;
c713216d
MG
4016
4017 /* Find all holes for the zone within the node */
4018 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4019
4020 /* No need to continue if prev_end_pfn is outside the zone */
4021 if (prev_end_pfn >= range_end_pfn)
4022 break;
4023
4024 /* Make sure the end of the zone is not within the hole */
4025 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4026 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4027
4028 /* Update the hole size cound and move on */
4029 if (start_pfn > range_start_pfn) {
4030 BUG_ON(prev_end_pfn > start_pfn);
4031 hole_pages += start_pfn - prev_end_pfn;
4032 }
4033 prev_end_pfn = early_node_map[i].end_pfn;
4034 }
4035
9c7cd687
MG
4036 /* Account for ranges past physical memory on this node */
4037 if (range_end_pfn > prev_end_pfn)
0c6cb974 4038 hole_pages += range_end_pfn -
9c7cd687
MG
4039 max(range_start_pfn, prev_end_pfn);
4040
c713216d
MG
4041 return hole_pages;
4042}
4043
4044/**
4045 * absent_pages_in_range - Return number of page frames in holes within a range
4046 * @start_pfn: The start PFN to start searching for holes
4047 * @end_pfn: The end PFN to stop searching for holes
4048 *
88ca3b94 4049 * It returns the number of pages frames in memory holes within a range.
c713216d
MG
4050 */
4051unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4052 unsigned long end_pfn)
4053{
4054 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4055}
4056
4057/* Return the number of page frames in holes in a zone on a node */
6ea6e688 4058static unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d
MG
4059 unsigned long zone_type,
4060 unsigned long *ignored)
4061{
9c7cd687
MG
4062 unsigned long node_start_pfn, node_end_pfn;
4063 unsigned long zone_start_pfn, zone_end_pfn;
4064
4065 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4066 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4067 node_start_pfn);
4068 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4069 node_end_pfn);
4070
2a1e274a
MG
4071 adjust_zone_range_for_zone_movable(nid, zone_type,
4072 node_start_pfn, node_end_pfn,
4073 &zone_start_pfn, &zone_end_pfn);
9c7cd687 4074 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
c713216d 4075}
0e0b864e 4076
c713216d 4077#else
6ea6e688 4078static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
c713216d
MG
4079 unsigned long zone_type,
4080 unsigned long *zones_size)
4081{
4082 return zones_size[zone_type];
4083}
4084
6ea6e688 4085static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
c713216d
MG
4086 unsigned long zone_type,
4087 unsigned long *zholes_size)
4088{
4089 if (!zholes_size)
4090 return 0;
4091
4092 return zholes_size[zone_type];
4093}
0e0b864e 4094
c713216d
MG
4095#endif
4096
a3142c8e 4097static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
c713216d
MG
4098 unsigned long *zones_size, unsigned long *zholes_size)
4099{
4100 unsigned long realtotalpages, totalpages = 0;
4101 enum zone_type i;
4102
4103 for (i = 0; i < MAX_NR_ZONES; i++)
4104 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4105 zones_size);
4106 pgdat->node_spanned_pages = totalpages;
4107
4108 realtotalpages = totalpages;
4109 for (i = 0; i < MAX_NR_ZONES; i++)
4110 realtotalpages -=
4111 zone_absent_pages_in_node(pgdat->node_id, i,
4112 zholes_size);
4113 pgdat->node_present_pages = realtotalpages;
4114 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4115 realtotalpages);
4116}
4117
835c134e
MG
4118#ifndef CONFIG_SPARSEMEM
4119/*
4120 * Calculate the size of the zone->blockflags rounded to an unsigned long
d9c23400
MG
4121 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4122 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
835c134e
MG
4123 * round what is now in bits to nearest long in bits, then return it in
4124 * bytes.
4125 */
4126static unsigned long __init usemap_size(unsigned long zonesize)
4127{
4128 unsigned long usemapsize;
4129
d9c23400
MG
4130 usemapsize = roundup(zonesize, pageblock_nr_pages);
4131 usemapsize = usemapsize >> pageblock_order;
835c134e
MG
4132 usemapsize *= NR_PAGEBLOCK_BITS;
4133 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4134
4135 return usemapsize / 8;
4136}
4137
4138static void __init setup_usemap(struct pglist_data *pgdat,
4139 struct zone *zone, unsigned long zonesize)
4140{
4141 unsigned long usemapsize = usemap_size(zonesize);
4142 zone->pageblock_flags = NULL;
58a01a45 4143 if (usemapsize)
835c134e 4144 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
835c134e
MG
4145}
4146#else
fa9f90be 4147static inline void setup_usemap(struct pglist_data *pgdat,
835c134e
MG
4148 struct zone *zone, unsigned long zonesize) {}
4149#endif /* CONFIG_SPARSEMEM */
4150
d9c23400 4151#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
ba72cb8c
MG
4152
4153/* Return a sensible default order for the pageblock size. */
4154static inline int pageblock_default_order(void)
4155{
4156 if (HPAGE_SHIFT > PAGE_SHIFT)
4157 return HUGETLB_PAGE_ORDER;
4158
4159 return MAX_ORDER-1;
4160}
4161
d9c23400
MG
4162/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4163static inline void __init set_pageblock_order(unsigned int order)
4164{
4165 /* Check that pageblock_nr_pages has not already been setup */
4166 if (pageblock_order)
4167 return;
4168
4169 /*
4170 * Assume the largest contiguous order of interest is a huge page.
4171 * This value may be variable depending on boot parameters on IA64
4172 */
4173 pageblock_order = order;
4174}
4175#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4176
ba72cb8c
MG
4177/*
4178 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4179 * and pageblock_default_order() are unused as pageblock_order is set
4180 * at compile-time. See include/linux/pageblock-flags.h for the values of
4181 * pageblock_order based on the kernel config
4182 */
4183static inline int pageblock_default_order(unsigned int order)
4184{
4185 return MAX_ORDER-1;
4186}
d9c23400
MG
4187#define set_pageblock_order(x) do {} while (0)
4188
4189#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4190
1da177e4
LT
4191/*
4192 * Set up the zone data structures:
4193 * - mark all pages reserved
4194 * - mark all memory queues empty
4195 * - clear the memory bitmaps
4196 */
b5a0e011 4197static void __paginginit free_area_init_core(struct pglist_data *pgdat,
1da177e4
LT
4198 unsigned long *zones_size, unsigned long *zholes_size)
4199{
2f1b6248 4200 enum zone_type j;
ed8ece2e 4201 int nid = pgdat->node_id;
1da177e4 4202 unsigned long zone_start_pfn = pgdat->node_start_pfn;
718127cc 4203 int ret;
1da177e4 4204
208d54e5 4205 pgdat_resize_init(pgdat);
1da177e4
LT
4206 pgdat->nr_zones = 0;
4207 init_waitqueue_head(&pgdat->kswapd_wait);
4208 pgdat->kswapd_max_order = 0;
52d4b9ac 4209 pgdat_page_cgroup_init(pgdat);
1da177e4
LT
4210
4211 for (j = 0; j < MAX_NR_ZONES; j++) {
4212 struct zone *zone = pgdat->node_zones + j;
0e0b864e 4213 unsigned long size, realsize, memmap_pages;
b69408e8 4214 enum lru_list l;
1da177e4 4215
c713216d
MG
4216 size = zone_spanned_pages_in_node(nid, j, zones_size);
4217 realsize = size - zone_absent_pages_in_node(nid, j,
4218 zholes_size);
1da177e4 4219
0e0b864e
MG
4220 /*
4221 * Adjust realsize so that it accounts for how much memory
4222 * is used by this zone for memmap. This affects the watermark
4223 * and per-cpu initialisations
4224 */
f7232154
JW
4225 memmap_pages =
4226 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
0e0b864e
MG
4227 if (realsize >= memmap_pages) {
4228 realsize -= memmap_pages;
5594c8c8
YL
4229 if (memmap_pages)
4230 printk(KERN_DEBUG
4231 " %s zone: %lu pages used for memmap\n",
4232 zone_names[j], memmap_pages);
0e0b864e
MG
4233 } else
4234 printk(KERN_WARNING
4235 " %s zone: %lu pages exceeds realsize %lu\n",
4236 zone_names[j], memmap_pages, realsize);
4237
6267276f
CL
4238 /* Account for reserved pages */
4239 if (j == 0 && realsize > dma_reserve) {
0e0b864e 4240 realsize -= dma_reserve;
d903ef9f 4241 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
6267276f 4242 zone_names[0], dma_reserve);
0e0b864e
MG
4243 }
4244
98d2b0eb 4245 if (!is_highmem_idx(j))
1da177e4
LT
4246 nr_kernel_pages += realsize;
4247 nr_all_pages += realsize;
4248
4249 zone->spanned_pages = size;
4250 zone->present_pages = realsize;
9614634f 4251#ifdef CONFIG_NUMA
d5f541ed 4252 zone->node = nid;
8417bba4 4253 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
9614634f 4254 / 100;
0ff38490 4255 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
9614634f 4256#endif
1da177e4
LT
4257 zone->name = zone_names[j];
4258 spin_lock_init(&zone->lock);
4259 spin_lock_init(&zone->lru_lock);
bdc8cb98 4260 zone_seqlock_init(zone);
1da177e4 4261 zone->zone_pgdat = pgdat;
1da177e4 4262
ed8ece2e 4263 zone_pcp_init(zone);
b69408e8
CL
4264 for_each_lru(l) {
4265 INIT_LIST_HEAD(&zone->lru[l].list);
f8629631 4266 zone->reclaim_stat.nr_saved_scan[l] = 0;
b69408e8 4267 }
6e901571
KM
4268 zone->reclaim_stat.recent_rotated[0] = 0;
4269 zone->reclaim_stat.recent_rotated[1] = 0;
4270 zone->reclaim_stat.recent_scanned[0] = 0;
4271 zone->reclaim_stat.recent_scanned[1] = 0;
2244b95a 4272 zap_zone_vm_stats(zone);
e815af95 4273 zone->flags = 0;
1da177e4
LT
4274 if (!size)
4275 continue;
4276
ba72cb8c 4277 set_pageblock_order(pageblock_default_order());
835c134e 4278 setup_usemap(pgdat, zone, size);
a2f3aa02
DH
4279 ret = init_currently_empty_zone(zone, zone_start_pfn,
4280 size, MEMMAP_EARLY);
718127cc 4281 BUG_ON(ret);
76cdd58e 4282 memmap_init(size, nid, j, zone_start_pfn);
1da177e4 4283 zone_start_pfn += size;
1da177e4
LT
4284 }
4285}
4286
577a32f6 4287static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
1da177e4 4288{
1da177e4
LT
4289 /* Skip empty nodes */
4290 if (!pgdat->node_spanned_pages)
4291 return;
4292
d41dee36 4293#ifdef CONFIG_FLAT_NODE_MEM_MAP
1da177e4
LT
4294 /* ia64 gets its own node_mem_map, before this, without bootmem */
4295 if (!pgdat->node_mem_map) {
e984bb43 4296 unsigned long size, start, end;
d41dee36
AW
4297 struct page *map;
4298
e984bb43
BP
4299 /*
4300 * The zone's endpoints aren't required to be MAX_ORDER
4301 * aligned but the node_mem_map endpoints must be in order
4302 * for the buddy allocator to function correctly.
4303 */
4304 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4305 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4306 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4307 size = (end - start) * sizeof(struct page);
6f167ec7
DH
4308 map = alloc_remap(pgdat->node_id, size);
4309 if (!map)
4310 map = alloc_bootmem_node(pgdat, size);
e984bb43 4311 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
1da177e4 4312 }
12d810c1 4313#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
4314 /*
4315 * With no DISCONTIG, the global mem_map is just set as node 0's
4316 */
c713216d 4317 if (pgdat == NODE_DATA(0)) {
1da177e4 4318 mem_map = NODE_DATA(0)->node_mem_map;
c713216d
MG
4319#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4320 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
467bc461 4321 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
c713216d
MG
4322#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4323 }
1da177e4 4324#endif
d41dee36 4325#endif /* CONFIG_FLAT_NODE_MEM_MAP */
1da177e4
LT
4326}
4327
9109fb7b
JW
4328void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4329 unsigned long node_start_pfn, unsigned long *zholes_size)
1da177e4 4330{
9109fb7b
JW
4331 pg_data_t *pgdat = NODE_DATA(nid);
4332
1da177e4
LT
4333 pgdat->node_id = nid;
4334 pgdat->node_start_pfn = node_start_pfn;
c713216d 4335 calculate_node_totalpages(pgdat, zones_size, zholes_size);
1da177e4
LT
4336
4337 alloc_node_mem_map(pgdat);
e8c27ac9
YL
4338#ifdef CONFIG_FLAT_NODE_MEM_MAP
4339 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4340 nid, (unsigned long)pgdat,
4341 (unsigned long)pgdat->node_mem_map);
4342#endif
1da177e4
LT
4343
4344 free_area_init_core(pgdat, zones_size, zholes_size);
4345}
4346
c713216d 4347#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
418508c1
MS
4348
4349#if MAX_NUMNODES > 1
4350/*
4351 * Figure out the number of possible node ids.
4352 */
4353static void __init setup_nr_node_ids(void)
4354{
4355 unsigned int node;
4356 unsigned int highest = 0;
4357
4358 for_each_node_mask(node, node_possible_map)
4359 highest = node;
4360 nr_node_ids = highest + 1;
4361}
4362#else
4363static inline void setup_nr_node_ids(void)
4364{
4365}
4366#endif
4367
c713216d
MG
4368/**
4369 * add_active_range - Register a range of PFNs backed by physical memory
4370 * @nid: The node ID the range resides on
4371 * @start_pfn: The start PFN of the available physical memory
4372 * @end_pfn: The end PFN of the available physical memory
4373 *
4374 * These ranges are stored in an early_node_map[] and later used by
4375 * free_area_init_nodes() to calculate zone sizes and holes. If the
4376 * range spans a memory hole, it is up to the architecture to ensure
4377 * the memory is not freed by the bootmem allocator. If possible
4378 * the range being registered will be merged with existing ranges.
4379 */
4380void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4381 unsigned long end_pfn)
4382{
4383 int i;
4384
6b74ab97
MG
4385 mminit_dprintk(MMINIT_TRACE, "memory_register",
4386 "Entering add_active_range(%d, %#lx, %#lx) "
4387 "%d entries of %d used\n",
4388 nid, start_pfn, end_pfn,
4389 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
c713216d 4390
2dbb51c4
MG
4391 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4392
c713216d
MG
4393 /* Merge with existing active regions if possible */
4394 for (i = 0; i < nr_nodemap_entries; i++) {
4395 if (early_node_map[i].nid != nid)
4396 continue;
4397
4398 /* Skip if an existing region covers this new one */
4399 if (start_pfn >= early_node_map[i].start_pfn &&
4400 end_pfn <= early_node_map[i].end_pfn)
4401 return;
4402
4403 /* Merge forward if suitable */
4404 if (start_pfn <= early_node_map[i].end_pfn &&
4405 end_pfn > early_node_map[i].end_pfn) {
4406 early_node_map[i].end_pfn = end_pfn;
4407 return;
4408 }
4409
4410 /* Merge backward if suitable */
d2dbe08d 4411 if (start_pfn < early_node_map[i].start_pfn &&
c713216d
MG
4412 end_pfn >= early_node_map[i].start_pfn) {
4413 early_node_map[i].start_pfn = start_pfn;
4414 return;
4415 }
4416 }
4417
4418 /* Check that early_node_map is large enough */
4419 if (i >= MAX_ACTIVE_REGIONS) {
4420 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4421 MAX_ACTIVE_REGIONS);
4422 return;
4423 }
4424
4425 early_node_map[i].nid = nid;
4426 early_node_map[i].start_pfn = start_pfn;
4427 early_node_map[i].end_pfn = end_pfn;
4428 nr_nodemap_entries = i + 1;
4429}
4430
4431/**
cc1050ba 4432 * remove_active_range - Shrink an existing registered range of PFNs
c713216d 4433 * @nid: The node id the range is on that should be shrunk
cc1050ba
YL
4434 * @start_pfn: The new PFN of the range
4435 * @end_pfn: The new PFN of the range
c713216d
MG
4436 *
4437 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
cc1a9d86
YL
4438 * The map is kept near the end physical page range that has already been
4439 * registered. This function allows an arch to shrink an existing registered
4440 * range.
c713216d 4441 */
cc1050ba
YL
4442void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4443 unsigned long end_pfn)
c713216d 4444{
cc1a9d86
YL
4445 int i, j;
4446 int removed = 0;
c713216d 4447
cc1050ba
YL
4448 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4449 nid, start_pfn, end_pfn);
4450
c713216d 4451 /* Find the old active region end and shrink */
cc1a9d86 4452 for_each_active_range_index_in_nid(i, nid) {
cc1050ba
YL
4453 if (early_node_map[i].start_pfn >= start_pfn &&
4454 early_node_map[i].end_pfn <= end_pfn) {
cc1a9d86 4455 /* clear it */
cc1050ba 4456 early_node_map[i].start_pfn = 0;
cc1a9d86
YL
4457 early_node_map[i].end_pfn = 0;
4458 removed = 1;
4459 continue;
4460 }
cc1050ba
YL
4461 if (early_node_map[i].start_pfn < start_pfn &&
4462 early_node_map[i].end_pfn > start_pfn) {
4463 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4464 early_node_map[i].end_pfn = start_pfn;
4465 if (temp_end_pfn > end_pfn)
4466 add_active_range(nid, end_pfn, temp_end_pfn);
4467 continue;
4468 }
4469 if (early_node_map[i].start_pfn >= start_pfn &&
4470 early_node_map[i].end_pfn > end_pfn &&
4471 early_node_map[i].start_pfn < end_pfn) {
4472 early_node_map[i].start_pfn = end_pfn;
cc1a9d86 4473 continue;
c713216d 4474 }
cc1a9d86
YL
4475 }
4476
4477 if (!removed)
4478 return;
4479
4480 /* remove the blank ones */
4481 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4482 if (early_node_map[i].nid != nid)
4483 continue;
4484 if (early_node_map[i].end_pfn)
4485 continue;
4486 /* we found it, get rid of it */
4487 for (j = i; j < nr_nodemap_entries - 1; j++)
4488 memcpy(&early_node_map[j], &early_node_map[j+1],
4489 sizeof(early_node_map[j]));
4490 j = nr_nodemap_entries - 1;
4491 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4492 nr_nodemap_entries--;
4493 }
c713216d
MG
4494}
4495
4496/**
4497 * remove_all_active_ranges - Remove all currently registered regions
88ca3b94 4498 *
c713216d
MG
4499 * During discovery, it may be found that a table like SRAT is invalid
4500 * and an alternative discovery method must be used. This function removes
4501 * all currently registered regions.
4502 */
88ca3b94 4503void __init remove_all_active_ranges(void)
c713216d
MG
4504{
4505 memset(early_node_map, 0, sizeof(early_node_map));
4506 nr_nodemap_entries = 0;
4507}
4508
4509/* Compare two active node_active_regions */
4510static int __init cmp_node_active_region(const void *a, const void *b)
4511{
4512 struct node_active_region *arange = (struct node_active_region *)a;
4513 struct node_active_region *brange = (struct node_active_region *)b;
4514
4515 /* Done this way to avoid overflows */
4516 if (arange->start_pfn > brange->start_pfn)
4517 return 1;
4518 if (arange->start_pfn < brange->start_pfn)
4519 return -1;
4520
4521 return 0;
4522}
4523
4524/* sort the node_map by start_pfn */
32996250 4525void __init sort_node_map(void)
c713216d
MG
4526{
4527 sort(early_node_map, (size_t)nr_nodemap_entries,
4528 sizeof(struct node_active_region),
4529 cmp_node_active_region, NULL);
4530}
4531
a6af2bc3 4532/* Find the lowest pfn for a node */
b69a7288 4533static unsigned long __init find_min_pfn_for_node(int nid)
c713216d
MG
4534{
4535 int i;
a6af2bc3 4536 unsigned long min_pfn = ULONG_MAX;
1abbfb41 4537
c713216d
MG
4538 /* Assuming a sorted map, the first range found has the starting pfn */
4539 for_each_active_range_index_in_nid(i, nid)
a6af2bc3 4540 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
c713216d 4541
a6af2bc3
MG
4542 if (min_pfn == ULONG_MAX) {
4543 printk(KERN_WARNING
2bc0d261 4544 "Could not find start_pfn for node %d\n", nid);
a6af2bc3
MG
4545 return 0;
4546 }
4547
4548 return min_pfn;
c713216d
MG
4549}
4550
4551/**
4552 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4553 *
4554 * It returns the minimum PFN based on information provided via
88ca3b94 4555 * add_active_range().
c713216d
MG
4556 */
4557unsigned long __init find_min_pfn_with_active_regions(void)
4558{
4559 return find_min_pfn_for_node(MAX_NUMNODES);
4560}
4561
37b07e41
LS
4562/*
4563 * early_calculate_totalpages()
4564 * Sum pages in active regions for movable zone.
4565 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4566 */
484f51f8 4567static unsigned long __init early_calculate_totalpages(void)
7e63efef
MG
4568{
4569 int i;
4570 unsigned long totalpages = 0;
4571
37b07e41
LS
4572 for (i = 0; i < nr_nodemap_entries; i++) {
4573 unsigned long pages = early_node_map[i].end_pfn -
7e63efef 4574 early_node_map[i].start_pfn;
37b07e41
LS
4575 totalpages += pages;
4576 if (pages)
4577 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4578 }
4579 return totalpages;
7e63efef
MG
4580}
4581
2a1e274a
MG
4582/*
4583 * Find the PFN the Movable zone begins in each node. Kernel memory
4584 * is spread evenly between nodes as long as the nodes have enough
4585 * memory. When they don't, some nodes will have more kernelcore than
4586 * others
4587 */
b69a7288 4588static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
2a1e274a
MG
4589{
4590 int i, nid;
4591 unsigned long usable_startpfn;
4592 unsigned long kernelcore_node, kernelcore_remaining;
66918dcd
YL
4593 /* save the state before borrow the nodemask */
4594 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
37b07e41
LS
4595 unsigned long totalpages = early_calculate_totalpages();
4596 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
2a1e274a 4597
7e63efef
MG
4598 /*
4599 * If movablecore was specified, calculate what size of
4600 * kernelcore that corresponds so that memory usable for
4601 * any allocation type is evenly spread. If both kernelcore
4602 * and movablecore are specified, then the value of kernelcore
4603 * will be used for required_kernelcore if it's greater than
4604 * what movablecore would have allowed.
4605 */
4606 if (required_movablecore) {
7e63efef
MG
4607 unsigned long corepages;
4608
4609 /*
4610 * Round-up so that ZONE_MOVABLE is at least as large as what
4611 * was requested by the user
4612 */
4613 required_movablecore =
4614 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4615 corepages = totalpages - required_movablecore;
4616
4617 required_kernelcore = max(required_kernelcore, corepages);
4618 }
4619
2a1e274a
MG
4620 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4621 if (!required_kernelcore)
66918dcd 4622 goto out;
2a1e274a
MG
4623
4624 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4625 find_usable_zone_for_movable();
4626 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4627
4628restart:
4629 /* Spread kernelcore memory as evenly as possible throughout nodes */
4630 kernelcore_node = required_kernelcore / usable_nodes;
37b07e41 4631 for_each_node_state(nid, N_HIGH_MEMORY) {
2a1e274a
MG
4632 /*
4633 * Recalculate kernelcore_node if the division per node
4634 * now exceeds what is necessary to satisfy the requested
4635 * amount of memory for the kernel
4636 */
4637 if (required_kernelcore < kernelcore_node)
4638 kernelcore_node = required_kernelcore / usable_nodes;
4639
4640 /*
4641 * As the map is walked, we track how much memory is usable
4642 * by the kernel using kernelcore_remaining. When it is
4643 * 0, the rest of the node is usable by ZONE_MOVABLE
4644 */
4645 kernelcore_remaining = kernelcore_node;
4646
4647 /* Go through each range of PFNs within this node */
4648 for_each_active_range_index_in_nid(i, nid) {
4649 unsigned long start_pfn, end_pfn;
4650 unsigned long size_pages;
4651
4652 start_pfn = max(early_node_map[i].start_pfn,
4653 zone_movable_pfn[nid]);
4654 end_pfn = early_node_map[i].end_pfn;
4655 if (start_pfn >= end_pfn)
4656 continue;
4657
4658 /* Account for what is only usable for kernelcore */
4659 if (start_pfn < usable_startpfn) {
4660 unsigned long kernel_pages;
4661 kernel_pages = min(end_pfn, usable_startpfn)
4662 - start_pfn;
4663
4664 kernelcore_remaining -= min(kernel_pages,
4665 kernelcore_remaining);
4666 required_kernelcore -= min(kernel_pages,
4667 required_kernelcore);
4668
4669 /* Continue if range is now fully accounted */
4670 if (end_pfn <= usable_startpfn) {
4671
4672 /*
4673 * Push zone_movable_pfn to the end so
4674 * that if we have to rebalance
4675 * kernelcore across nodes, we will
4676 * not double account here
4677 */
4678 zone_movable_pfn[nid] = end_pfn;
4679 continue;
4680 }
4681 start_pfn = usable_startpfn;
4682 }
4683
4684 /*
4685 * The usable PFN range for ZONE_MOVABLE is from
4686 * start_pfn->end_pfn. Calculate size_pages as the
4687 * number of pages used as kernelcore
4688 */
4689 size_pages = end_pfn - start_pfn;
4690 if (size_pages > kernelcore_remaining)
4691 size_pages = kernelcore_remaining;
4692 zone_movable_pfn[nid] = start_pfn + size_pages;
4693
4694 /*
4695 * Some kernelcore has been met, update counts and
4696 * break if the kernelcore for this node has been
4697 * satisified
4698 */
4699 required_kernelcore -= min(required_kernelcore,
4700 size_pages);
4701 kernelcore_remaining -= size_pages;
4702 if (!kernelcore_remaining)
4703 break;
4704 }
4705 }
4706
4707 /*
4708 * If there is still required_kernelcore, we do another pass with one
4709 * less node in the count. This will push zone_movable_pfn[nid] further
4710 * along on the nodes that still have memory until kernelcore is
4711 * satisified
4712 */
4713 usable_nodes--;
4714 if (usable_nodes && required_kernelcore > usable_nodes)
4715 goto restart;
4716
4717 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4718 for (nid = 0; nid < MAX_NUMNODES; nid++)
4719 zone_movable_pfn[nid] =
4720 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
66918dcd
YL
4721
4722out:
4723 /* restore the node_state */
4724 node_states[N_HIGH_MEMORY] = saved_node_state;
2a1e274a
MG
4725}
4726
37b07e41
LS
4727/* Any regular memory on that node ? */
4728static void check_for_regular_memory(pg_data_t *pgdat)
4729{
4730#ifdef CONFIG_HIGHMEM
4731 enum zone_type zone_type;
4732
4733 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4734 struct zone *zone = &pgdat->node_zones[zone_type];
4735 if (zone->present_pages)
4736 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4737 }
4738#endif
4739}
4740
c713216d
MG
4741/**
4742 * free_area_init_nodes - Initialise all pg_data_t and zone data
88ca3b94 4743 * @max_zone_pfn: an array of max PFNs for each zone
c713216d
MG
4744 *
4745 * This will call free_area_init_node() for each active node in the system.
4746 * Using the page ranges provided by add_active_range(), the size of each
4747 * zone in each node and their holes is calculated. If the maximum PFN
4748 * between two adjacent zones match, it is assumed that the zone is empty.
4749 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4750 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4751 * starts where the previous one ended. For example, ZONE_DMA32 starts
4752 * at arch_max_dma_pfn.
4753 */
4754void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4755{
4756 unsigned long nid;
db99100d 4757 int i;
c713216d 4758
a6af2bc3
MG
4759 /* Sort early_node_map as initialisation assumes it is sorted */
4760 sort_node_map();
4761
c713216d
MG
4762 /* Record where the zone boundaries are */
4763 memset(arch_zone_lowest_possible_pfn, 0,
4764 sizeof(arch_zone_lowest_possible_pfn));
4765 memset(arch_zone_highest_possible_pfn, 0,
4766 sizeof(arch_zone_highest_possible_pfn));
4767 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4768 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4769 for (i = 1; i < MAX_NR_ZONES; i++) {
2a1e274a
MG
4770 if (i == ZONE_MOVABLE)
4771 continue;
c713216d
MG
4772 arch_zone_lowest_possible_pfn[i] =
4773 arch_zone_highest_possible_pfn[i-1];
4774 arch_zone_highest_possible_pfn[i] =
4775 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4776 }
2a1e274a
MG
4777 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4778 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4779
4780 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4781 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4782 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
c713216d 4783
c713216d
MG
4784 /* Print out the zone ranges */
4785 printk("Zone PFN ranges:\n");
2a1e274a
MG
4786 for (i = 0; i < MAX_NR_ZONES; i++) {
4787 if (i == ZONE_MOVABLE)
4788 continue;
72f0ba02
DR
4789 printk(" %-8s ", zone_names[i]);
4790 if (arch_zone_lowest_possible_pfn[i] ==
4791 arch_zone_highest_possible_pfn[i])
4792 printk("empty\n");
4793 else
4794 printk("%0#10lx -> %0#10lx\n",
c713216d
MG
4795 arch_zone_lowest_possible_pfn[i],
4796 arch_zone_highest_possible_pfn[i]);
2a1e274a
MG
4797 }
4798
4799 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4800 printk("Movable zone start PFN for each node\n");
4801 for (i = 0; i < MAX_NUMNODES; i++) {
4802 if (zone_movable_pfn[i])
4803 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4804 }
c713216d
MG
4805
4806 /* Print out the early_node_map[] */
4807 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4808 for (i = 0; i < nr_nodemap_entries; i++)
5dab8ec1 4809 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
c713216d
MG
4810 early_node_map[i].start_pfn,
4811 early_node_map[i].end_pfn);
4812
4813 /* Initialise every node */
708614e6 4814 mminit_verify_pageflags_layout();
8ef82866 4815 setup_nr_node_ids();
c713216d
MG
4816 for_each_online_node(nid) {
4817 pg_data_t *pgdat = NODE_DATA(nid);
9109fb7b 4818 free_area_init_node(nid, NULL,
c713216d 4819 find_min_pfn_for_node(nid), NULL);
37b07e41
LS
4820
4821 /* Any memory on that node */
4822 if (pgdat->node_present_pages)
4823 node_set_state(nid, N_HIGH_MEMORY);
4824 check_for_regular_memory(pgdat);
c713216d
MG
4825 }
4826}
2a1e274a 4827
7e63efef 4828static int __init cmdline_parse_core(char *p, unsigned long *core)
2a1e274a
MG
4829{
4830 unsigned long long coremem;
4831 if (!p)
4832 return -EINVAL;
4833
4834 coremem = memparse(p, &p);
7e63efef 4835 *core = coremem >> PAGE_SHIFT;
2a1e274a 4836
7e63efef 4837 /* Paranoid check that UL is enough for the coremem value */
2a1e274a
MG
4838 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4839
4840 return 0;
4841}
ed7ed365 4842
7e63efef
MG
4843/*
4844 * kernelcore=size sets the amount of memory for use for allocations that
4845 * cannot be reclaimed or migrated.
4846 */
4847static int __init cmdline_parse_kernelcore(char *p)
4848{
4849 return cmdline_parse_core(p, &required_kernelcore);
4850}
4851
4852/*
4853 * movablecore=size sets the amount of memory for use for allocations that
4854 * can be reclaimed or migrated.
4855 */
4856static int __init cmdline_parse_movablecore(char *p)
4857{
4858 return cmdline_parse_core(p, &required_movablecore);
4859}
4860
ed7ed365 4861early_param("kernelcore", cmdline_parse_kernelcore);
7e63efef 4862early_param("movablecore", cmdline_parse_movablecore);
ed7ed365 4863
c713216d
MG
4864#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4865
0e0b864e 4866/**
88ca3b94
RD
4867 * set_dma_reserve - set the specified number of pages reserved in the first zone
4868 * @new_dma_reserve: The number of pages to mark reserved
0e0b864e
MG
4869 *
4870 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4871 * In the DMA zone, a significant percentage may be consumed by kernel image
4872 * and other unfreeable allocations which can skew the watermarks badly. This
88ca3b94
RD
4873 * function may optionally be used to account for unfreeable pages in the
4874 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4875 * smaller per-cpu batchsize.
0e0b864e
MG
4876 */
4877void __init set_dma_reserve(unsigned long new_dma_reserve)
4878{
4879 dma_reserve = new_dma_reserve;
4880}
4881
1da177e4
LT
4882void __init free_area_init(unsigned long *zones_size)
4883{
9109fb7b 4884 free_area_init_node(0, zones_size,
1da177e4
LT
4885 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4886}
1da177e4 4887
1da177e4
LT
4888static int page_alloc_cpu_notify(struct notifier_block *self,
4889 unsigned long action, void *hcpu)
4890{
4891 int cpu = (unsigned long)hcpu;
1da177e4 4892
8bb78442 4893 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
9f8f2172
CL
4894 drain_pages(cpu);
4895
4896 /*
4897 * Spill the event counters of the dead processor
4898 * into the current processors event counters.
4899 * This artificially elevates the count of the current
4900 * processor.
4901 */
f8891e5e 4902 vm_events_fold_cpu(cpu);
9f8f2172
CL
4903
4904 /*
4905 * Zero the differential counters of the dead processor
4906 * so that the vm statistics are consistent.
4907 *
4908 * This is only okay since the processor is dead and cannot
4909 * race with what we are doing.
4910 */
2244b95a 4911 refresh_cpu_vm_stats(cpu);
1da177e4
LT
4912 }
4913 return NOTIFY_OK;
4914}
1da177e4
LT
4915
4916void __init page_alloc_init(void)
4917{
4918 hotcpu_notifier(page_alloc_cpu_notify, 0);
4919}
4920
cb45b0e9
HA
4921/*
4922 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4923 * or min_free_kbytes changes.
4924 */
4925static void calculate_totalreserve_pages(void)
4926{
4927 struct pglist_data *pgdat;
4928 unsigned long reserve_pages = 0;
2f6726e5 4929 enum zone_type i, j;
cb45b0e9
HA
4930
4931 for_each_online_pgdat(pgdat) {
4932 for (i = 0; i < MAX_NR_ZONES; i++) {
4933 struct zone *zone = pgdat->node_zones + i;
4934 unsigned long max = 0;
4935
4936 /* Find valid and maximum lowmem_reserve in the zone */
4937 for (j = i; j < MAX_NR_ZONES; j++) {
4938 if (zone->lowmem_reserve[j] > max)
4939 max = zone->lowmem_reserve[j];
4940 }
4941
41858966
MG
4942 /* we treat the high watermark as reserved pages. */
4943 max += high_wmark_pages(zone);
cb45b0e9
HA
4944
4945 if (max > zone->present_pages)
4946 max = zone->present_pages;
4947 reserve_pages += max;
4948 }
4949 }
4950 totalreserve_pages = reserve_pages;
4951}
4952
1da177e4
LT
4953/*
4954 * setup_per_zone_lowmem_reserve - called whenever
4955 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4956 * has a correct pages reserved value, so an adequate number of
4957 * pages are left in the zone after a successful __alloc_pages().
4958 */
4959static void setup_per_zone_lowmem_reserve(void)
4960{
4961 struct pglist_data *pgdat;
2f6726e5 4962 enum zone_type j, idx;
1da177e4 4963
ec936fc5 4964 for_each_online_pgdat(pgdat) {
1da177e4
LT
4965 for (j = 0; j < MAX_NR_ZONES; j++) {
4966 struct zone *zone = pgdat->node_zones + j;
4967 unsigned long present_pages = zone->present_pages;
4968
4969 zone->lowmem_reserve[j] = 0;
4970
2f6726e5
CL
4971 idx = j;
4972 while (idx) {
1da177e4
LT
4973 struct zone *lower_zone;
4974
2f6726e5
CL
4975 idx--;
4976
1da177e4
LT
4977 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4978 sysctl_lowmem_reserve_ratio[idx] = 1;
4979
4980 lower_zone = pgdat->node_zones + idx;
4981 lower_zone->lowmem_reserve[j] = present_pages /
4982 sysctl_lowmem_reserve_ratio[idx];
4983 present_pages += lower_zone->present_pages;
4984 }
4985 }
4986 }
cb45b0e9
HA
4987
4988 /* update totalreserve_pages */
4989 calculate_totalreserve_pages();
1da177e4
LT
4990}
4991
88ca3b94 4992/**
bc75d33f 4993 * setup_per_zone_wmarks - called when min_free_kbytes changes
bce7394a 4994 * or when memory is hot-{added|removed}
88ca3b94 4995 *
bc75d33f
MK
4996 * Ensures that the watermark[min,low,high] values for each zone are set
4997 * correctly with respect to min_free_kbytes.
1da177e4 4998 */
bc75d33f 4999void setup_per_zone_wmarks(void)
1da177e4
LT
5000{
5001 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5002 unsigned long lowmem_pages = 0;
5003 struct zone *zone;
5004 unsigned long flags;
5005
5006 /* Calculate total number of !ZONE_HIGHMEM pages */
5007 for_each_zone(zone) {
5008 if (!is_highmem(zone))
5009 lowmem_pages += zone->present_pages;
5010 }
5011
5012 for_each_zone(zone) {
ac924c60
AM
5013 u64 tmp;
5014
1125b4e3 5015 spin_lock_irqsave(&zone->lock, flags);
ac924c60
AM
5016 tmp = (u64)pages_min * zone->present_pages;
5017 do_div(tmp, lowmem_pages);
1da177e4
LT
5018 if (is_highmem(zone)) {
5019 /*
669ed175
NP
5020 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5021 * need highmem pages, so cap pages_min to a small
5022 * value here.
5023 *
41858966 5024 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
669ed175
NP
5025 * deltas controls asynch page reclaim, and so should
5026 * not be capped for highmem.
1da177e4
LT
5027 */
5028 int min_pages;
5029
5030 min_pages = zone->present_pages / 1024;
5031 if (min_pages < SWAP_CLUSTER_MAX)
5032 min_pages = SWAP_CLUSTER_MAX;
5033 if (min_pages > 128)
5034 min_pages = 128;
41858966 5035 zone->watermark[WMARK_MIN] = min_pages;
1da177e4 5036 } else {
669ed175
NP
5037 /*
5038 * If it's a lowmem zone, reserve a number of pages
1da177e4
LT
5039 * proportionate to the zone's size.
5040 */
41858966 5041 zone->watermark[WMARK_MIN] = tmp;
1da177e4
LT
5042 }
5043
41858966
MG
5044 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5045 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
56fd56b8 5046 setup_zone_migrate_reserve(zone);
1125b4e3 5047 spin_unlock_irqrestore(&zone->lock, flags);
1da177e4 5048 }
cb45b0e9
HA
5049
5050 /* update totalreserve_pages */
5051 calculate_totalreserve_pages();
1da177e4
LT
5052}
5053
55a4462a 5054/*
556adecb
RR
5055 * The inactive anon list should be small enough that the VM never has to
5056 * do too much work, but large enough that each inactive page has a chance
5057 * to be referenced again before it is swapped out.
5058 *
5059 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5060 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5061 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5062 * the anonymous pages are kept on the inactive list.
5063 *
5064 * total target max
5065 * memory ratio inactive anon
5066 * -------------------------------------
5067 * 10MB 1 5MB
5068 * 100MB 1 50MB
5069 * 1GB 3 250MB
5070 * 10GB 10 0.9GB
5071 * 100GB 31 3GB
5072 * 1TB 101 10GB
5073 * 10TB 320 32GB
5074 */
96cb4df5 5075void calculate_zone_inactive_ratio(struct zone *zone)
556adecb 5076{
96cb4df5 5077 unsigned int gb, ratio;
556adecb 5078
96cb4df5
MK
5079 /* Zone size in gigabytes */
5080 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5081 if (gb)
556adecb 5082 ratio = int_sqrt(10 * gb);
96cb4df5
MK
5083 else
5084 ratio = 1;
556adecb 5085
96cb4df5
MK
5086 zone->inactive_ratio = ratio;
5087}
556adecb 5088
96cb4df5
MK
5089static void __init setup_per_zone_inactive_ratio(void)
5090{
5091 struct zone *zone;
5092
5093 for_each_zone(zone)
5094 calculate_zone_inactive_ratio(zone);
556adecb
RR
5095}
5096
1da177e4
LT
5097/*
5098 * Initialise min_free_kbytes.
5099 *
5100 * For small machines we want it small (128k min). For large machines
5101 * we want it large (64MB max). But it is not linear, because network
5102 * bandwidth does not increase linearly with machine size. We use
5103 *
5104 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5105 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5106 *
5107 * which yields
5108 *
5109 * 16MB: 512k
5110 * 32MB: 724k
5111 * 64MB: 1024k
5112 * 128MB: 1448k
5113 * 256MB: 2048k
5114 * 512MB: 2896k
5115 * 1024MB: 4096k
5116 * 2048MB: 5792k
5117 * 4096MB: 8192k
5118 * 8192MB: 11584k
5119 * 16384MB: 16384k
5120 */
bc75d33f 5121static int __init init_per_zone_wmark_min(void)
1da177e4
LT
5122{
5123 unsigned long lowmem_kbytes;
5124
5125 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5126
5127 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5128 if (min_free_kbytes < 128)
5129 min_free_kbytes = 128;
5130 if (min_free_kbytes > 65536)
5131 min_free_kbytes = 65536;
bc75d33f 5132 setup_per_zone_wmarks();
1da177e4 5133 setup_per_zone_lowmem_reserve();
556adecb 5134 setup_per_zone_inactive_ratio();
1da177e4
LT
5135 return 0;
5136}
bc75d33f 5137module_init(init_per_zone_wmark_min)
1da177e4
LT
5138
5139/*
5140 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5141 * that we can call two helper functions whenever min_free_kbytes
5142 * changes.
5143 */
5144int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
8d65af78 5145 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 5146{
8d65af78 5147 proc_dointvec(table, write, buffer, length, ppos);
3b1d92c5 5148 if (write)
bc75d33f 5149 setup_per_zone_wmarks();
1da177e4
LT
5150 return 0;
5151}
5152
9614634f
CL
5153#ifdef CONFIG_NUMA
5154int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
8d65af78 5155 void __user *buffer, size_t *length, loff_t *ppos)
9614634f
CL
5156{
5157 struct zone *zone;
5158 int rc;
5159
8d65af78 5160 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
9614634f
CL
5161 if (rc)
5162 return rc;
5163
5164 for_each_zone(zone)
8417bba4 5165 zone->min_unmapped_pages = (zone->present_pages *
9614634f
CL
5166 sysctl_min_unmapped_ratio) / 100;
5167 return 0;
5168}
0ff38490
CL
5169
5170int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
8d65af78 5171 void __user *buffer, size_t *length, loff_t *ppos)
0ff38490
CL
5172{
5173 struct zone *zone;
5174 int rc;
5175
8d65af78 5176 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
0ff38490
CL
5177 if (rc)
5178 return rc;
5179
5180 for_each_zone(zone)
5181 zone->min_slab_pages = (zone->present_pages *
5182 sysctl_min_slab_ratio) / 100;
5183 return 0;
5184}
9614634f
CL
5185#endif
5186
1da177e4
LT
5187/*
5188 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5189 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5190 * whenever sysctl_lowmem_reserve_ratio changes.
5191 *
5192 * The reserve ratio obviously has absolutely no relation with the
41858966 5193 * minimum watermarks. The lowmem reserve ratio can only make sense
1da177e4
LT
5194 * if in function of the boot time zone sizes.
5195 */
5196int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
8d65af78 5197 void __user *buffer, size_t *length, loff_t *ppos)
1da177e4 5198{
8d65af78 5199 proc_dointvec_minmax(table, write, buffer, length, ppos);
1da177e4
LT
5200 setup_per_zone_lowmem_reserve();
5201 return 0;
5202}
5203
8ad4b1fb
RS
5204/*
5205 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5206 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5207 * can have before it gets flushed back to buddy allocator.
5208 */
5209
5210int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
8d65af78 5211 void __user *buffer, size_t *length, loff_t *ppos)
8ad4b1fb
RS
5212{
5213 struct zone *zone;
5214 unsigned int cpu;
5215 int ret;
5216
8d65af78 5217 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8ad4b1fb
RS
5218 if (!write || (ret == -EINVAL))
5219 return ret;
364df0eb 5220 for_each_populated_zone(zone) {
99dcc3e5 5221 for_each_possible_cpu(cpu) {
8ad4b1fb
RS
5222 unsigned long high;
5223 high = zone->present_pages / percpu_pagelist_fraction;
99dcc3e5
CL
5224 setup_pagelist_highmark(
5225 per_cpu_ptr(zone->pageset, cpu), high);
8ad4b1fb
RS
5226 }
5227 }
5228 return 0;
5229}
5230
f034b5d4 5231int hashdist = HASHDIST_DEFAULT;
1da177e4
LT
5232
5233#ifdef CONFIG_NUMA
5234static int __init set_hashdist(char *str)
5235{
5236 if (!str)
5237 return 0;
5238 hashdist = simple_strtoul(str, &str, 0);
5239 return 1;
5240}
5241__setup("hashdist=", set_hashdist);
5242#endif
5243
5244/*
5245 * allocate a large system hash table from bootmem
5246 * - it is assumed that the hash table must contain an exact power-of-2
5247 * quantity of entries
5248 * - limit is the number of hash buckets, not the total allocation size
5249 */
5250void *__init alloc_large_system_hash(const char *tablename,
5251 unsigned long bucketsize,
5252 unsigned long numentries,
5253 int scale,
5254 int flags,
5255 unsigned int *_hash_shift,
5256 unsigned int *_hash_mask,
5257 unsigned long limit)
5258{
5259 unsigned long long max = limit;
5260 unsigned long log2qty, size;
5261 void *table = NULL;
5262
5263 /* allow the kernel cmdline to have a say */
5264 if (!numentries) {
5265 /* round applicable memory size up to nearest megabyte */
04903664 5266 numentries = nr_kernel_pages;
1da177e4
LT
5267 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5268 numentries >>= 20 - PAGE_SHIFT;
5269 numentries <<= 20 - PAGE_SHIFT;
5270
5271 /* limit to 1 bucket per 2^scale bytes of low memory */
5272 if (scale > PAGE_SHIFT)
5273 numentries >>= (scale - PAGE_SHIFT);
5274 else
5275 numentries <<= (PAGE_SHIFT - scale);
9ab37b8f
PM
5276
5277 /* Make sure we've got at least a 0-order allocation.. */
2c85f51d
JB
5278 if (unlikely(flags & HASH_SMALL)) {
5279 /* Makes no sense without HASH_EARLY */
5280 WARN_ON(!(flags & HASH_EARLY));
5281 if (!(numentries >> *_hash_shift)) {
5282 numentries = 1UL << *_hash_shift;
5283 BUG_ON(!numentries);
5284 }
5285 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
9ab37b8f 5286 numentries = PAGE_SIZE / bucketsize;
1da177e4 5287 }
6e692ed3 5288 numentries = roundup_pow_of_two(numentries);
1da177e4
LT
5289
5290 /* limit allocation size to 1/16 total memory by default */
5291 if (max == 0) {
5292 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5293 do_div(max, bucketsize);
5294 }
5295
5296 if (numentries > max)
5297 numentries = max;
5298
f0d1b0b3 5299 log2qty = ilog2(numentries);
1da177e4
LT
5300
5301 do {
5302 size = bucketsize << log2qty;
5303 if (flags & HASH_EARLY)
74768ed8 5304 table = alloc_bootmem_nopanic(size);
1da177e4
LT
5305 else if (hashdist)
5306 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5307 else {
1037b83b
ED
5308 /*
5309 * If bucketsize is not a power-of-two, we may free
a1dd268c
MG
5310 * some pages at the end of hash table which
5311 * alloc_pages_exact() automatically does
1037b83b 5312 */
264ef8a9 5313 if (get_order(size) < MAX_ORDER) {
a1dd268c 5314 table = alloc_pages_exact(size, GFP_ATOMIC);
264ef8a9
CM
5315 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5316 }
1da177e4
LT
5317 }
5318 } while (!table && size > PAGE_SIZE && --log2qty);
5319
5320 if (!table)
5321 panic("Failed to allocate %s hash table\n", tablename);
5322
f241e660 5323 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
1da177e4 5324 tablename,
f241e660 5325 (1UL << log2qty),
f0d1b0b3 5326 ilog2(size) - PAGE_SHIFT,
1da177e4
LT
5327 size);
5328
5329 if (_hash_shift)
5330 *_hash_shift = log2qty;
5331 if (_hash_mask)
5332 *_hash_mask = (1 << log2qty) - 1;
5333
5334 return table;
5335}
a117e66e 5336
835c134e
MG
5337/* Return a pointer to the bitmap storing bits affecting a block of pages */
5338static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5339 unsigned long pfn)
5340{
5341#ifdef CONFIG_SPARSEMEM
5342 return __pfn_to_section(pfn)->pageblock_flags;
5343#else
5344 return zone->pageblock_flags;
5345#endif /* CONFIG_SPARSEMEM */
5346}
5347
5348static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5349{
5350#ifdef CONFIG_SPARSEMEM
5351 pfn &= (PAGES_PER_SECTION-1);
d9c23400 5352 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
835c134e
MG
5353#else
5354 pfn = pfn - zone->zone_start_pfn;
d9c23400 5355 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
835c134e
MG
5356#endif /* CONFIG_SPARSEMEM */
5357}
5358
5359/**
d9c23400 5360 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
835c134e
MG
5361 * @page: The page within the block of interest
5362 * @start_bitidx: The first bit of interest to retrieve
5363 * @end_bitidx: The last bit of interest
5364 * returns pageblock_bits flags
5365 */
5366unsigned long get_pageblock_flags_group(struct page *page,
5367 int start_bitidx, int end_bitidx)
5368{
5369 struct zone *zone;
5370 unsigned long *bitmap;
5371 unsigned long pfn, bitidx;
5372 unsigned long flags = 0;
5373 unsigned long value = 1;
5374
5375 zone = page_zone(page);
5376 pfn = page_to_pfn(page);
5377 bitmap = get_pageblock_bitmap(zone, pfn);
5378 bitidx = pfn_to_bitidx(zone, pfn);
5379
5380 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5381 if (test_bit(bitidx + start_bitidx, bitmap))
5382 flags |= value;
6220ec78 5383
835c134e
MG
5384 return flags;
5385}
5386
5387/**
d9c23400 5388 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
835c134e
MG
5389 * @page: The page within the block of interest
5390 * @start_bitidx: The first bit of interest
5391 * @end_bitidx: The last bit of interest
5392 * @flags: The flags to set
5393 */
5394void set_pageblock_flags_group(struct page *page, unsigned long flags,
5395 int start_bitidx, int end_bitidx)
5396{
5397 struct zone *zone;
5398 unsigned long *bitmap;
5399 unsigned long pfn, bitidx;
5400 unsigned long value = 1;
5401
5402 zone = page_zone(page);
5403 pfn = page_to_pfn(page);
5404 bitmap = get_pageblock_bitmap(zone, pfn);
5405 bitidx = pfn_to_bitidx(zone, pfn);
86051ca5
KH
5406 VM_BUG_ON(pfn < zone->zone_start_pfn);
5407 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
835c134e
MG
5408
5409 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5410 if (flags & value)
5411 __set_bit(bitidx + start_bitidx, bitmap);
5412 else
5413 __clear_bit(bitidx + start_bitidx, bitmap);
5414}
a5d76b54
KH
5415
5416/*
5417 * This is designed as sub function...plz see page_isolation.c also.
5418 * set/clear page block's type to be ISOLATE.
5419 * page allocater never alloc memory from ISOLATE block.
5420 */
5421
49ac8255
KH
5422static int
5423__count_immobile_pages(struct zone *zone, struct page *page, int count)
5424{
5425 unsigned long pfn, iter, found;
5426 /*
5427 * For avoiding noise data, lru_add_drain_all() should be called
5428 * If ZONE_MOVABLE, the zone never contains immobile pages
5429 */
5430 if (zone_idx(zone) == ZONE_MOVABLE)
5431 return true;
5432
5433 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5434 return true;
5435
5436 pfn = page_to_pfn(page);
5437 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5438 unsigned long check = pfn + iter;
5439
29723fcc 5440 if (!pfn_valid_within(check))
49ac8255 5441 continue;
29723fcc 5442
49ac8255
KH
5443 page = pfn_to_page(check);
5444 if (!page_count(page)) {
5445 if (PageBuddy(page))
5446 iter += (1 << page_order(page)) - 1;
5447 continue;
5448 }
5449 if (!PageLRU(page))
5450 found++;
5451 /*
5452 * If there are RECLAIMABLE pages, we need to check it.
5453 * But now, memory offline itself doesn't call shrink_slab()
5454 * and it still to be fixed.
5455 */
5456 /*
5457 * If the page is not RAM, page_count()should be 0.
5458 * we don't need more check. This is an _used_ not-movable page.
5459 *
5460 * The problematic thing here is PG_reserved pages. PG_reserved
5461 * is set to both of a memory hole page and a _used_ kernel
5462 * page at boot.
5463 */
5464 if (found > count)
5465 return false;
5466 }
5467 return true;
5468}
5469
5470bool is_pageblock_removable_nolock(struct page *page)
5471{
5472 struct zone *zone = page_zone(page);
5473 return __count_immobile_pages(zone, page, 0);
5474}
5475
a5d76b54
KH
5476int set_migratetype_isolate(struct page *page)
5477{
5478 struct zone *zone;
49ac8255 5479 unsigned long flags, pfn;
925cc71e
RJ
5480 struct memory_isolate_notify arg;
5481 int notifier_ret;
a5d76b54 5482 int ret = -EBUSY;
8e7e40d9 5483 int zone_idx;
a5d76b54
KH
5484
5485 zone = page_zone(page);
8e7e40d9 5486 zone_idx = zone_idx(zone);
925cc71e 5487
a5d76b54 5488 spin_lock_irqsave(&zone->lock, flags);
925cc71e
RJ
5489
5490 pfn = page_to_pfn(page);
5491 arg.start_pfn = pfn;
5492 arg.nr_pages = pageblock_nr_pages;
5493 arg.pages_found = 0;
5494
a5d76b54 5495 /*
925cc71e
RJ
5496 * It may be possible to isolate a pageblock even if the
5497 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5498 * notifier chain is used by balloon drivers to return the
5499 * number of pages in a range that are held by the balloon
5500 * driver to shrink memory. If all the pages are accounted for
5501 * by balloons, are free, or on the LRU, isolation can continue.
5502 * Later, for example, when memory hotplug notifier runs, these
5503 * pages reported as "can be isolated" should be isolated(freed)
5504 * by the balloon driver through the memory notifier chain.
a5d76b54 5505 */
925cc71e
RJ
5506 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5507 notifier_ret = notifier_to_errno(notifier_ret);
4b20477f 5508 if (notifier_ret)
a5d76b54 5509 goto out;
49ac8255
KH
5510 /*
5511 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5512 * We just check MOVABLE pages.
5513 */
5514 if (__count_immobile_pages(zone, page, arg.pages_found))
925cc71e
RJ
5515 ret = 0;
5516
49ac8255
KH
5517 /*
5518 * immobile means "not-on-lru" paes. If immobile is larger than
5519 * removable-by-driver pages reported by notifier, we'll fail.
5520 */
5521
a5d76b54 5522out:
925cc71e
RJ
5523 if (!ret) {
5524 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5525 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5526 }
5527
a5d76b54
KH
5528 spin_unlock_irqrestore(&zone->lock, flags);
5529 if (!ret)
9f8f2172 5530 drain_all_pages();
a5d76b54
KH
5531 return ret;
5532}
5533
5534void unset_migratetype_isolate(struct page *page)
5535{
5536 struct zone *zone;
5537 unsigned long flags;
5538 zone = page_zone(page);
5539 spin_lock_irqsave(&zone->lock, flags);
5540 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5541 goto out;
5542 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5543 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5544out:
5545 spin_unlock_irqrestore(&zone->lock, flags);
5546}
0c0e6195
KH
5547
5548#ifdef CONFIG_MEMORY_HOTREMOVE
5549/*
5550 * All pages in the range must be isolated before calling this.
5551 */
5552void
5553__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5554{
5555 struct page *page;
5556 struct zone *zone;
5557 int order, i;
5558 unsigned long pfn;
5559 unsigned long flags;
5560 /* find the first valid pfn */
5561 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5562 if (pfn_valid(pfn))
5563 break;
5564 if (pfn == end_pfn)
5565 return;
5566 zone = page_zone(pfn_to_page(pfn));
5567 spin_lock_irqsave(&zone->lock, flags);
5568 pfn = start_pfn;
5569 while (pfn < end_pfn) {
5570 if (!pfn_valid(pfn)) {
5571 pfn++;
5572 continue;
5573 }
5574 page = pfn_to_page(pfn);
5575 BUG_ON(page_count(page));
5576 BUG_ON(!PageBuddy(page));
5577 order = page_order(page);
5578#ifdef CONFIG_DEBUG_VM
5579 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5580 pfn, 1 << order, end_pfn);
5581#endif
5582 list_del(&page->lru);
5583 rmv_page_order(page);
5584 zone->free_area[order].nr_free--;
5585 __mod_zone_page_state(zone, NR_FREE_PAGES,
5586 - (1UL << order));
5587 for (i = 0; i < (1 << order); i++)
5588 SetPageReserved((page+i));
5589 pfn += (1 << order);
5590 }
5591 spin_unlock_irqrestore(&zone->lock, flags);
5592}
5593#endif
8d22ba1b
WF
5594
5595#ifdef CONFIG_MEMORY_FAILURE
5596bool is_free_buddy_page(struct page *page)
5597{
5598 struct zone *zone = page_zone(page);
5599 unsigned long pfn = page_to_pfn(page);
5600 unsigned long flags;
5601 int order;
5602
5603 spin_lock_irqsave(&zone->lock, flags);
5604 for (order = 0; order < MAX_ORDER; order++) {
5605 struct page *page_head = page - (pfn & ((1 << order) - 1));
5606
5607 if (PageBuddy(page_head) && page_order(page_head) >= order)
5608 break;
5609 }
5610 spin_unlock_irqrestore(&zone->lock, flags);
5611
5612 return order < MAX_ORDER;
5613}
5614#endif
718a3821
WF
5615
5616static struct trace_print_flags pageflag_names[] = {
5617 {1UL << PG_locked, "locked" },
5618 {1UL << PG_error, "error" },
5619 {1UL << PG_referenced, "referenced" },
5620 {1UL << PG_uptodate, "uptodate" },
5621 {1UL << PG_dirty, "dirty" },
5622 {1UL << PG_lru, "lru" },
5623 {1UL << PG_active, "active" },
5624 {1UL << PG_slab, "slab" },
5625 {1UL << PG_owner_priv_1, "owner_priv_1" },
5626 {1UL << PG_arch_1, "arch_1" },
5627 {1UL << PG_reserved, "reserved" },
5628 {1UL << PG_private, "private" },
5629 {1UL << PG_private_2, "private_2" },
5630 {1UL << PG_writeback, "writeback" },
5631#ifdef CONFIG_PAGEFLAGS_EXTENDED
5632 {1UL << PG_head, "head" },
5633 {1UL << PG_tail, "tail" },
5634#else
5635 {1UL << PG_compound, "compound" },
5636#endif
5637 {1UL << PG_swapcache, "swapcache" },
5638 {1UL << PG_mappedtodisk, "mappedtodisk" },
5639 {1UL << PG_reclaim, "reclaim" },
718a3821
WF
5640 {1UL << PG_swapbacked, "swapbacked" },
5641 {1UL << PG_unevictable, "unevictable" },
5642#ifdef CONFIG_MMU
5643 {1UL << PG_mlocked, "mlocked" },
5644#endif
5645#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5646 {1UL << PG_uncached, "uncached" },
5647#endif
5648#ifdef CONFIG_MEMORY_FAILURE
5649 {1UL << PG_hwpoison, "hwpoison" },
5650#endif
5651 {-1UL, NULL },
5652};
5653
5654static void dump_page_flags(unsigned long flags)
5655{
5656 const char *delim = "";
5657 unsigned long mask;
5658 int i;
5659
5660 printk(KERN_ALERT "page flags: %#lx(", flags);
5661
5662 /* remove zone id */
5663 flags &= (1UL << NR_PAGEFLAGS) - 1;
5664
5665 for (i = 0; pageflag_names[i].name && flags; i++) {
5666
5667 mask = pageflag_names[i].mask;
5668 if ((flags & mask) != mask)
5669 continue;
5670
5671 flags &= ~mask;
5672 printk("%s%s", delim, pageflag_names[i].name);
5673 delim = "|";
5674 }
5675
5676 /* check for left over flags */
5677 if (flags)
5678 printk("%s%#lx", delim, flags);
5679
5680 printk(")\n");
5681}
5682
5683void dump_page(struct page *page)
5684{
5685 printk(KERN_ALERT
5686 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
4e9f64c4 5687 page, atomic_read(&page->_count), page_mapcount(page),
718a3821
WF
5688 page->mapping, page->index);
5689 dump_page_flags(page->flags);
f212ad7c 5690 mem_cgroup_print_bad_page(page);
718a3821 5691}