]>
Commit | Line | Data |
---|---|---|
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 LT |
23 | #include <linux/bootmem.h> |
24 | #include <linux/compiler.h> | |
9f158333 | 25 | #include <linux/kernel.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/suspend.h> | |
28 | #include <linux/pagevec.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/slab.h> | |
5a3135c2 | 31 | #include <linux/oom.h> |
1da177e4 LT |
32 | #include <linux/notifier.h> |
33 | #include <linux/topology.h> | |
34 | #include <linux/sysctl.h> | |
35 | #include <linux/cpu.h> | |
36 | #include <linux/cpuset.h> | |
bdc8cb98 | 37 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
38 | #include <linux/nodemask.h> |
39 | #include <linux/vmalloc.h> | |
4be38e35 | 40 | #include <linux/mempolicy.h> |
6811378e | 41 | #include <linux/stop_machine.h> |
c713216d MG |
42 | #include <linux/sort.h> |
43 | #include <linux/pfn.h> | |
3fcfab16 | 44 | #include <linux/backing-dev.h> |
933e312e | 45 | #include <linux/fault-inject.h> |
a5d76b54 | 46 | #include <linux/page-isolation.h> |
52d4b9ac | 47 | #include <linux/page_cgroup.h> |
3ac7fe5a | 48 | #include <linux/debugobjects.h> |
dbb1f81c | 49 | #include <linux/kmemleak.h> |
1da177e4 LT |
50 | |
51 | #include <asm/tlbflush.h> | |
ac924c60 | 52 | #include <asm/div64.h> |
1da177e4 LT |
53 | #include "internal.h" |
54 | ||
55 | /* | |
13808910 | 56 | * Array of node states. |
1da177e4 | 57 | */ |
13808910 CL |
58 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
59 | [N_POSSIBLE] = NODE_MASK_ALL, | |
60 | [N_ONLINE] = { { [0] = 1UL } }, | |
61 | #ifndef CONFIG_NUMA | |
62 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
63 | #ifdef CONFIG_HIGHMEM | |
64 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
65 | #endif | |
66 | [N_CPU] = { { [0] = 1UL } }, | |
67 | #endif /* NUMA */ | |
68 | }; | |
69 | EXPORT_SYMBOL(node_states); | |
70 | ||
6c231b7b | 71 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 72 | unsigned long totalreserve_pages __read_mostly; |
22b31eec | 73 | unsigned long highest_memmap_pfn __read_mostly; |
8ad4b1fb | 74 | int percpu_pagelist_fraction; |
1da177e4 | 75 | |
d9c23400 MG |
76 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
77 | int pageblock_order __read_mostly; | |
78 | #endif | |
79 | ||
d98c7a09 | 80 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 81 | |
1da177e4 LT |
82 | /* |
83 | * results with 256, 32 in the lowmem_reserve sysctl: | |
84 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
85 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
86 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
87 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
88 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
89 | * |
90 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
91 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 92 | */ |
2f1b6248 | 93 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 94 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 95 | 256, |
4b51d669 | 96 | #endif |
fb0e7942 | 97 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 98 | 256, |
fb0e7942 | 99 | #endif |
e53ef38d | 100 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 101 | 32, |
e53ef38d | 102 | #endif |
2a1e274a | 103 | 32, |
2f1b6248 | 104 | }; |
1da177e4 LT |
105 | |
106 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 107 | |
15ad7cdc | 108 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 109 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 110 | "DMA", |
4b51d669 | 111 | #endif |
fb0e7942 | 112 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 113 | "DMA32", |
fb0e7942 | 114 | #endif |
2f1b6248 | 115 | "Normal", |
e53ef38d | 116 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 117 | "HighMem", |
e53ef38d | 118 | #endif |
2a1e274a | 119 | "Movable", |
2f1b6248 CL |
120 | }; |
121 | ||
1da177e4 LT |
122 | int min_free_kbytes = 1024; |
123 | ||
86356ab1 YG |
124 | unsigned long __meminitdata nr_kernel_pages; |
125 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 126 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 127 | |
c713216d MG |
128 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
129 | /* | |
183ff22b | 130 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
131 | * ranges of memory (RAM) that may be registered with add_active_range(). |
132 | * Ranges passed to add_active_range() will be merged if possible | |
133 | * so the number of times add_active_range() can be called is | |
134 | * related to the number of nodes and the number of holes | |
135 | */ | |
136 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
137 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
138 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
139 | #else | |
140 | #if MAX_NUMNODES >= 32 | |
141 | /* If there can be many nodes, allow up to 50 holes per node */ | |
142 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
143 | #else | |
144 | /* By default, allow up to 256 distinct regions */ | |
145 | #define MAX_ACTIVE_REGIONS 256 | |
146 | #endif | |
147 | #endif | |
148 | ||
98011f56 JB |
149 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
150 | static int __meminitdata nr_nodemap_entries; | |
151 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
152 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
b69a7288 | 153 | static unsigned long __initdata required_kernelcore; |
484f51f8 | 154 | static unsigned long __initdata required_movablecore; |
b69a7288 | 155 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
156 | |
157 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
158 | int movable_zone; | |
159 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
160 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
161 | ||
418508c1 MS |
162 | #if MAX_NUMNODES > 1 |
163 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
164 | EXPORT_SYMBOL(nr_node_ids); | |
165 | #endif | |
166 | ||
9ef9acb0 MG |
167 | int page_group_by_mobility_disabled __read_mostly; |
168 | ||
b2a0ac88 MG |
169 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
170 | { | |
49255c61 MG |
171 | |
172 | if (unlikely(page_group_by_mobility_disabled)) | |
173 | migratetype = MIGRATE_UNMOVABLE; | |
174 | ||
b2a0ac88 MG |
175 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
176 | PB_migrate, PB_migrate_end); | |
177 | } | |
178 | ||
13e7444b | 179 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 180 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 181 | { |
bdc8cb98 DH |
182 | int ret = 0; |
183 | unsigned seq; | |
184 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 185 | |
bdc8cb98 DH |
186 | do { |
187 | seq = zone_span_seqbegin(zone); | |
188 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
189 | ret = 1; | |
190 | else if (pfn < zone->zone_start_pfn) | |
191 | ret = 1; | |
192 | } while (zone_span_seqretry(zone, seq)); | |
193 | ||
194 | return ret; | |
c6a57e19 DH |
195 | } |
196 | ||
197 | static int page_is_consistent(struct zone *zone, struct page *page) | |
198 | { | |
14e07298 | 199 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 200 | return 0; |
1da177e4 | 201 | if (zone != page_zone(page)) |
c6a57e19 DH |
202 | return 0; |
203 | ||
204 | return 1; | |
205 | } | |
206 | /* | |
207 | * Temporary debugging check for pages not lying within a given zone. | |
208 | */ | |
209 | static int bad_range(struct zone *zone, struct page *page) | |
210 | { | |
211 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 212 | return 1; |
c6a57e19 DH |
213 | if (!page_is_consistent(zone, page)) |
214 | return 1; | |
215 | ||
1da177e4 LT |
216 | return 0; |
217 | } | |
13e7444b NP |
218 | #else |
219 | static inline int bad_range(struct zone *zone, struct page *page) | |
220 | { | |
221 | return 0; | |
222 | } | |
223 | #endif | |
224 | ||
224abf92 | 225 | static void bad_page(struct page *page) |
1da177e4 | 226 | { |
d936cf9b HD |
227 | static unsigned long resume; |
228 | static unsigned long nr_shown; | |
229 | static unsigned long nr_unshown; | |
230 | ||
231 | /* | |
232 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
233 | * or allow a steady drip of one report per second. | |
234 | */ | |
235 | if (nr_shown == 60) { | |
236 | if (time_before(jiffies, resume)) { | |
237 | nr_unshown++; | |
238 | goto out; | |
239 | } | |
240 | if (nr_unshown) { | |
1e9e6365 HD |
241 | printk(KERN_ALERT |
242 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
243 | nr_unshown); |
244 | nr_unshown = 0; | |
245 | } | |
246 | nr_shown = 0; | |
247 | } | |
248 | if (nr_shown++ == 0) | |
249 | resume = jiffies + 60 * HZ; | |
250 | ||
1e9e6365 | 251 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 252 | current->comm, page_to_pfn(page)); |
1e9e6365 | 253 | printk(KERN_ALERT |
3dc14741 HD |
254 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
255 | page, (void *)page->flags, page_count(page), | |
256 | page_mapcount(page), page->mapping, page->index); | |
3dc14741 | 257 | |
1da177e4 | 258 | dump_stack(); |
d936cf9b | 259 | out: |
8cc3b392 HD |
260 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
261 | __ClearPageBuddy(page); | |
9f158333 | 262 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
263 | } |
264 | ||
1da177e4 LT |
265 | /* |
266 | * Higher-order pages are called "compound pages". They are structured thusly: | |
267 | * | |
268 | * The first PAGE_SIZE page is called the "head page". | |
269 | * | |
270 | * The remaining PAGE_SIZE pages are called "tail pages". | |
271 | * | |
272 | * All pages have PG_compound set. All pages have their ->private pointing at | |
273 | * the head page (even the head page has this). | |
274 | * | |
41d78ba5 HD |
275 | * The first tail page's ->lru.next holds the address of the compound page's |
276 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
277 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 278 | */ |
d98c7a09 HD |
279 | |
280 | static void free_compound_page(struct page *page) | |
281 | { | |
d85f3385 | 282 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
283 | } |
284 | ||
01ad1c08 | 285 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
286 | { |
287 | int i; | |
288 | int nr_pages = 1 << order; | |
289 | ||
290 | set_compound_page_dtor(page, free_compound_page); | |
291 | set_compound_order(page, order); | |
292 | __SetPageHead(page); | |
293 | for (i = 1; i < nr_pages; i++) { | |
294 | struct page *p = page + i; | |
295 | ||
296 | __SetPageTail(p); | |
297 | p->first_page = page; | |
298 | } | |
299 | } | |
300 | ||
301 | #ifdef CONFIG_HUGETLBFS | |
302 | void prep_compound_gigantic_page(struct page *page, unsigned long order) | |
1da177e4 LT |
303 | { |
304 | int i; | |
305 | int nr_pages = 1 << order; | |
6babc32c | 306 | struct page *p = page + 1; |
1da177e4 | 307 | |
33f2ef89 | 308 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 309 | set_compound_order(page, order); |
6d777953 | 310 | __SetPageHead(page); |
18229df5 | 311 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { |
d85f3385 | 312 | __SetPageTail(p); |
d85f3385 | 313 | p->first_page = page; |
1da177e4 LT |
314 | } |
315 | } | |
18229df5 | 316 | #endif |
1da177e4 | 317 | |
8cc3b392 | 318 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
319 | { |
320 | int i; | |
321 | int nr_pages = 1 << order; | |
8cc3b392 | 322 | int bad = 0; |
1da177e4 | 323 | |
8cc3b392 HD |
324 | if (unlikely(compound_order(page) != order) || |
325 | unlikely(!PageHead(page))) { | |
224abf92 | 326 | bad_page(page); |
8cc3b392 HD |
327 | bad++; |
328 | } | |
1da177e4 | 329 | |
6d777953 | 330 | __ClearPageHead(page); |
8cc3b392 | 331 | |
18229df5 AW |
332 | for (i = 1; i < nr_pages; i++) { |
333 | struct page *p = page + i; | |
1da177e4 | 334 | |
e713a21d | 335 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 336 | bad_page(page); |
8cc3b392 HD |
337 | bad++; |
338 | } | |
d85f3385 | 339 | __ClearPageTail(p); |
1da177e4 | 340 | } |
8cc3b392 HD |
341 | |
342 | return bad; | |
1da177e4 | 343 | } |
1da177e4 | 344 | |
17cf4406 NP |
345 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
346 | { | |
347 | int i; | |
348 | ||
6626c5d5 AM |
349 | /* |
350 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
351 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
352 | */ | |
725d704e | 353 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
354 | for (i = 0; i < (1 << order); i++) |
355 | clear_highpage(page + i); | |
356 | } | |
357 | ||
6aa3001b AM |
358 | static inline void set_page_order(struct page *page, int order) |
359 | { | |
4c21e2f2 | 360 | set_page_private(page, order); |
676165a8 | 361 | __SetPageBuddy(page); |
1da177e4 LT |
362 | } |
363 | ||
364 | static inline void rmv_page_order(struct page *page) | |
365 | { | |
676165a8 | 366 | __ClearPageBuddy(page); |
4c21e2f2 | 367 | set_page_private(page, 0); |
1da177e4 LT |
368 | } |
369 | ||
370 | /* | |
371 | * Locate the struct page for both the matching buddy in our | |
372 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
373 | * | |
374 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
375 | * the following equation: | |
376 | * B2 = B1 ^ (1 << O) | |
377 | * For example, if the starting buddy (buddy2) is #8 its order | |
378 | * 1 buddy is #10: | |
379 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
380 | * | |
381 | * 2) Any buddy B will have an order O+1 parent P which | |
382 | * satisfies the following equation: | |
383 | * P = B & ~(1 << O) | |
384 | * | |
d6e05edc | 385 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
386 | */ |
387 | static inline struct page * | |
388 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
389 | { | |
390 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
391 | ||
392 | return page + (buddy_idx - page_idx); | |
393 | } | |
394 | ||
395 | static inline unsigned long | |
396 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
397 | { | |
398 | return (page_idx & ~(1 << order)); | |
399 | } | |
400 | ||
401 | /* | |
402 | * This function checks whether a page is free && is the buddy | |
403 | * we can do coalesce a page and its buddy if | |
13e7444b | 404 | * (a) the buddy is not in a hole && |
676165a8 | 405 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
406 | * (c) a page and its buddy have the same order && |
407 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
408 | * |
409 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
410 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 411 | * |
676165a8 | 412 | * For recording page's order, we use page_private(page). |
1da177e4 | 413 | */ |
cb2b95e1 AW |
414 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
415 | int order) | |
1da177e4 | 416 | { |
14e07298 | 417 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 418 | return 0; |
13e7444b | 419 | |
cb2b95e1 AW |
420 | if (page_zone_id(page) != page_zone_id(buddy)) |
421 | return 0; | |
422 | ||
423 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
424 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 425 | return 1; |
676165a8 | 426 | } |
6aa3001b | 427 | return 0; |
1da177e4 LT |
428 | } |
429 | ||
430 | /* | |
431 | * Freeing function for a buddy system allocator. | |
432 | * | |
433 | * The concept of a buddy system is to maintain direct-mapped table | |
434 | * (containing bit values) for memory blocks of various "orders". | |
435 | * The bottom level table contains the map for the smallest allocatable | |
436 | * units of memory (here, pages), and each level above it describes | |
437 | * pairs of units from the levels below, hence, "buddies". | |
438 | * At a high level, all that happens here is marking the table entry | |
439 | * at the bottom level available, and propagating the changes upward | |
440 | * as necessary, plus some accounting needed to play nicely with other | |
441 | * parts of the VM system. | |
442 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 443 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 444 | * order is recorded in page_private(page) field. |
1da177e4 LT |
445 | * So when we are allocating or freeing one, we can derive the state of the |
446 | * other. That is, if we allocate a small block, and both were | |
447 | * free, the remainder of the region must be split into blocks. | |
448 | * If a block is freed, and its buddy is also free, then this | |
449 | * triggers coalescing into a block of larger size. | |
450 | * | |
451 | * -- wli | |
452 | */ | |
453 | ||
48db57f8 | 454 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
455 | struct zone *zone, unsigned int order) |
456 | { | |
457 | unsigned long page_idx; | |
458 | int order_size = 1 << order; | |
b2a0ac88 | 459 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 460 | |
224abf92 | 461 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
462 | if (unlikely(destroy_compound_page(page, order))) |
463 | return; | |
1da177e4 LT |
464 | |
465 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
466 | ||
725d704e NP |
467 | VM_BUG_ON(page_idx & (order_size - 1)); |
468 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 469 | |
d23ad423 | 470 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
471 | while (order < MAX_ORDER-1) { |
472 | unsigned long combined_idx; | |
1da177e4 LT |
473 | struct page *buddy; |
474 | ||
1da177e4 | 475 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 476 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 477 | break; |
13e7444b | 478 | |
3c82d0ce | 479 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 480 | list_del(&buddy->lru); |
b2a0ac88 | 481 | zone->free_area[order].nr_free--; |
1da177e4 | 482 | rmv_page_order(buddy); |
13e7444b | 483 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
484 | page = page + (combined_idx - page_idx); |
485 | page_idx = combined_idx; | |
486 | order++; | |
487 | } | |
488 | set_page_order(page, order); | |
b2a0ac88 MG |
489 | list_add(&page->lru, |
490 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
491 | zone->free_area[order].nr_free++; |
492 | } | |
493 | ||
224abf92 | 494 | static inline int free_pages_check(struct page *page) |
1da177e4 | 495 | { |
985737cf | 496 | free_page_mlock(page); |
92be2e33 NP |
497 | if (unlikely(page_mapcount(page) | |
498 | (page->mapping != NULL) | | |
499 | (page_count(page) != 0) | | |
8cc3b392 | 500 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
224abf92 | 501 | bad_page(page); |
79f4b7bf | 502 | return 1; |
8cc3b392 | 503 | } |
79f4b7bf HD |
504 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
505 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
506 | return 0; | |
1da177e4 LT |
507 | } |
508 | ||
509 | /* | |
510 | * Frees a list of pages. | |
511 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 512 | * count is the number of pages to free. |
1da177e4 LT |
513 | * |
514 | * If the zone was previously in an "all pages pinned" state then look to | |
515 | * see if this freeing clears that state. | |
516 | * | |
517 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
518 | * pinned" detection logic. | |
519 | */ | |
48db57f8 NP |
520 | static void free_pages_bulk(struct zone *zone, int count, |
521 | struct list_head *list, int order) | |
1da177e4 | 522 | { |
c54ad30c | 523 | spin_lock(&zone->lock); |
e815af95 | 524 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 525 | zone->pages_scanned = 0; |
48db57f8 NP |
526 | while (count--) { |
527 | struct page *page; | |
528 | ||
725d704e | 529 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 530 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 531 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 532 | list_del(&page->lru); |
48db57f8 | 533 | __free_one_page(page, zone, order); |
1da177e4 | 534 | } |
c54ad30c | 535 | spin_unlock(&zone->lock); |
1da177e4 LT |
536 | } |
537 | ||
48db57f8 | 538 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 539 | { |
006d22d9 | 540 | spin_lock(&zone->lock); |
e815af95 | 541 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 542 | zone->pages_scanned = 0; |
0798e519 | 543 | __free_one_page(page, zone, order); |
006d22d9 | 544 | spin_unlock(&zone->lock); |
48db57f8 NP |
545 | } |
546 | ||
547 | static void __free_pages_ok(struct page *page, unsigned int order) | |
548 | { | |
549 | unsigned long flags; | |
1da177e4 | 550 | int i; |
8cc3b392 | 551 | int bad = 0; |
1da177e4 | 552 | |
1da177e4 | 553 | for (i = 0 ; i < (1 << order) ; ++i) |
8cc3b392 HD |
554 | bad += free_pages_check(page + i); |
555 | if (bad) | |
689bcebf HD |
556 | return; |
557 | ||
3ac7fe5a | 558 | if (!PageHighMem(page)) { |
9858db50 | 559 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
560 | debug_check_no_obj_freed(page_address(page), |
561 | PAGE_SIZE << order); | |
562 | } | |
dafb1367 | 563 | arch_free_page(page, order); |
48db57f8 | 564 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 565 | |
c54ad30c | 566 | local_irq_save(flags); |
f8891e5e | 567 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 568 | free_one_page(page_zone(page), page, order); |
c54ad30c | 569 | local_irq_restore(flags); |
1da177e4 LT |
570 | } |
571 | ||
a226f6c8 DH |
572 | /* |
573 | * permit the bootmem allocator to evade page validation on high-order frees | |
574 | */ | |
af370fb8 | 575 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
576 | { |
577 | if (order == 0) { | |
578 | __ClearPageReserved(page); | |
579 | set_page_count(page, 0); | |
7835e98b | 580 | set_page_refcounted(page); |
545b1ea9 | 581 | __free_page(page); |
a226f6c8 | 582 | } else { |
a226f6c8 DH |
583 | int loop; |
584 | ||
545b1ea9 | 585 | prefetchw(page); |
a226f6c8 DH |
586 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
587 | struct page *p = &page[loop]; | |
588 | ||
545b1ea9 NP |
589 | if (loop + 1 < BITS_PER_LONG) |
590 | prefetchw(p + 1); | |
a226f6c8 DH |
591 | __ClearPageReserved(p); |
592 | set_page_count(p, 0); | |
593 | } | |
594 | ||
7835e98b | 595 | set_page_refcounted(page); |
545b1ea9 | 596 | __free_pages(page, order); |
a226f6c8 DH |
597 | } |
598 | } | |
599 | ||
1da177e4 LT |
600 | |
601 | /* | |
602 | * The order of subdivision here is critical for the IO subsystem. | |
603 | * Please do not alter this order without good reasons and regression | |
604 | * testing. Specifically, as large blocks of memory are subdivided, | |
605 | * the order in which smaller blocks are delivered depends on the order | |
606 | * they're subdivided in this function. This is the primary factor | |
607 | * influencing the order in which pages are delivered to the IO | |
608 | * subsystem according to empirical testing, and this is also justified | |
609 | * by considering the behavior of a buddy system containing a single | |
610 | * large block of memory acted on by a series of small allocations. | |
611 | * This behavior is a critical factor in sglist merging's success. | |
612 | * | |
613 | * -- wli | |
614 | */ | |
085cc7d5 | 615 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
616 | int low, int high, struct free_area *area, |
617 | int migratetype) | |
1da177e4 LT |
618 | { |
619 | unsigned long size = 1 << high; | |
620 | ||
621 | while (high > low) { | |
622 | area--; | |
623 | high--; | |
624 | size >>= 1; | |
725d704e | 625 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 626 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
627 | area->nr_free++; |
628 | set_page_order(&page[size], high); | |
629 | } | |
1da177e4 LT |
630 | } |
631 | ||
1da177e4 LT |
632 | /* |
633 | * This page is about to be returned from the page allocator | |
634 | */ | |
17cf4406 | 635 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 636 | { |
92be2e33 NP |
637 | if (unlikely(page_mapcount(page) | |
638 | (page->mapping != NULL) | | |
639 | (page_count(page) != 0) | | |
8cc3b392 | 640 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
224abf92 | 641 | bad_page(page); |
689bcebf | 642 | return 1; |
8cc3b392 | 643 | } |
689bcebf | 644 | |
4c21e2f2 | 645 | set_page_private(page, 0); |
7835e98b | 646 | set_page_refcounted(page); |
cc102509 NP |
647 | |
648 | arch_alloc_page(page, order); | |
1da177e4 | 649 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
650 | |
651 | if (gfp_flags & __GFP_ZERO) | |
652 | prep_zero_page(page, order, gfp_flags); | |
653 | ||
654 | if (order && (gfp_flags & __GFP_COMP)) | |
655 | prep_compound_page(page, order); | |
656 | ||
689bcebf | 657 | return 0; |
1da177e4 LT |
658 | } |
659 | ||
56fd56b8 MG |
660 | /* |
661 | * Go through the free lists for the given migratetype and remove | |
662 | * the smallest available page from the freelists | |
663 | */ | |
664 | static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
665 | int migratetype) | |
666 | { | |
667 | unsigned int current_order; | |
668 | struct free_area * area; | |
669 | struct page *page; | |
670 | ||
671 | /* Find a page of the appropriate size in the preferred list */ | |
672 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
673 | area = &(zone->free_area[current_order]); | |
674 | if (list_empty(&area->free_list[migratetype])) | |
675 | continue; | |
676 | ||
677 | page = list_entry(area->free_list[migratetype].next, | |
678 | struct page, lru); | |
679 | list_del(&page->lru); | |
680 | rmv_page_order(page); | |
681 | area->nr_free--; | |
682 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); | |
683 | expand(zone, page, order, current_order, area, migratetype); | |
684 | return page; | |
685 | } | |
686 | ||
687 | return NULL; | |
688 | } | |
689 | ||
690 | ||
b2a0ac88 MG |
691 | /* |
692 | * This array describes the order lists are fallen back to when | |
693 | * the free lists for the desirable migrate type are depleted | |
694 | */ | |
695 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
696 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
697 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
698 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
699 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
700 | }; |
701 | ||
c361be55 MG |
702 | /* |
703 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 704 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
705 | * boundary. If alignment is required, use move_freepages_block() |
706 | */ | |
b69a7288 AB |
707 | static int move_freepages(struct zone *zone, |
708 | struct page *start_page, struct page *end_page, | |
709 | int migratetype) | |
c361be55 MG |
710 | { |
711 | struct page *page; | |
712 | unsigned long order; | |
d100313f | 713 | int pages_moved = 0; |
c361be55 MG |
714 | |
715 | #ifndef CONFIG_HOLES_IN_ZONE | |
716 | /* | |
717 | * page_zone is not safe to call in this context when | |
718 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
719 | * anyway as we check zone boundaries in move_freepages_block(). | |
720 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 721 | * grouping pages by mobility |
c361be55 MG |
722 | */ |
723 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
724 | #endif | |
725 | ||
726 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
727 | /* Make sure we are not inadvertently changing nodes */ |
728 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
729 | ||
c361be55 MG |
730 | if (!pfn_valid_within(page_to_pfn(page))) { |
731 | page++; | |
732 | continue; | |
733 | } | |
734 | ||
735 | if (!PageBuddy(page)) { | |
736 | page++; | |
737 | continue; | |
738 | } | |
739 | ||
740 | order = page_order(page); | |
741 | list_del(&page->lru); | |
742 | list_add(&page->lru, | |
743 | &zone->free_area[order].free_list[migratetype]); | |
744 | page += 1 << order; | |
d100313f | 745 | pages_moved += 1 << order; |
c361be55 MG |
746 | } |
747 | ||
d100313f | 748 | return pages_moved; |
c361be55 MG |
749 | } |
750 | ||
b69a7288 AB |
751 | static int move_freepages_block(struct zone *zone, struct page *page, |
752 | int migratetype) | |
c361be55 MG |
753 | { |
754 | unsigned long start_pfn, end_pfn; | |
755 | struct page *start_page, *end_page; | |
756 | ||
757 | start_pfn = page_to_pfn(page); | |
d9c23400 | 758 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 759 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
760 | end_page = start_page + pageblock_nr_pages - 1; |
761 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
762 | |
763 | /* Do not cross zone boundaries */ | |
764 | if (start_pfn < zone->zone_start_pfn) | |
765 | start_page = page; | |
766 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
767 | return 0; | |
768 | ||
769 | return move_freepages(zone, start_page, end_page, migratetype); | |
770 | } | |
771 | ||
b2a0ac88 MG |
772 | /* Remove an element from the buddy allocator from the fallback list */ |
773 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
774 | int start_migratetype) | |
775 | { | |
776 | struct free_area * area; | |
777 | int current_order; | |
778 | struct page *page; | |
779 | int migratetype, i; | |
780 | ||
781 | /* Find the largest possible block of pages in the other list */ | |
782 | for (current_order = MAX_ORDER-1; current_order >= order; | |
783 | --current_order) { | |
784 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
785 | migratetype = fallbacks[start_migratetype][i]; | |
786 | ||
56fd56b8 MG |
787 | /* MIGRATE_RESERVE handled later if necessary */ |
788 | if (migratetype == MIGRATE_RESERVE) | |
789 | continue; | |
e010487d | 790 | |
b2a0ac88 MG |
791 | area = &(zone->free_area[current_order]); |
792 | if (list_empty(&area->free_list[migratetype])) | |
793 | continue; | |
794 | ||
795 | page = list_entry(area->free_list[migratetype].next, | |
796 | struct page, lru); | |
797 | area->nr_free--; | |
798 | ||
799 | /* | |
c361be55 | 800 | * If breaking a large block of pages, move all free |
46dafbca MG |
801 | * pages to the preferred allocation list. If falling |
802 | * back for a reclaimable kernel allocation, be more | |
803 | * agressive about taking ownership of free pages | |
b2a0ac88 | 804 | */ |
d9c23400 | 805 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
46dafbca MG |
806 | start_migratetype == MIGRATE_RECLAIMABLE) { |
807 | unsigned long pages; | |
808 | pages = move_freepages_block(zone, page, | |
809 | start_migratetype); | |
810 | ||
811 | /* Claim the whole block if over half of it is free */ | |
d9c23400 | 812 | if (pages >= (1 << (pageblock_order-1))) |
46dafbca MG |
813 | set_pageblock_migratetype(page, |
814 | start_migratetype); | |
815 | ||
b2a0ac88 | 816 | migratetype = start_migratetype; |
c361be55 | 817 | } |
b2a0ac88 MG |
818 | |
819 | /* Remove the page from the freelists */ | |
820 | list_del(&page->lru); | |
821 | rmv_page_order(page); | |
822 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
823 | -(1UL << order)); | |
824 | ||
d9c23400 | 825 | if (current_order == pageblock_order) |
b2a0ac88 MG |
826 | set_pageblock_migratetype(page, |
827 | start_migratetype); | |
828 | ||
829 | expand(zone, page, order, current_order, area, migratetype); | |
830 | return page; | |
831 | } | |
832 | } | |
833 | ||
56fd56b8 MG |
834 | /* Use MIGRATE_RESERVE rather than fail an allocation */ |
835 | return __rmqueue_smallest(zone, order, MIGRATE_RESERVE); | |
b2a0ac88 MG |
836 | } |
837 | ||
56fd56b8 | 838 | /* |
1da177e4 LT |
839 | * Do the hard work of removing an element from the buddy allocator. |
840 | * Call me with the zone->lock already held. | |
841 | */ | |
b2a0ac88 MG |
842 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
843 | int migratetype) | |
1da177e4 | 844 | { |
1da177e4 LT |
845 | struct page *page; |
846 | ||
56fd56b8 | 847 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 848 | |
56fd56b8 MG |
849 | if (unlikely(!page)) |
850 | page = __rmqueue_fallback(zone, order, migratetype); | |
b2a0ac88 MG |
851 | |
852 | return page; | |
1da177e4 LT |
853 | } |
854 | ||
855 | /* | |
856 | * Obtain a specified number of elements from the buddy allocator, all under | |
857 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
858 | * Returns the number of new pages which were placed at *list. | |
859 | */ | |
860 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
861 | unsigned long count, struct list_head *list, |
862 | int migratetype) | |
1da177e4 | 863 | { |
1da177e4 | 864 | int i; |
1da177e4 | 865 | |
c54ad30c | 866 | spin_lock(&zone->lock); |
1da177e4 | 867 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 868 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 869 | if (unlikely(page == NULL)) |
1da177e4 | 870 | break; |
81eabcbe MG |
871 | |
872 | /* | |
873 | * Split buddy pages returned by expand() are received here | |
874 | * in physical page order. The page is added to the callers and | |
875 | * list and the list head then moves forward. From the callers | |
876 | * perspective, the linked list is ordered by page number in | |
877 | * some conditions. This is useful for IO devices that can | |
878 | * merge IO requests if the physical pages are ordered | |
879 | * properly. | |
880 | */ | |
535131e6 MG |
881 | list_add(&page->lru, list); |
882 | set_page_private(page, migratetype); | |
81eabcbe | 883 | list = &page->lru; |
1da177e4 | 884 | } |
c54ad30c | 885 | spin_unlock(&zone->lock); |
085cc7d5 | 886 | return i; |
1da177e4 LT |
887 | } |
888 | ||
4ae7c039 | 889 | #ifdef CONFIG_NUMA |
8fce4d8e | 890 | /* |
4037d452 CL |
891 | * Called from the vmstat counter updater to drain pagesets of this |
892 | * currently executing processor on remote nodes after they have | |
893 | * expired. | |
894 | * | |
879336c3 CL |
895 | * Note that this function must be called with the thread pinned to |
896 | * a single processor. | |
8fce4d8e | 897 | */ |
4037d452 | 898 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 899 | { |
4ae7c039 | 900 | unsigned long flags; |
4037d452 | 901 | int to_drain; |
4ae7c039 | 902 | |
4037d452 CL |
903 | local_irq_save(flags); |
904 | if (pcp->count >= pcp->batch) | |
905 | to_drain = pcp->batch; | |
906 | else | |
907 | to_drain = pcp->count; | |
908 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
909 | pcp->count -= to_drain; | |
910 | local_irq_restore(flags); | |
4ae7c039 CL |
911 | } |
912 | #endif | |
913 | ||
9f8f2172 CL |
914 | /* |
915 | * Drain pages of the indicated processor. | |
916 | * | |
917 | * The processor must either be the current processor and the | |
918 | * thread pinned to the current processor or a processor that | |
919 | * is not online. | |
920 | */ | |
921 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 922 | { |
c54ad30c | 923 | unsigned long flags; |
1da177e4 | 924 | struct zone *zone; |
1da177e4 | 925 | |
ee99c71c | 926 | for_each_populated_zone(zone) { |
1da177e4 | 927 | struct per_cpu_pageset *pset; |
3dfa5721 | 928 | struct per_cpu_pages *pcp; |
1da177e4 | 929 | |
e7c8d5c9 | 930 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
931 | |
932 | pcp = &pset->pcp; | |
933 | local_irq_save(flags); | |
934 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
935 | pcp->count = 0; | |
936 | local_irq_restore(flags); | |
1da177e4 LT |
937 | } |
938 | } | |
1da177e4 | 939 | |
9f8f2172 CL |
940 | /* |
941 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
942 | */ | |
943 | void drain_local_pages(void *arg) | |
944 | { | |
945 | drain_pages(smp_processor_id()); | |
946 | } | |
947 | ||
948 | /* | |
949 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
950 | */ | |
951 | void drain_all_pages(void) | |
952 | { | |
15c8b6c1 | 953 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
954 | } |
955 | ||
296699de | 956 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
957 | |
958 | void mark_free_pages(struct zone *zone) | |
959 | { | |
f623f0db RW |
960 | unsigned long pfn, max_zone_pfn; |
961 | unsigned long flags; | |
b2a0ac88 | 962 | int order, t; |
1da177e4 LT |
963 | struct list_head *curr; |
964 | ||
965 | if (!zone->spanned_pages) | |
966 | return; | |
967 | ||
968 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
969 | |
970 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
971 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
972 | if (pfn_valid(pfn)) { | |
973 | struct page *page = pfn_to_page(pfn); | |
974 | ||
7be98234 RW |
975 | if (!swsusp_page_is_forbidden(page)) |
976 | swsusp_unset_page_free(page); | |
f623f0db | 977 | } |
1da177e4 | 978 | |
b2a0ac88 MG |
979 | for_each_migratetype_order(order, t) { |
980 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 981 | unsigned long i; |
1da177e4 | 982 | |
f623f0db RW |
983 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
984 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 985 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 986 | } |
b2a0ac88 | 987 | } |
1da177e4 LT |
988 | spin_unlock_irqrestore(&zone->lock, flags); |
989 | } | |
e2c55dc8 | 990 | #endif /* CONFIG_PM */ |
1da177e4 | 991 | |
1da177e4 LT |
992 | /* |
993 | * Free a 0-order page | |
994 | */ | |
920c7a5d | 995 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
996 | { |
997 | struct zone *zone = page_zone(page); | |
998 | struct per_cpu_pages *pcp; | |
999 | unsigned long flags; | |
1000 | ||
1da177e4 LT |
1001 | if (PageAnon(page)) |
1002 | page->mapping = NULL; | |
224abf92 | 1003 | if (free_pages_check(page)) |
689bcebf HD |
1004 | return; |
1005 | ||
3ac7fe5a | 1006 | if (!PageHighMem(page)) { |
9858db50 | 1007 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); |
3ac7fe5a TG |
1008 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); |
1009 | } | |
dafb1367 | 1010 | arch_free_page(page, 0); |
689bcebf HD |
1011 | kernel_map_pages(page, 1, 0); |
1012 | ||
3dfa5721 | 1013 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
1da177e4 | 1014 | local_irq_save(flags); |
f8891e5e | 1015 | __count_vm_event(PGFREE); |
3dfa5721 CL |
1016 | if (cold) |
1017 | list_add_tail(&page->lru, &pcp->list); | |
1018 | else | |
1019 | list_add(&page->lru, &pcp->list); | |
535131e6 | 1020 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 1021 | pcp->count++; |
48db57f8 NP |
1022 | if (pcp->count >= pcp->high) { |
1023 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
1024 | pcp->count -= pcp->batch; | |
1025 | } | |
1da177e4 LT |
1026 | local_irq_restore(flags); |
1027 | put_cpu(); | |
1028 | } | |
1029 | ||
920c7a5d | 1030 | void free_hot_page(struct page *page) |
1da177e4 LT |
1031 | { |
1032 | free_hot_cold_page(page, 0); | |
1033 | } | |
1034 | ||
920c7a5d | 1035 | void free_cold_page(struct page *page) |
1da177e4 LT |
1036 | { |
1037 | free_hot_cold_page(page, 1); | |
1038 | } | |
1039 | ||
8dfcc9ba NP |
1040 | /* |
1041 | * split_page takes a non-compound higher-order page, and splits it into | |
1042 | * n (1<<order) sub-pages: page[0..n] | |
1043 | * Each sub-page must be freed individually. | |
1044 | * | |
1045 | * Note: this is probably too low level an operation for use in drivers. | |
1046 | * Please consult with lkml before using this in your driver. | |
1047 | */ | |
1048 | void split_page(struct page *page, unsigned int order) | |
1049 | { | |
1050 | int i; | |
1051 | ||
725d704e NP |
1052 | VM_BUG_ON(PageCompound(page)); |
1053 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
1054 | for (i = 1; i < (1 << order); i++) |
1055 | set_page_refcounted(page + i); | |
8dfcc9ba | 1056 | } |
8dfcc9ba | 1057 | |
1da177e4 LT |
1058 | /* |
1059 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1060 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1061 | * or two. | |
1062 | */ | |
18ea7e71 | 1063 | static struct page *buffered_rmqueue(struct zone *preferred_zone, |
a74609fa | 1064 | struct zone *zone, int order, gfp_t gfp_flags) |
1da177e4 LT |
1065 | { |
1066 | unsigned long flags; | |
689bcebf | 1067 | struct page *page; |
1da177e4 | 1068 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1069 | int cpu; |
64c5e135 | 1070 | int migratetype = allocflags_to_migratetype(gfp_flags); |
1da177e4 | 1071 | |
689bcebf | 1072 | again: |
a74609fa | 1073 | cpu = get_cpu(); |
48db57f8 | 1074 | if (likely(order == 0)) { |
1da177e4 LT |
1075 | struct per_cpu_pages *pcp; |
1076 | ||
3dfa5721 | 1077 | pcp = &zone_pcp(zone, cpu)->pcp; |
1da177e4 | 1078 | local_irq_save(flags); |
a74609fa | 1079 | if (!pcp->count) { |
941c7105 | 1080 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 1081 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
1082 | if (unlikely(!pcp->count)) |
1083 | goto failed; | |
1da177e4 | 1084 | } |
b92a6edd | 1085 | |
535131e6 | 1086 | /* Find a page of the appropriate migrate type */ |
3dfa5721 CL |
1087 | if (cold) { |
1088 | list_for_each_entry_reverse(page, &pcp->list, lru) | |
1089 | if (page_private(page) == migratetype) | |
1090 | break; | |
1091 | } else { | |
1092 | list_for_each_entry(page, &pcp->list, lru) | |
1093 | if (page_private(page) == migratetype) | |
1094 | break; | |
1095 | } | |
535131e6 | 1096 | |
b92a6edd MG |
1097 | /* Allocate more to the pcp list if necessary */ |
1098 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 MG |
1099 | pcp->count += rmqueue_bulk(zone, 0, |
1100 | pcp->batch, &pcp->list, migratetype); | |
1101 | page = list_entry(pcp->list.next, struct page, lru); | |
535131e6 | 1102 | } |
b92a6edd MG |
1103 | |
1104 | list_del(&page->lru); | |
1105 | pcp->count--; | |
7fb1d9fc | 1106 | } else { |
1da177e4 | 1107 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1108 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1109 | spin_unlock(&zone->lock); |
1110 | if (!page) | |
1111 | goto failed; | |
1da177e4 LT |
1112 | } |
1113 | ||
f8891e5e | 1114 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1115 | zone_statistics(preferred_zone, zone); |
a74609fa NP |
1116 | local_irq_restore(flags); |
1117 | put_cpu(); | |
1da177e4 | 1118 | |
725d704e | 1119 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1120 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1121 | goto again; |
1da177e4 | 1122 | return page; |
a74609fa NP |
1123 | |
1124 | failed: | |
1125 | local_irq_restore(flags); | |
1126 | put_cpu(); | |
1127 | return NULL; | |
1da177e4 LT |
1128 | } |
1129 | ||
7fb1d9fc | 1130 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
1131 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
1132 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
1133 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
1134 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1135 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1136 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1137 | |
933e312e AM |
1138 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1139 | ||
1140 | static struct fail_page_alloc_attr { | |
1141 | struct fault_attr attr; | |
1142 | ||
1143 | u32 ignore_gfp_highmem; | |
1144 | u32 ignore_gfp_wait; | |
54114994 | 1145 | u32 min_order; |
933e312e AM |
1146 | |
1147 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1148 | ||
1149 | struct dentry *ignore_gfp_highmem_file; | |
1150 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1151 | struct dentry *min_order_file; |
933e312e AM |
1152 | |
1153 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1154 | ||
1155 | } fail_page_alloc = { | |
1156 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1157 | .ignore_gfp_wait = 1, |
1158 | .ignore_gfp_highmem = 1, | |
54114994 | 1159 | .min_order = 1, |
933e312e AM |
1160 | }; |
1161 | ||
1162 | static int __init setup_fail_page_alloc(char *str) | |
1163 | { | |
1164 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1165 | } | |
1166 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1167 | ||
1168 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1169 | { | |
54114994 AM |
1170 | if (order < fail_page_alloc.min_order) |
1171 | return 0; | |
933e312e AM |
1172 | if (gfp_mask & __GFP_NOFAIL) |
1173 | return 0; | |
1174 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1175 | return 0; | |
1176 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1177 | return 0; | |
1178 | ||
1179 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1180 | } | |
1181 | ||
1182 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1183 | ||
1184 | static int __init fail_page_alloc_debugfs(void) | |
1185 | { | |
1186 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1187 | struct dentry *dir; | |
1188 | int err; | |
1189 | ||
1190 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1191 | "fail_page_alloc"); | |
1192 | if (err) | |
1193 | return err; | |
1194 | dir = fail_page_alloc.attr.dentries.dir; | |
1195 | ||
1196 | fail_page_alloc.ignore_gfp_wait_file = | |
1197 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1198 | &fail_page_alloc.ignore_gfp_wait); | |
1199 | ||
1200 | fail_page_alloc.ignore_gfp_highmem_file = | |
1201 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1202 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1203 | fail_page_alloc.min_order_file = |
1204 | debugfs_create_u32("min-order", mode, dir, | |
1205 | &fail_page_alloc.min_order); | |
933e312e AM |
1206 | |
1207 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1208 | !fail_page_alloc.ignore_gfp_highmem_file || |
1209 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1210 | err = -ENOMEM; |
1211 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1212 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1213 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1214 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1215 | } | |
1216 | ||
1217 | return err; | |
1218 | } | |
1219 | ||
1220 | late_initcall(fail_page_alloc_debugfs); | |
1221 | ||
1222 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1223 | ||
1224 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1225 | ||
1226 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1227 | { | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1232 | ||
1da177e4 LT |
1233 | /* |
1234 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1235 | * of the allocation. | |
1236 | */ | |
1237 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1238 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1239 | { |
1240 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1241 | long min = mark; |
1242 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1243 | int o; |
1244 | ||
7fb1d9fc | 1245 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1246 | min -= min / 2; |
7fb1d9fc | 1247 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1248 | min -= min / 4; |
1249 | ||
1250 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1251 | return 0; | |
1252 | for (o = 0; o < order; o++) { | |
1253 | /* At the next order, this order's pages become unavailable */ | |
1254 | free_pages -= z->free_area[o].nr_free << o; | |
1255 | ||
1256 | /* Require fewer higher order pages to be free */ | |
1257 | min >>= 1; | |
1258 | ||
1259 | if (free_pages <= min) | |
1260 | return 0; | |
1261 | } | |
1262 | return 1; | |
1263 | } | |
1264 | ||
9276b1bc PJ |
1265 | #ifdef CONFIG_NUMA |
1266 | /* | |
1267 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1268 | * skip over zones that are not allowed by the cpuset, or that have | |
1269 | * been recently (in last second) found to be nearly full. See further | |
1270 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1271 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1272 | * |
1273 | * If the zonelist cache is present in the passed in zonelist, then | |
1274 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1275 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1276 | * |
1277 | * If the zonelist cache is not available for this zonelist, does | |
1278 | * nothing and returns NULL. | |
1279 | * | |
1280 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1281 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1282 | * | |
1283 | * We hold off even calling zlc_setup, until after we've checked the | |
1284 | * first zone in the zonelist, on the theory that most allocations will | |
1285 | * be satisfied from that first zone, so best to examine that zone as | |
1286 | * quickly as we can. | |
1287 | */ | |
1288 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1289 | { | |
1290 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1291 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1292 | ||
1293 | zlc = zonelist->zlcache_ptr; | |
1294 | if (!zlc) | |
1295 | return NULL; | |
1296 | ||
f05111f5 | 1297 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1298 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1299 | zlc->last_full_zap = jiffies; | |
1300 | } | |
1301 | ||
1302 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1303 | &cpuset_current_mems_allowed : | |
37b07e41 | 1304 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1305 | return allowednodes; |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1310 | * if it is worth looking at further for free memory: | |
1311 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1312 | * bit set in the zonelist_cache fullzones BITMAP). | |
1313 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1314 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1315 | * Return true (non-zero) if zone is worth looking at further, or | |
1316 | * else return false (zero) if it is not. | |
1317 | * | |
1318 | * This check -ignores- the distinction between various watermarks, | |
1319 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1320 | * found to be full for any variation of these watermarks, it will | |
1321 | * be considered full for up to one second by all requests, unless | |
1322 | * we are so low on memory on all allowed nodes that we are forced | |
1323 | * into the second scan of the zonelist. | |
1324 | * | |
1325 | * In the second scan we ignore this zonelist cache and exactly | |
1326 | * apply the watermarks to all zones, even it is slower to do so. | |
1327 | * We are low on memory in the second scan, and should leave no stone | |
1328 | * unturned looking for a free page. | |
1329 | */ | |
dd1a239f | 1330 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1331 | nodemask_t *allowednodes) |
1332 | { | |
1333 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1334 | int i; /* index of *z in zonelist zones */ | |
1335 | int n; /* node that zone *z is on */ | |
1336 | ||
1337 | zlc = zonelist->zlcache_ptr; | |
1338 | if (!zlc) | |
1339 | return 1; | |
1340 | ||
dd1a239f | 1341 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1342 | n = zlc->z_to_n[i]; |
1343 | ||
1344 | /* This zone is worth trying if it is allowed but not full */ | |
1345 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1350 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1351 | * from that zone don't waste time re-examining it. | |
1352 | */ | |
dd1a239f | 1353 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1354 | { |
1355 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1356 | int i; /* index of *z in zonelist zones */ | |
1357 | ||
1358 | zlc = zonelist->zlcache_ptr; | |
1359 | if (!zlc) | |
1360 | return; | |
1361 | ||
dd1a239f | 1362 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1363 | |
1364 | set_bit(i, zlc->fullzones); | |
1365 | } | |
1366 | ||
1367 | #else /* CONFIG_NUMA */ | |
1368 | ||
1369 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1370 | { | |
1371 | return NULL; | |
1372 | } | |
1373 | ||
dd1a239f | 1374 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1375 | nodemask_t *allowednodes) |
1376 | { | |
1377 | return 1; | |
1378 | } | |
1379 | ||
dd1a239f | 1380 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1381 | { |
1382 | } | |
1383 | #endif /* CONFIG_NUMA */ | |
1384 | ||
7fb1d9fc | 1385 | /* |
0798e519 | 1386 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1387 | * a page. |
1388 | */ | |
1389 | static struct page * | |
19770b32 | 1390 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
54a6eb5c | 1391 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags) |
753ee728 | 1392 | { |
dd1a239f | 1393 | struct zoneref *z; |
7fb1d9fc | 1394 | struct page *page = NULL; |
54a6eb5c | 1395 | int classzone_idx; |
18ea7e71 | 1396 | struct zone *zone, *preferred_zone; |
9276b1bc PJ |
1397 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1398 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1399 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1400 | |
19770b32 MG |
1401 | (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask, |
1402 | &preferred_zone); | |
7eb54824 AW |
1403 | if (!preferred_zone) |
1404 | return NULL; | |
1405 | ||
19770b32 | 1406 | classzone_idx = zone_idx(preferred_zone); |
7fb1d9fc | 1407 | |
b3c466ce MG |
1408 | if (WARN_ON_ONCE(order >= MAX_ORDER)) |
1409 | return NULL; | |
1410 | ||
9276b1bc | 1411 | zonelist_scan: |
7fb1d9fc | 1412 | /* |
9276b1bc | 1413 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1414 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1415 | */ | |
19770b32 MG |
1416 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1417 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1418 | if (NUMA_BUILD && zlc_active && |
1419 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1420 | continue; | |
7fb1d9fc | 1421 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1422 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1423 | goto try_next_zone; |
7fb1d9fc RS |
1424 | |
1425 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1426 | unsigned long mark; |
1427 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1428 | mark = zone->pages_min; |
3148890b | 1429 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1430 | mark = zone->pages_low; |
3148890b | 1431 | else |
1192d526 | 1432 | mark = zone->pages_high; |
0798e519 PJ |
1433 | if (!zone_watermark_ok(zone, order, mark, |
1434 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1435 | if (!zone_reclaim_mode || |
1192d526 | 1436 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1437 | goto this_zone_full; |
0798e519 | 1438 | } |
7fb1d9fc RS |
1439 | } |
1440 | ||
18ea7e71 | 1441 | page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask); |
0798e519 | 1442 | if (page) |
7fb1d9fc | 1443 | break; |
9276b1bc PJ |
1444 | this_zone_full: |
1445 | if (NUMA_BUILD) | |
1446 | zlc_mark_zone_full(zonelist, z); | |
1447 | try_next_zone: | |
1448 | if (NUMA_BUILD && !did_zlc_setup) { | |
1449 | /* we do zlc_setup after the first zone is tried */ | |
1450 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1451 | zlc_active = 1; | |
1452 | did_zlc_setup = 1; | |
1453 | } | |
54a6eb5c | 1454 | } |
9276b1bc PJ |
1455 | |
1456 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1457 | /* Disable zlc cache for second zonelist scan */ | |
1458 | zlc_active = 0; | |
1459 | goto zonelist_scan; | |
1460 | } | |
7fb1d9fc | 1461 | return page; |
753ee728 MH |
1462 | } |
1463 | ||
11e33f6a MG |
1464 | static inline int |
1465 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1466 | unsigned long pages_reclaimed) | |
1da177e4 | 1467 | { |
11e33f6a MG |
1468 | /* Do not loop if specifically requested */ |
1469 | if (gfp_mask & __GFP_NORETRY) | |
1470 | return 0; | |
1da177e4 | 1471 | |
11e33f6a MG |
1472 | /* |
1473 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1474 | * means __GFP_NOFAIL, but that may not be true in other | |
1475 | * implementations. | |
1476 | */ | |
1477 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1478 | return 1; | |
1479 | ||
1480 | /* | |
1481 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1482 | * specified, then we retry until we no longer reclaim any pages | |
1483 | * (above), or we've reclaimed an order of pages at least as | |
1484 | * large as the allocation's order. In both cases, if the | |
1485 | * allocation still fails, we stop retrying. | |
1486 | */ | |
1487 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1488 | return 1; | |
cf40bd16 | 1489 | |
11e33f6a MG |
1490 | /* |
1491 | * Don't let big-order allocations loop unless the caller | |
1492 | * explicitly requests that. | |
1493 | */ | |
1494 | if (gfp_mask & __GFP_NOFAIL) | |
1495 | return 1; | |
1da177e4 | 1496 | |
11e33f6a MG |
1497 | return 0; |
1498 | } | |
933e312e | 1499 | |
11e33f6a MG |
1500 | static inline struct page * |
1501 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1502 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1503 | nodemask_t *nodemask) | |
1504 | { | |
1505 | struct page *page; | |
1506 | ||
1507 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
1508 | if (!try_set_zone_oom(zonelist, gfp_mask)) { | |
1509 | schedule_timeout_uninterruptible(1); | |
1da177e4 LT |
1510 | return NULL; |
1511 | } | |
6b1de916 | 1512 | |
11e33f6a MG |
1513 | /* |
1514 | * Go through the zonelist yet one more time, keep very high watermark | |
1515 | * here, this is only to catch a parallel oom killing, we must fail if | |
1516 | * we're still under heavy pressure. | |
1517 | */ | |
1518 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1519 | order, zonelist, high_zoneidx, | |
1520 | ALLOC_WMARK_HIGH|ALLOC_CPUSET); | |
7fb1d9fc | 1521 | if (page) |
11e33f6a MG |
1522 | goto out; |
1523 | ||
1524 | /* The OOM killer will not help higher order allocs */ | |
1525 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1526 | goto out; | |
1527 | ||
1528 | /* Exhausted what can be done so it's blamo time */ | |
1529 | out_of_memory(zonelist, gfp_mask, order); | |
1530 | ||
1531 | out: | |
1532 | clear_zonelist_oom(zonelist, gfp_mask); | |
1533 | return page; | |
1534 | } | |
1535 | ||
1536 | /* The really slow allocator path where we enter direct reclaim */ | |
1537 | static inline struct page * | |
1538 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1539 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1540 | nodemask_t *nodemask, int alloc_flags, unsigned long *did_some_progress) | |
1541 | { | |
1542 | struct page *page = NULL; | |
1543 | struct reclaim_state reclaim_state; | |
1544 | struct task_struct *p = current; | |
1545 | ||
1546 | cond_resched(); | |
1547 | ||
1548 | /* We now go into synchronous reclaim */ | |
1549 | cpuset_memory_pressure_bump(); | |
1550 | ||
1551 | /* | |
1552 | * The task's cpuset might have expanded its set of allowable nodes | |
1553 | */ | |
1554 | p->flags |= PF_MEMALLOC; | |
1555 | lockdep_set_current_reclaim_state(gfp_mask); | |
1556 | reclaim_state.reclaimed_slab = 0; | |
1557 | p->reclaim_state = &reclaim_state; | |
1558 | ||
1559 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1560 | ||
1561 | p->reclaim_state = NULL; | |
1562 | lockdep_clear_current_reclaim_state(); | |
1563 | p->flags &= ~PF_MEMALLOC; | |
1564 | ||
1565 | cond_resched(); | |
1566 | ||
1567 | if (order != 0) | |
1568 | drain_all_pages(); | |
1569 | ||
1570 | if (likely(*did_some_progress)) | |
1571 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
1572 | zonelist, high_zoneidx, alloc_flags); | |
1573 | return page; | |
1574 | } | |
1575 | ||
1576 | static inline int | |
1577 | is_allocation_high_priority(struct task_struct *p, gfp_t gfp_mask) | |
1578 | { | |
1579 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
1580 | && !in_interrupt()) | |
1581 | return 1; | |
1582 | return 0; | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * This is called in the allocator slow-path if the allocation request is of | |
1587 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1588 | */ | |
1589 | static inline struct page * | |
1590 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1591 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1592 | nodemask_t *nodemask) | |
1593 | { | |
1594 | struct page *page; | |
1595 | ||
1596 | do { | |
1597 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
1598 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS); | |
1599 | ||
1600 | if (!page && gfp_mask & __GFP_NOFAIL) | |
1601 | congestion_wait(WRITE, HZ/50); | |
1602 | } while (!page && (gfp_mask & __GFP_NOFAIL)); | |
1603 | ||
1604 | return page; | |
1605 | } | |
1606 | ||
1607 | static inline | |
1608 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
1609 | enum zone_type high_zoneidx) | |
1610 | { | |
1611 | struct zoneref *z; | |
1612 | struct zone *zone; | |
1613 | ||
1614 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) | |
1615 | wakeup_kswapd(zone, order); | |
1616 | } | |
1617 | ||
1618 | static inline struct page * | |
1619 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
1620 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1621 | nodemask_t *nodemask) | |
1622 | { | |
1623 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1624 | struct page *page = NULL; | |
1625 | int alloc_flags; | |
1626 | unsigned long pages_reclaimed = 0; | |
1627 | unsigned long did_some_progress; | |
1628 | struct task_struct *p = current; | |
1da177e4 | 1629 | |
952f3b51 CL |
1630 | /* |
1631 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1632 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1633 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1634 | * using a larger set of nodes after it has established that the | |
1635 | * allowed per node queues are empty and that nodes are | |
1636 | * over allocated. | |
1637 | */ | |
1638 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1639 | goto nopage; | |
1640 | ||
11e33f6a | 1641 | wake_all_kswapd(order, zonelist, high_zoneidx); |
1da177e4 | 1642 | |
9bf2229f | 1643 | /* |
7fb1d9fc RS |
1644 | * OK, we're below the kswapd watermark and have kicked background |
1645 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1646 | * to how we want to proceed. | |
1647 | * | |
1648 | * The caller may dip into page reserves a bit more if the caller | |
1649 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1650 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1651 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1652 | */ |
3148890b | 1653 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1654 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1655 | alloc_flags |= ALLOC_HARDER; | |
1656 | if (gfp_mask & __GFP_HIGH) | |
1657 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1658 | if (wait) |
1659 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 | 1660 | |
11e33f6a | 1661 | restart: |
1da177e4 LT |
1662 | /* |
1663 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1664 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1665 | * |
1666 | * This is the last chance, in general, before the goto nopage. | |
1667 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1668 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1669 | */ |
19770b32 | 1670 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
54a6eb5c | 1671 | high_zoneidx, alloc_flags); |
7fb1d9fc RS |
1672 | if (page) |
1673 | goto got_pg; | |
1da177e4 | 1674 | |
b43a57bb | 1675 | rebalance: |
11e33f6a MG |
1676 | /* Allocate without watermarks if the context allows */ |
1677 | if (is_allocation_high_priority(p, gfp_mask)) { | |
1678 | /* Do not dip into emergency reserves if specified */ | |
b84a35be | 1679 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { |
11e33f6a MG |
1680 | page = __alloc_pages_high_priority(gfp_mask, order, |
1681 | zonelist, high_zoneidx, nodemask); | |
7fb1d9fc RS |
1682 | if (page) |
1683 | goto got_pg; | |
1da177e4 | 1684 | } |
11e33f6a MG |
1685 | |
1686 | /* Ensure no recursion into the allocator */ | |
1da177e4 LT |
1687 | goto nopage; |
1688 | } | |
1689 | ||
1690 | /* Atomic allocations - we can't balance anything */ | |
1691 | if (!wait) | |
1692 | goto nopage; | |
1693 | ||
11e33f6a MG |
1694 | /* Try direct reclaim and then allocating */ |
1695 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
1696 | zonelist, high_zoneidx, | |
1697 | nodemask, | |
1698 | alloc_flags, &did_some_progress); | |
1699 | if (page) | |
1700 | goto got_pg; | |
1da177e4 | 1701 | |
11e33f6a MG |
1702 | /* |
1703 | * If we failed to make any progress reclaiming, then we are | |
1704 | * running out of options and have to consider going OOM | |
1705 | */ | |
1706 | if (!did_some_progress) { | |
1707 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
1708 | page = __alloc_pages_may_oom(gfp_mask, order, | |
1709 | zonelist, high_zoneidx, | |
1710 | nodemask); | |
1711 | if (page) | |
1712 | goto got_pg; | |
1da177e4 | 1713 | |
11e33f6a MG |
1714 | /* |
1715 | * The OOM killer does not trigger for high-order allocations | |
1716 | * but if no progress is being made, there are no other | |
1717 | * options and retrying is unlikely to help | |
1718 | */ | |
1719 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1720 | goto nopage; | |
e2c55dc8 | 1721 | |
ff0ceb9d DR |
1722 | goto restart; |
1723 | } | |
1da177e4 LT |
1724 | } |
1725 | ||
11e33f6a | 1726 | /* Check if we should retry the allocation */ |
a41f24ea | 1727 | pages_reclaimed += did_some_progress; |
11e33f6a MG |
1728 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
1729 | /* Wait for some write requests to complete then retry */ | |
3fcfab16 | 1730 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1731 | goto rebalance; |
1732 | } | |
1733 | ||
1734 | nopage: | |
1735 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1736 | printk(KERN_WARNING "%s: page allocation failure." | |
1737 | " order:%d, mode:0x%x\n", | |
1738 | p->comm, order, gfp_mask); | |
1739 | dump_stack(); | |
578c2fd6 | 1740 | show_mem(); |
1da177e4 | 1741 | } |
1da177e4 | 1742 | got_pg: |
1da177e4 | 1743 | return page; |
11e33f6a MG |
1744 | |
1745 | } | |
1746 | ||
1747 | /* | |
1748 | * This is the 'heart' of the zoned buddy allocator. | |
1749 | */ | |
1750 | struct page * | |
1751 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
1752 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1753 | { | |
1754 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1755 | struct page *page; | |
1756 | ||
1757 | lockdep_trace_alloc(gfp_mask); | |
1758 | ||
1759 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1760 | ||
1761 | if (should_fail_alloc_page(gfp_mask, order)) | |
1762 | return NULL; | |
1763 | ||
1764 | /* | |
1765 | * Check the zones suitable for the gfp_mask contain at least one | |
1766 | * valid zone. It's possible to have an empty zonelist as a result | |
1767 | * of GFP_THISNODE and a memoryless node | |
1768 | */ | |
1769 | if (unlikely(!zonelist->_zonerefs->zone)) | |
1770 | return NULL; | |
1771 | ||
1772 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, | |
1773 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET); | |
1774 | if (unlikely(!page)) | |
1775 | page = __alloc_pages_slowpath(gfp_mask, order, | |
1776 | zonelist, high_zoneidx, nodemask); | |
1777 | ||
1778 | return page; | |
1da177e4 | 1779 | } |
d239171e | 1780 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
1781 | |
1782 | /* | |
1783 | * Common helper functions. | |
1784 | */ | |
920c7a5d | 1785 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1786 | { |
1787 | struct page * page; | |
1788 | page = alloc_pages(gfp_mask, order); | |
1789 | if (!page) | |
1790 | return 0; | |
1791 | return (unsigned long) page_address(page); | |
1792 | } | |
1793 | ||
1794 | EXPORT_SYMBOL(__get_free_pages); | |
1795 | ||
920c7a5d | 1796 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1797 | { |
1798 | struct page * page; | |
1799 | ||
1800 | /* | |
1801 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1802 | * a highmem page | |
1803 | */ | |
725d704e | 1804 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1805 | |
1806 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1807 | if (page) | |
1808 | return (unsigned long) page_address(page); | |
1809 | return 0; | |
1810 | } | |
1811 | ||
1812 | EXPORT_SYMBOL(get_zeroed_page); | |
1813 | ||
1814 | void __pagevec_free(struct pagevec *pvec) | |
1815 | { | |
1816 | int i = pagevec_count(pvec); | |
1817 | ||
1818 | while (--i >= 0) | |
1819 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1820 | } | |
1821 | ||
920c7a5d | 1822 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 1823 | { |
b5810039 | 1824 | if (put_page_testzero(page)) { |
1da177e4 LT |
1825 | if (order == 0) |
1826 | free_hot_page(page); | |
1827 | else | |
1828 | __free_pages_ok(page, order); | |
1829 | } | |
1830 | } | |
1831 | ||
1832 | EXPORT_SYMBOL(__free_pages); | |
1833 | ||
920c7a5d | 1834 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
1835 | { |
1836 | if (addr != 0) { | |
725d704e | 1837 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1838 | __free_pages(virt_to_page((void *)addr), order); |
1839 | } | |
1840 | } | |
1841 | ||
1842 | EXPORT_SYMBOL(free_pages); | |
1843 | ||
2be0ffe2 TT |
1844 | /** |
1845 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
1846 | * @size: the number of bytes to allocate | |
1847 | * @gfp_mask: GFP flags for the allocation | |
1848 | * | |
1849 | * This function is similar to alloc_pages(), except that it allocates the | |
1850 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
1851 | * allocate memory in power-of-two pages. | |
1852 | * | |
1853 | * This function is also limited by MAX_ORDER. | |
1854 | * | |
1855 | * Memory allocated by this function must be released by free_pages_exact(). | |
1856 | */ | |
1857 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
1858 | { | |
1859 | unsigned int order = get_order(size); | |
1860 | unsigned long addr; | |
1861 | ||
1862 | addr = __get_free_pages(gfp_mask, order); | |
1863 | if (addr) { | |
1864 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
1865 | unsigned long used = addr + PAGE_ALIGN(size); | |
1866 | ||
1867 | split_page(virt_to_page(addr), order); | |
1868 | while (used < alloc_end) { | |
1869 | free_page(used); | |
1870 | used += PAGE_SIZE; | |
1871 | } | |
1872 | } | |
1873 | ||
1874 | return (void *)addr; | |
1875 | } | |
1876 | EXPORT_SYMBOL(alloc_pages_exact); | |
1877 | ||
1878 | /** | |
1879 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
1880 | * @virt: the value returned by alloc_pages_exact. | |
1881 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
1882 | * | |
1883 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
1884 | */ | |
1885 | void free_pages_exact(void *virt, size_t size) | |
1886 | { | |
1887 | unsigned long addr = (unsigned long)virt; | |
1888 | unsigned long end = addr + PAGE_ALIGN(size); | |
1889 | ||
1890 | while (addr < end) { | |
1891 | free_page(addr); | |
1892 | addr += PAGE_SIZE; | |
1893 | } | |
1894 | } | |
1895 | EXPORT_SYMBOL(free_pages_exact); | |
1896 | ||
1da177e4 LT |
1897 | static unsigned int nr_free_zone_pages(int offset) |
1898 | { | |
dd1a239f | 1899 | struct zoneref *z; |
54a6eb5c MG |
1900 | struct zone *zone; |
1901 | ||
e310fd43 | 1902 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
1903 | unsigned int sum = 0; |
1904 | ||
0e88460d | 1905 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 1906 | |
54a6eb5c | 1907 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 MB |
1908 | unsigned long size = zone->present_pages; |
1909 | unsigned long high = zone->pages_high; | |
1910 | if (size > high) | |
1911 | sum += size - high; | |
1da177e4 LT |
1912 | } |
1913 | ||
1914 | return sum; | |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1919 | */ | |
1920 | unsigned int nr_free_buffer_pages(void) | |
1921 | { | |
af4ca457 | 1922 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1923 | } |
c2f1a551 | 1924 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1925 | |
1926 | /* | |
1927 | * Amount of free RAM allocatable within all zones | |
1928 | */ | |
1929 | unsigned int nr_free_pagecache_pages(void) | |
1930 | { | |
2a1e274a | 1931 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1932 | } |
08e0f6a9 CL |
1933 | |
1934 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1935 | { |
08e0f6a9 | 1936 | if (NUMA_BUILD) |
25ba77c1 | 1937 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1938 | } |
1da177e4 | 1939 | |
1da177e4 LT |
1940 | void si_meminfo(struct sysinfo *val) |
1941 | { | |
1942 | val->totalram = totalram_pages; | |
1943 | val->sharedram = 0; | |
d23ad423 | 1944 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1945 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1946 | val->totalhigh = totalhigh_pages; |
1947 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1948 | val->mem_unit = PAGE_SIZE; |
1949 | } | |
1950 | ||
1951 | EXPORT_SYMBOL(si_meminfo); | |
1952 | ||
1953 | #ifdef CONFIG_NUMA | |
1954 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1955 | { | |
1956 | pg_data_t *pgdat = NODE_DATA(nid); | |
1957 | ||
1958 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1959 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1960 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1961 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1962 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1963 | NR_FREE_PAGES); | |
98d2b0eb CL |
1964 | #else |
1965 | val->totalhigh = 0; | |
1966 | val->freehigh = 0; | |
1967 | #endif | |
1da177e4 LT |
1968 | val->mem_unit = PAGE_SIZE; |
1969 | } | |
1970 | #endif | |
1971 | ||
1972 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1973 | ||
1974 | /* | |
1975 | * Show free area list (used inside shift_scroll-lock stuff) | |
1976 | * We also calculate the percentage fragmentation. We do this by counting the | |
1977 | * memory on each free list with the exception of the first item on the list. | |
1978 | */ | |
1979 | void show_free_areas(void) | |
1980 | { | |
c7241913 | 1981 | int cpu; |
1da177e4 LT |
1982 | struct zone *zone; |
1983 | ||
ee99c71c | 1984 | for_each_populated_zone(zone) { |
c7241913 JS |
1985 | show_node(zone); |
1986 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1987 | |
6b482c67 | 1988 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1989 | struct per_cpu_pageset *pageset; |
1990 | ||
e7c8d5c9 | 1991 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1992 | |
3dfa5721 CL |
1993 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
1994 | cpu, pageset->pcp.high, | |
1995 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
1996 | } |
1997 | } | |
1998 | ||
7b854121 LS |
1999 | printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n" |
2000 | " inactive_file:%lu" | |
2001 | //TODO: check/adjust line lengths | |
2002 | #ifdef CONFIG_UNEVICTABLE_LRU | |
2003 | " unevictable:%lu" | |
2004 | #endif | |
2005 | " dirty:%lu writeback:%lu unstable:%lu\n" | |
d23ad423 | 2006 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe RR |
2007 | global_page_state(NR_ACTIVE_ANON), |
2008 | global_page_state(NR_ACTIVE_FILE), | |
2009 | global_page_state(NR_INACTIVE_ANON), | |
2010 | global_page_state(NR_INACTIVE_FILE), | |
7b854121 LS |
2011 | #ifdef CONFIG_UNEVICTABLE_LRU |
2012 | global_page_state(NR_UNEVICTABLE), | |
2013 | #endif | |
b1e7a8fd | 2014 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2015 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2016 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2017 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
2018 | global_page_state(NR_SLAB_RECLAIMABLE) + |
2019 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2020 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
2021 | global_page_state(NR_PAGETABLE), |
2022 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2023 | |
ee99c71c | 2024 | for_each_populated_zone(zone) { |
1da177e4 LT |
2025 | int i; |
2026 | ||
2027 | show_node(zone); | |
2028 | printk("%s" | |
2029 | " free:%lukB" | |
2030 | " min:%lukB" | |
2031 | " low:%lukB" | |
2032 | " high:%lukB" | |
4f98a2fe RR |
2033 | " active_anon:%lukB" |
2034 | " inactive_anon:%lukB" | |
2035 | " active_file:%lukB" | |
2036 | " inactive_file:%lukB" | |
7b854121 LS |
2037 | #ifdef CONFIG_UNEVICTABLE_LRU |
2038 | " unevictable:%lukB" | |
2039 | #endif | |
1da177e4 LT |
2040 | " present:%lukB" |
2041 | " pages_scanned:%lu" | |
2042 | " all_unreclaimable? %s" | |
2043 | "\n", | |
2044 | zone->name, | |
d23ad423 | 2045 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
2046 | K(zone->pages_min), |
2047 | K(zone->pages_low), | |
2048 | K(zone->pages_high), | |
4f98a2fe RR |
2049 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2050 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2051 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2052 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 LS |
2053 | #ifdef CONFIG_UNEVICTABLE_LRU |
2054 | K(zone_page_state(zone, NR_UNEVICTABLE)), | |
2055 | #endif | |
1da177e4 LT |
2056 | K(zone->present_pages), |
2057 | zone->pages_scanned, | |
e815af95 | 2058 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
2059 | ); |
2060 | printk("lowmem_reserve[]:"); | |
2061 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2062 | printk(" %lu", zone->lowmem_reserve[i]); | |
2063 | printk("\n"); | |
2064 | } | |
2065 | ||
ee99c71c | 2066 | for_each_populated_zone(zone) { |
8f9de51a | 2067 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
2068 | |
2069 | show_node(zone); | |
2070 | printk("%s: ", zone->name); | |
1da177e4 LT |
2071 | |
2072 | spin_lock_irqsave(&zone->lock, flags); | |
2073 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2074 | nr[order] = zone->free_area[order].nr_free; |
2075 | total += nr[order] << order; | |
1da177e4 LT |
2076 | } |
2077 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2078 | for (order = 0; order < MAX_ORDER; order++) |
2079 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2080 | printk("= %lukB\n", K(total)); |
2081 | } | |
2082 | ||
e6f3602d LW |
2083 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2084 | ||
1da177e4 LT |
2085 | show_swap_cache_info(); |
2086 | } | |
2087 | ||
19770b32 MG |
2088 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2089 | { | |
2090 | zoneref->zone = zone; | |
2091 | zoneref->zone_idx = zone_idx(zone); | |
2092 | } | |
2093 | ||
1da177e4 LT |
2094 | /* |
2095 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2096 | * |
2097 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2098 | */ |
f0c0b2b8 KH |
2099 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2100 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2101 | { |
1a93205b CL |
2102 | struct zone *zone; |
2103 | ||
98d2b0eb | 2104 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2105 | zone_type++; |
02a68a5e CL |
2106 | |
2107 | do { | |
2f6726e5 | 2108 | zone_type--; |
070f8032 | 2109 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2110 | if (populated_zone(zone)) { |
dd1a239f MG |
2111 | zoneref_set_zone(zone, |
2112 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2113 | check_highest_zone(zone_type); |
1da177e4 | 2114 | } |
02a68a5e | 2115 | |
2f6726e5 | 2116 | } while (zone_type); |
070f8032 | 2117 | return nr_zones; |
1da177e4 LT |
2118 | } |
2119 | ||
f0c0b2b8 KH |
2120 | |
2121 | /* | |
2122 | * zonelist_order: | |
2123 | * 0 = automatic detection of better ordering. | |
2124 | * 1 = order by ([node] distance, -zonetype) | |
2125 | * 2 = order by (-zonetype, [node] distance) | |
2126 | * | |
2127 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2128 | * the same zonelist. So only NUMA can configure this param. | |
2129 | */ | |
2130 | #define ZONELIST_ORDER_DEFAULT 0 | |
2131 | #define ZONELIST_ORDER_NODE 1 | |
2132 | #define ZONELIST_ORDER_ZONE 2 | |
2133 | ||
2134 | /* zonelist order in the kernel. | |
2135 | * set_zonelist_order() will set this to NODE or ZONE. | |
2136 | */ | |
2137 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2138 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2139 | ||
2140 | ||
1da177e4 | 2141 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2142 | /* The value user specified ....changed by config */ |
2143 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2144 | /* string for sysctl */ | |
2145 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2146 | char numa_zonelist_order[16] = "default"; | |
2147 | ||
2148 | /* | |
2149 | * interface for configure zonelist ordering. | |
2150 | * command line option "numa_zonelist_order" | |
2151 | * = "[dD]efault - default, automatic configuration. | |
2152 | * = "[nN]ode - order by node locality, then by zone within node | |
2153 | * = "[zZ]one - order by zone, then by locality within zone | |
2154 | */ | |
2155 | ||
2156 | static int __parse_numa_zonelist_order(char *s) | |
2157 | { | |
2158 | if (*s == 'd' || *s == 'D') { | |
2159 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2160 | } else if (*s == 'n' || *s == 'N') { | |
2161 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2162 | } else if (*s == 'z' || *s == 'Z') { | |
2163 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2164 | } else { | |
2165 | printk(KERN_WARNING | |
2166 | "Ignoring invalid numa_zonelist_order value: " | |
2167 | "%s\n", s); | |
2168 | return -EINVAL; | |
2169 | } | |
2170 | return 0; | |
2171 | } | |
2172 | ||
2173 | static __init int setup_numa_zonelist_order(char *s) | |
2174 | { | |
2175 | if (s) | |
2176 | return __parse_numa_zonelist_order(s); | |
2177 | return 0; | |
2178 | } | |
2179 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2180 | ||
2181 | /* | |
2182 | * sysctl handler for numa_zonelist_order | |
2183 | */ | |
2184 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
2185 | struct file *file, void __user *buffer, size_t *length, | |
2186 | loff_t *ppos) | |
2187 | { | |
2188 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2189 | int ret; | |
2190 | ||
2191 | if (write) | |
2192 | strncpy(saved_string, (char*)table->data, | |
2193 | NUMA_ZONELIST_ORDER_LEN); | |
2194 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
2195 | if (ret) | |
2196 | return ret; | |
2197 | if (write) { | |
2198 | int oldval = user_zonelist_order; | |
2199 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2200 | /* | |
2201 | * bogus value. restore saved string | |
2202 | */ | |
2203 | strncpy((char*)table->data, saved_string, | |
2204 | NUMA_ZONELIST_ORDER_LEN); | |
2205 | user_zonelist_order = oldval; | |
2206 | } else if (oldval != user_zonelist_order) | |
2207 | build_all_zonelists(); | |
2208 | } | |
2209 | return 0; | |
2210 | } | |
2211 | ||
2212 | ||
1da177e4 | 2213 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
2214 | static int node_load[MAX_NUMNODES]; |
2215 | ||
1da177e4 | 2216 | /** |
4dc3b16b | 2217 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2218 | * @node: node whose fallback list we're appending |
2219 | * @used_node_mask: nodemask_t of already used nodes | |
2220 | * | |
2221 | * We use a number of factors to determine which is the next node that should | |
2222 | * appear on a given node's fallback list. The node should not have appeared | |
2223 | * already in @node's fallback list, and it should be the next closest node | |
2224 | * according to the distance array (which contains arbitrary distance values | |
2225 | * from each node to each node in the system), and should also prefer nodes | |
2226 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2227 | * on them otherwise. | |
2228 | * It returns -1 if no node is found. | |
2229 | */ | |
f0c0b2b8 | 2230 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2231 | { |
4cf808eb | 2232 | int n, val; |
1da177e4 LT |
2233 | int min_val = INT_MAX; |
2234 | int best_node = -1; | |
a70f7302 | 2235 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2236 | |
4cf808eb LT |
2237 | /* Use the local node if we haven't already */ |
2238 | if (!node_isset(node, *used_node_mask)) { | |
2239 | node_set(node, *used_node_mask); | |
2240 | return node; | |
2241 | } | |
1da177e4 | 2242 | |
37b07e41 | 2243 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2244 | |
2245 | /* Don't want a node to appear more than once */ | |
2246 | if (node_isset(n, *used_node_mask)) | |
2247 | continue; | |
2248 | ||
1da177e4 LT |
2249 | /* Use the distance array to find the distance */ |
2250 | val = node_distance(node, n); | |
2251 | ||
4cf808eb LT |
2252 | /* Penalize nodes under us ("prefer the next node") */ |
2253 | val += (n < node); | |
2254 | ||
1da177e4 | 2255 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2256 | tmp = cpumask_of_node(n); |
2257 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2258 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2259 | ||
2260 | /* Slight preference for less loaded node */ | |
2261 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2262 | val += node_load[n]; | |
2263 | ||
2264 | if (val < min_val) { | |
2265 | min_val = val; | |
2266 | best_node = n; | |
2267 | } | |
2268 | } | |
2269 | ||
2270 | if (best_node >= 0) | |
2271 | node_set(best_node, *used_node_mask); | |
2272 | ||
2273 | return best_node; | |
2274 | } | |
2275 | ||
f0c0b2b8 KH |
2276 | |
2277 | /* | |
2278 | * Build zonelists ordered by node and zones within node. | |
2279 | * This results in maximum locality--normal zone overflows into local | |
2280 | * DMA zone, if any--but risks exhausting DMA zone. | |
2281 | */ | |
2282 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2283 | { |
f0c0b2b8 | 2284 | int j; |
1da177e4 | 2285 | struct zonelist *zonelist; |
f0c0b2b8 | 2286 | |
54a6eb5c | 2287 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2288 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2289 | ; |
2290 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2291 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2292 | zonelist->_zonerefs[j].zone = NULL; |
2293 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2294 | } |
2295 | ||
523b9458 CL |
2296 | /* |
2297 | * Build gfp_thisnode zonelists | |
2298 | */ | |
2299 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2300 | { | |
523b9458 CL |
2301 | int j; |
2302 | struct zonelist *zonelist; | |
2303 | ||
54a6eb5c MG |
2304 | zonelist = &pgdat->node_zonelists[1]; |
2305 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2306 | zonelist->_zonerefs[j].zone = NULL; |
2307 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2308 | } |
2309 | ||
f0c0b2b8 KH |
2310 | /* |
2311 | * Build zonelists ordered by zone and nodes within zones. | |
2312 | * This results in conserving DMA zone[s] until all Normal memory is | |
2313 | * exhausted, but results in overflowing to remote node while memory | |
2314 | * may still exist in local DMA zone. | |
2315 | */ | |
2316 | static int node_order[MAX_NUMNODES]; | |
2317 | ||
2318 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2319 | { | |
f0c0b2b8 KH |
2320 | int pos, j, node; |
2321 | int zone_type; /* needs to be signed */ | |
2322 | struct zone *z; | |
2323 | struct zonelist *zonelist; | |
2324 | ||
54a6eb5c MG |
2325 | zonelist = &pgdat->node_zonelists[0]; |
2326 | pos = 0; | |
2327 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2328 | for (j = 0; j < nr_nodes; j++) { | |
2329 | node = node_order[j]; | |
2330 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2331 | if (populated_zone(z)) { | |
dd1a239f MG |
2332 | zoneref_set_zone(z, |
2333 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2334 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2335 | } |
2336 | } | |
f0c0b2b8 | 2337 | } |
dd1a239f MG |
2338 | zonelist->_zonerefs[pos].zone = NULL; |
2339 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2340 | } |
2341 | ||
2342 | static int default_zonelist_order(void) | |
2343 | { | |
2344 | int nid, zone_type; | |
2345 | unsigned long low_kmem_size,total_size; | |
2346 | struct zone *z; | |
2347 | int average_size; | |
2348 | /* | |
2349 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2350 | * If they are really small and used heavily, the system can fall | |
2351 | * into OOM very easily. | |
2352 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2353 | */ | |
2354 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2355 | low_kmem_size = 0; | |
2356 | total_size = 0; | |
2357 | for_each_online_node(nid) { | |
2358 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2359 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2360 | if (populated_zone(z)) { | |
2361 | if (zone_type < ZONE_NORMAL) | |
2362 | low_kmem_size += z->present_pages; | |
2363 | total_size += z->present_pages; | |
2364 | } | |
2365 | } | |
2366 | } | |
2367 | if (!low_kmem_size || /* there are no DMA area. */ | |
2368 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2369 | return ZONELIST_ORDER_NODE; | |
2370 | /* | |
2371 | * look into each node's config. | |
2372 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2373 | * local memory, NODE_ORDER may be suitable. | |
2374 | */ | |
37b07e41 LS |
2375 | average_size = total_size / |
2376 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2377 | for_each_online_node(nid) { |
2378 | low_kmem_size = 0; | |
2379 | total_size = 0; | |
2380 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2381 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2382 | if (populated_zone(z)) { | |
2383 | if (zone_type < ZONE_NORMAL) | |
2384 | low_kmem_size += z->present_pages; | |
2385 | total_size += z->present_pages; | |
2386 | } | |
2387 | } | |
2388 | if (low_kmem_size && | |
2389 | total_size > average_size && /* ignore small node */ | |
2390 | low_kmem_size > total_size * 70/100) | |
2391 | return ZONELIST_ORDER_NODE; | |
2392 | } | |
2393 | return ZONELIST_ORDER_ZONE; | |
2394 | } | |
2395 | ||
2396 | static void set_zonelist_order(void) | |
2397 | { | |
2398 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2399 | current_zonelist_order = default_zonelist_order(); | |
2400 | else | |
2401 | current_zonelist_order = user_zonelist_order; | |
2402 | } | |
2403 | ||
2404 | static void build_zonelists(pg_data_t *pgdat) | |
2405 | { | |
2406 | int j, node, load; | |
2407 | enum zone_type i; | |
1da177e4 | 2408 | nodemask_t used_mask; |
f0c0b2b8 KH |
2409 | int local_node, prev_node; |
2410 | struct zonelist *zonelist; | |
2411 | int order = current_zonelist_order; | |
1da177e4 LT |
2412 | |
2413 | /* initialize zonelists */ | |
523b9458 | 2414 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2415 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2416 | zonelist->_zonerefs[0].zone = NULL; |
2417 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2418 | } |
2419 | ||
2420 | /* NUMA-aware ordering of nodes */ | |
2421 | local_node = pgdat->node_id; | |
2422 | load = num_online_nodes(); | |
2423 | prev_node = local_node; | |
2424 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2425 | |
2426 | memset(node_load, 0, sizeof(node_load)); | |
2427 | memset(node_order, 0, sizeof(node_order)); | |
2428 | j = 0; | |
2429 | ||
1da177e4 | 2430 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2431 | int distance = node_distance(local_node, node); |
2432 | ||
2433 | /* | |
2434 | * If another node is sufficiently far away then it is better | |
2435 | * to reclaim pages in a zone before going off node. | |
2436 | */ | |
2437 | if (distance > RECLAIM_DISTANCE) | |
2438 | zone_reclaim_mode = 1; | |
2439 | ||
1da177e4 LT |
2440 | /* |
2441 | * We don't want to pressure a particular node. | |
2442 | * So adding penalty to the first node in same | |
2443 | * distance group to make it round-robin. | |
2444 | */ | |
9eeff239 | 2445 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2446 | node_load[node] = load; |
2447 | ||
1da177e4 LT |
2448 | prev_node = node; |
2449 | load--; | |
f0c0b2b8 KH |
2450 | if (order == ZONELIST_ORDER_NODE) |
2451 | build_zonelists_in_node_order(pgdat, node); | |
2452 | else | |
2453 | node_order[j++] = node; /* remember order */ | |
2454 | } | |
1da177e4 | 2455 | |
f0c0b2b8 KH |
2456 | if (order == ZONELIST_ORDER_ZONE) { |
2457 | /* calculate node order -- i.e., DMA last! */ | |
2458 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2459 | } |
523b9458 CL |
2460 | |
2461 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2462 | } |
2463 | ||
9276b1bc | 2464 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2465 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2466 | { |
54a6eb5c MG |
2467 | struct zonelist *zonelist; |
2468 | struct zonelist_cache *zlc; | |
dd1a239f | 2469 | struct zoneref *z; |
9276b1bc | 2470 | |
54a6eb5c MG |
2471 | zonelist = &pgdat->node_zonelists[0]; |
2472 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2473 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2474 | for (z = zonelist->_zonerefs; z->zone; z++) |
2475 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2476 | } |
2477 | ||
f0c0b2b8 | 2478 | |
1da177e4 LT |
2479 | #else /* CONFIG_NUMA */ |
2480 | ||
f0c0b2b8 KH |
2481 | static void set_zonelist_order(void) |
2482 | { | |
2483 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2484 | } | |
2485 | ||
2486 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2487 | { |
19655d34 | 2488 | int node, local_node; |
54a6eb5c MG |
2489 | enum zone_type j; |
2490 | struct zonelist *zonelist; | |
1da177e4 LT |
2491 | |
2492 | local_node = pgdat->node_id; | |
1da177e4 | 2493 | |
54a6eb5c MG |
2494 | zonelist = &pgdat->node_zonelists[0]; |
2495 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2496 | |
54a6eb5c MG |
2497 | /* |
2498 | * Now we build the zonelist so that it contains the zones | |
2499 | * of all the other nodes. | |
2500 | * We don't want to pressure a particular node, so when | |
2501 | * building the zones for node N, we make sure that the | |
2502 | * zones coming right after the local ones are those from | |
2503 | * node N+1 (modulo N) | |
2504 | */ | |
2505 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2506 | if (!node_online(node)) | |
2507 | continue; | |
2508 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2509 | MAX_NR_ZONES - 1); | |
1da177e4 | 2510 | } |
54a6eb5c MG |
2511 | for (node = 0; node < local_node; node++) { |
2512 | if (!node_online(node)) | |
2513 | continue; | |
2514 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2515 | MAX_NR_ZONES - 1); | |
2516 | } | |
2517 | ||
dd1a239f MG |
2518 | zonelist->_zonerefs[j].zone = NULL; |
2519 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
2520 | } |
2521 | ||
9276b1bc | 2522 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2523 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2524 | { |
54a6eb5c | 2525 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
2526 | } |
2527 | ||
1da177e4 LT |
2528 | #endif /* CONFIG_NUMA */ |
2529 | ||
9b1a4d38 | 2530 | /* return values int ....just for stop_machine() */ |
f0c0b2b8 | 2531 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2532 | { |
6811378e | 2533 | int nid; |
9276b1bc PJ |
2534 | |
2535 | for_each_online_node(nid) { | |
7ea1530a CL |
2536 | pg_data_t *pgdat = NODE_DATA(nid); |
2537 | ||
2538 | build_zonelists(pgdat); | |
2539 | build_zonelist_cache(pgdat); | |
9276b1bc | 2540 | } |
6811378e YG |
2541 | return 0; |
2542 | } | |
2543 | ||
f0c0b2b8 | 2544 | void build_all_zonelists(void) |
6811378e | 2545 | { |
f0c0b2b8 KH |
2546 | set_zonelist_order(); |
2547 | ||
6811378e | 2548 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2549 | __build_all_zonelists(NULL); |
68ad8df4 | 2550 | mminit_verify_zonelist(); |
6811378e YG |
2551 | cpuset_init_current_mems_allowed(); |
2552 | } else { | |
183ff22b | 2553 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 2554 | of zonelist */ |
9b1a4d38 | 2555 | stop_machine(__build_all_zonelists, NULL, NULL); |
6811378e YG |
2556 | /* cpuset refresh routine should be here */ |
2557 | } | |
bd1e22b8 | 2558 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2559 | /* |
2560 | * Disable grouping by mobility if the number of pages in the | |
2561 | * system is too low to allow the mechanism to work. It would be | |
2562 | * more accurate, but expensive to check per-zone. This check is | |
2563 | * made on memory-hotadd so a system can start with mobility | |
2564 | * disabled and enable it later | |
2565 | */ | |
d9c23400 | 2566 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2567 | page_group_by_mobility_disabled = 1; |
2568 | else | |
2569 | page_group_by_mobility_disabled = 0; | |
2570 | ||
2571 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2572 | "Total pages: %ld\n", | |
f0c0b2b8 KH |
2573 | num_online_nodes(), |
2574 | zonelist_order_name[current_zonelist_order], | |
9ef9acb0 | 2575 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2576 | vm_total_pages); |
2577 | #ifdef CONFIG_NUMA | |
2578 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2579 | #endif | |
1da177e4 LT |
2580 | } |
2581 | ||
2582 | /* | |
2583 | * Helper functions to size the waitqueue hash table. | |
2584 | * Essentially these want to choose hash table sizes sufficiently | |
2585 | * large so that collisions trying to wait on pages are rare. | |
2586 | * But in fact, the number of active page waitqueues on typical | |
2587 | * systems is ridiculously low, less than 200. So this is even | |
2588 | * conservative, even though it seems large. | |
2589 | * | |
2590 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2591 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2592 | */ | |
2593 | #define PAGES_PER_WAITQUEUE 256 | |
2594 | ||
cca448fe | 2595 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2596 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2597 | { |
2598 | unsigned long size = 1; | |
2599 | ||
2600 | pages /= PAGES_PER_WAITQUEUE; | |
2601 | ||
2602 | while (size < pages) | |
2603 | size <<= 1; | |
2604 | ||
2605 | /* | |
2606 | * Once we have dozens or even hundreds of threads sleeping | |
2607 | * on IO we've got bigger problems than wait queue collision. | |
2608 | * Limit the size of the wait table to a reasonable size. | |
2609 | */ | |
2610 | size = min(size, 4096UL); | |
2611 | ||
2612 | return max(size, 4UL); | |
2613 | } | |
cca448fe YG |
2614 | #else |
2615 | /* | |
2616 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2617 | * a suitable size for its wait_table. So we use the maximum size now. | |
2618 | * | |
2619 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2620 | * | |
2621 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2622 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2623 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2624 | * | |
2625 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2626 | * or more by the traditional way. (See above). It equals: | |
2627 | * | |
2628 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2629 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2630 | * powerpc (64K page size) : = (32G +16M)byte. | |
2631 | */ | |
2632 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2633 | { | |
2634 | return 4096UL; | |
2635 | } | |
2636 | #endif | |
1da177e4 LT |
2637 | |
2638 | /* | |
2639 | * This is an integer logarithm so that shifts can be used later | |
2640 | * to extract the more random high bits from the multiplicative | |
2641 | * hash function before the remainder is taken. | |
2642 | */ | |
2643 | static inline unsigned long wait_table_bits(unsigned long size) | |
2644 | { | |
2645 | return ffz(~size); | |
2646 | } | |
2647 | ||
2648 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2649 | ||
56fd56b8 | 2650 | /* |
d9c23400 | 2651 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
56fd56b8 MG |
2652 | * of blocks reserved is based on zone->pages_min. The memory within the |
2653 | * reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
2654 | * higher will lead to a bigger reserve which will get freed as contiguous | |
2655 | * blocks as reclaim kicks in | |
2656 | */ | |
2657 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2658 | { | |
2659 | unsigned long start_pfn, pfn, end_pfn; | |
2660 | struct page *page; | |
2661 | unsigned long reserve, block_migratetype; | |
2662 | ||
2663 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2664 | start_pfn = zone->zone_start_pfn; | |
2665 | end_pfn = start_pfn + zone->spanned_pages; | |
d9c23400 MG |
2666 | reserve = roundup(zone->pages_min, pageblock_nr_pages) >> |
2667 | pageblock_order; | |
56fd56b8 | 2668 | |
d9c23400 | 2669 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2670 | if (!pfn_valid(pfn)) |
2671 | continue; | |
2672 | page = pfn_to_page(pfn); | |
2673 | ||
344c790e AL |
2674 | /* Watch out for overlapping nodes */ |
2675 | if (page_to_nid(page) != zone_to_nid(zone)) | |
2676 | continue; | |
2677 | ||
56fd56b8 MG |
2678 | /* Blocks with reserved pages will never free, skip them. */ |
2679 | if (PageReserved(page)) | |
2680 | continue; | |
2681 | ||
2682 | block_migratetype = get_pageblock_migratetype(page); | |
2683 | ||
2684 | /* If this block is reserved, account for it */ | |
2685 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2686 | reserve--; | |
2687 | continue; | |
2688 | } | |
2689 | ||
2690 | /* Suitable for reserving if this block is movable */ | |
2691 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2692 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2693 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2694 | reserve--; | |
2695 | continue; | |
2696 | } | |
2697 | ||
2698 | /* | |
2699 | * If the reserve is met and this is a previous reserved block, | |
2700 | * take it back | |
2701 | */ | |
2702 | if (block_migratetype == MIGRATE_RESERVE) { | |
2703 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2704 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2705 | } | |
2706 | } | |
2707 | } | |
ac0e5b7a | 2708 | |
1da177e4 LT |
2709 | /* |
2710 | * Initially all pages are reserved - free ones are freed | |
2711 | * up by free_all_bootmem() once the early boot process is | |
2712 | * done. Non-atomic initialization, single-pass. | |
2713 | */ | |
c09b4240 | 2714 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2715 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2716 | { |
1da177e4 | 2717 | struct page *page; |
29751f69 AW |
2718 | unsigned long end_pfn = start_pfn + size; |
2719 | unsigned long pfn; | |
86051ca5 | 2720 | struct zone *z; |
1da177e4 | 2721 | |
22b31eec HD |
2722 | if (highest_memmap_pfn < end_pfn - 1) |
2723 | highest_memmap_pfn = end_pfn - 1; | |
2724 | ||
86051ca5 | 2725 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 2726 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2727 | /* |
2728 | * There can be holes in boot-time mem_map[]s | |
2729 | * handed to this function. They do not | |
2730 | * exist on hotplugged memory. | |
2731 | */ | |
2732 | if (context == MEMMAP_EARLY) { | |
2733 | if (!early_pfn_valid(pfn)) | |
2734 | continue; | |
2735 | if (!early_pfn_in_nid(pfn, nid)) | |
2736 | continue; | |
2737 | } | |
d41dee36 AW |
2738 | page = pfn_to_page(pfn); |
2739 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 2740 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 2741 | init_page_count(page); |
1da177e4 LT |
2742 | reset_page_mapcount(page); |
2743 | SetPageReserved(page); | |
b2a0ac88 MG |
2744 | /* |
2745 | * Mark the block movable so that blocks are reserved for | |
2746 | * movable at startup. This will force kernel allocations | |
2747 | * to reserve their blocks rather than leaking throughout | |
2748 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2749 | * kernel allocations are made. Later some blocks near |
2750 | * the start are marked MIGRATE_RESERVE by | |
2751 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
2752 | * |
2753 | * bitmap is created for zone's valid pfn range. but memmap | |
2754 | * can be created for invalid pages (for alignment) | |
2755 | * check here not to call set_pageblock_migratetype() against | |
2756 | * pfn out of zone. | |
b2a0ac88 | 2757 | */ |
86051ca5 KH |
2758 | if ((z->zone_start_pfn <= pfn) |
2759 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
2760 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 2761 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2762 | |
1da177e4 LT |
2763 | INIT_LIST_HEAD(&page->lru); |
2764 | #ifdef WANT_PAGE_VIRTUAL | |
2765 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2766 | if (!is_highmem_idx(zone)) | |
3212c6be | 2767 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2768 | #endif |
1da177e4 LT |
2769 | } |
2770 | } | |
2771 | ||
1e548deb | 2772 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 2773 | { |
b2a0ac88 MG |
2774 | int order, t; |
2775 | for_each_migratetype_order(order, t) { | |
2776 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2777 | zone->free_area[order].nr_free = 0; |
2778 | } | |
2779 | } | |
2780 | ||
2781 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2782 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2783 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2784 | #endif |
2785 | ||
1d6f4e60 | 2786 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 2787 | { |
3a6be87f | 2788 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
2789 | int batch; |
2790 | ||
2791 | /* | |
2792 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2793 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2794 | * |
2795 | * OK, so we don't know how big the cache is. So guess. | |
2796 | */ | |
2797 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2798 | if (batch * PAGE_SIZE > 512 * 1024) |
2799 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2800 | batch /= 4; /* We effectively *= 4 below */ |
2801 | if (batch < 1) | |
2802 | batch = 1; | |
2803 | ||
2804 | /* | |
0ceaacc9 NP |
2805 | * Clamp the batch to a 2^n - 1 value. Having a power |
2806 | * of 2 value was found to be more likely to have | |
2807 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2808 | * |
0ceaacc9 NP |
2809 | * For example if 2 tasks are alternately allocating |
2810 | * batches of pages, one task can end up with a lot | |
2811 | * of pages of one half of the possible page colors | |
2812 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2813 | */ |
9155203a | 2814 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 2815 | |
e7c8d5c9 | 2816 | return batch; |
3a6be87f DH |
2817 | |
2818 | #else | |
2819 | /* The deferral and batching of frees should be suppressed under NOMMU | |
2820 | * conditions. | |
2821 | * | |
2822 | * The problem is that NOMMU needs to be able to allocate large chunks | |
2823 | * of contiguous memory as there's no hardware page translation to | |
2824 | * assemble apparent contiguous memory from discontiguous pages. | |
2825 | * | |
2826 | * Queueing large contiguous runs of pages for batching, however, | |
2827 | * causes the pages to actually be freed in smaller chunks. As there | |
2828 | * can be a significant delay between the individual batches being | |
2829 | * recycled, this leads to the once large chunks of space being | |
2830 | * fragmented and becoming unavailable for high-order allocations. | |
2831 | */ | |
2832 | return 0; | |
2833 | #endif | |
e7c8d5c9 CL |
2834 | } |
2835 | ||
b69a7288 | 2836 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
2837 | { |
2838 | struct per_cpu_pages *pcp; | |
2839 | ||
1c6fe946 MD |
2840 | memset(p, 0, sizeof(*p)); |
2841 | ||
3dfa5721 | 2842 | pcp = &p->pcp; |
2caaad41 | 2843 | pcp->count = 0; |
2caaad41 CL |
2844 | pcp->high = 6 * batch; |
2845 | pcp->batch = max(1UL, 1 * batch); | |
2846 | INIT_LIST_HEAD(&pcp->list); | |
2caaad41 CL |
2847 | } |
2848 | ||
8ad4b1fb RS |
2849 | /* |
2850 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2851 | * to the value high for the pageset p. | |
2852 | */ | |
2853 | ||
2854 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2855 | unsigned long high) | |
2856 | { | |
2857 | struct per_cpu_pages *pcp; | |
2858 | ||
3dfa5721 | 2859 | pcp = &p->pcp; |
8ad4b1fb RS |
2860 | pcp->high = high; |
2861 | pcp->batch = max(1UL, high/4); | |
2862 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2863 | pcp->batch = PAGE_SHIFT * 8; | |
2864 | } | |
2865 | ||
2866 | ||
e7c8d5c9 CL |
2867 | #ifdef CONFIG_NUMA |
2868 | /* | |
2caaad41 CL |
2869 | * Boot pageset table. One per cpu which is going to be used for all |
2870 | * zones and all nodes. The parameters will be set in such a way | |
2871 | * that an item put on a list will immediately be handed over to | |
2872 | * the buddy list. This is safe since pageset manipulation is done | |
2873 | * with interrupts disabled. | |
2874 | * | |
2875 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2876 | * |
2877 | * The boot_pagesets must be kept even after bootup is complete for | |
2878 | * unused processors and/or zones. They do play a role for bootstrapping | |
2879 | * hotplugged processors. | |
2880 | * | |
2881 | * zoneinfo_show() and maybe other functions do | |
2882 | * not check if the processor is online before following the pageset pointer. | |
2883 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2884 | */ |
88a2a4ac | 2885 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2886 | |
2887 | /* | |
2888 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2889 | * per cpu pageset array in struct zone. |
2890 | */ | |
6292d9aa | 2891 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2892 | { |
2893 | struct zone *zone, *dzone; | |
37c0708d CL |
2894 | int node = cpu_to_node(cpu); |
2895 | ||
2896 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 | 2897 | |
ee99c71c | 2898 | for_each_populated_zone(zone) { |
23316bc8 | 2899 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2900 | GFP_KERNEL, node); |
23316bc8 | 2901 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2902 | goto bad; |
e7c8d5c9 | 2903 | |
23316bc8 | 2904 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2905 | |
2906 | if (percpu_pagelist_fraction) | |
2907 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2908 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2909 | } |
2910 | ||
2911 | return 0; | |
2912 | bad: | |
2913 | for_each_zone(dzone) { | |
64191688 AM |
2914 | if (!populated_zone(dzone)) |
2915 | continue; | |
e7c8d5c9 CL |
2916 | if (dzone == zone) |
2917 | break; | |
23316bc8 NP |
2918 | kfree(zone_pcp(dzone, cpu)); |
2919 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2920 | } |
2921 | return -ENOMEM; | |
2922 | } | |
2923 | ||
2924 | static inline void free_zone_pagesets(int cpu) | |
2925 | { | |
e7c8d5c9 CL |
2926 | struct zone *zone; |
2927 | ||
2928 | for_each_zone(zone) { | |
2929 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2930 | ||
f3ef9ead DR |
2931 | /* Free per_cpu_pageset if it is slab allocated */ |
2932 | if (pset != &boot_pageset[cpu]) | |
2933 | kfree(pset); | |
e7c8d5c9 | 2934 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2935 | } |
e7c8d5c9 CL |
2936 | } |
2937 | ||
9c7b216d | 2938 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2939 | unsigned long action, |
2940 | void *hcpu) | |
2941 | { | |
2942 | int cpu = (long)hcpu; | |
2943 | int ret = NOTIFY_OK; | |
2944 | ||
2945 | switch (action) { | |
ce421c79 | 2946 | case CPU_UP_PREPARE: |
8bb78442 | 2947 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2948 | if (process_zones(cpu)) |
2949 | ret = NOTIFY_BAD; | |
2950 | break; | |
2951 | case CPU_UP_CANCELED: | |
8bb78442 | 2952 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2953 | case CPU_DEAD: |
8bb78442 | 2954 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2955 | free_zone_pagesets(cpu); |
2956 | break; | |
2957 | default: | |
2958 | break; | |
e7c8d5c9 CL |
2959 | } |
2960 | return ret; | |
2961 | } | |
2962 | ||
74b85f37 | 2963 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2964 | { &pageset_cpuup_callback, NULL, 0 }; |
2965 | ||
78d9955b | 2966 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2967 | { |
2968 | int err; | |
2969 | ||
2970 | /* Initialize per_cpu_pageset for cpu 0. | |
2971 | * A cpuup callback will do this for every cpu | |
2972 | * as it comes online | |
2973 | */ | |
2974 | err = process_zones(smp_processor_id()); | |
2975 | BUG_ON(err); | |
2976 | register_cpu_notifier(&pageset_notifier); | |
2977 | } | |
2978 | ||
2979 | #endif | |
2980 | ||
577a32f6 | 2981 | static noinline __init_refok |
cca448fe | 2982 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2983 | { |
2984 | int i; | |
2985 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2986 | size_t alloc_size; |
ed8ece2e DH |
2987 | |
2988 | /* | |
2989 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2990 | * per zone. | |
2991 | */ | |
02b694de YG |
2992 | zone->wait_table_hash_nr_entries = |
2993 | wait_table_hash_nr_entries(zone_size_pages); | |
2994 | zone->wait_table_bits = | |
2995 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2996 | alloc_size = zone->wait_table_hash_nr_entries |
2997 | * sizeof(wait_queue_head_t); | |
2998 | ||
cd94b9db | 2999 | if (!slab_is_available()) { |
cca448fe YG |
3000 | zone->wait_table = (wait_queue_head_t *) |
3001 | alloc_bootmem_node(pgdat, alloc_size); | |
3002 | } else { | |
3003 | /* | |
3004 | * This case means that a zone whose size was 0 gets new memory | |
3005 | * via memory hot-add. | |
3006 | * But it may be the case that a new node was hot-added. In | |
3007 | * this case vmalloc() will not be able to use this new node's | |
3008 | * memory - this wait_table must be initialized to use this new | |
3009 | * node itself as well. | |
3010 | * To use this new node's memory, further consideration will be | |
3011 | * necessary. | |
3012 | */ | |
8691f3a7 | 3013 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3014 | } |
3015 | if (!zone->wait_table) | |
3016 | return -ENOMEM; | |
ed8ece2e | 3017 | |
02b694de | 3018 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3019 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3020 | |
3021 | return 0; | |
ed8ece2e DH |
3022 | } |
3023 | ||
c09b4240 | 3024 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
3025 | { |
3026 | int cpu; | |
3027 | unsigned long batch = zone_batchsize(zone); | |
3028 | ||
3029 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3030 | #ifdef CONFIG_NUMA | |
3031 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 3032 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
3033 | setup_pageset(&boot_pageset[cpu],0); |
3034 | #else | |
3035 | setup_pageset(zone_pcp(zone,cpu), batch); | |
3036 | #endif | |
3037 | } | |
f5335c0f AB |
3038 | if (zone->present_pages) |
3039 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
3040 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
3041 | } |
3042 | ||
718127cc YG |
3043 | __meminit int init_currently_empty_zone(struct zone *zone, |
3044 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3045 | unsigned long size, |
3046 | enum memmap_context context) | |
ed8ece2e DH |
3047 | { |
3048 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3049 | int ret; |
3050 | ret = zone_wait_table_init(zone, size); | |
3051 | if (ret) | |
3052 | return ret; | |
ed8ece2e DH |
3053 | pgdat->nr_zones = zone_idx(zone) + 1; |
3054 | ||
ed8ece2e DH |
3055 | zone->zone_start_pfn = zone_start_pfn; |
3056 | ||
708614e6 MG |
3057 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3058 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3059 | pgdat->node_id, | |
3060 | (unsigned long)zone_idx(zone), | |
3061 | zone_start_pfn, (zone_start_pfn + size)); | |
3062 | ||
1e548deb | 3063 | zone_init_free_lists(zone); |
718127cc YG |
3064 | |
3065 | return 0; | |
ed8ece2e DH |
3066 | } |
3067 | ||
c713216d MG |
3068 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3069 | /* | |
3070 | * Basic iterator support. Return the first range of PFNs for a node | |
3071 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3072 | */ | |
a3142c8e | 3073 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
3074 | { |
3075 | int i; | |
3076 | ||
3077 | for (i = 0; i < nr_nodemap_entries; i++) | |
3078 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3079 | return i; | |
3080 | ||
3081 | return -1; | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 3086 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 3087 | */ |
a3142c8e | 3088 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
3089 | { |
3090 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3091 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3092 | return index; | |
3093 | ||
3094 | return -1; | |
3095 | } | |
3096 | ||
3097 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3098 | /* | |
3099 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3100 | * Architectures may implement their own version but if add_active_range() | |
3101 | * was used and there are no special requirements, this is a convenient | |
3102 | * alternative | |
3103 | */ | |
f2dbcfa7 | 3104 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
3105 | { |
3106 | int i; | |
3107 | ||
3108 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3109 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3110 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3111 | ||
3112 | if (start_pfn <= pfn && pfn < end_pfn) | |
3113 | return early_node_map[i].nid; | |
3114 | } | |
cc2559bc KH |
3115 | /* This is a memory hole */ |
3116 | return -1; | |
c713216d MG |
3117 | } |
3118 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3119 | ||
f2dbcfa7 KH |
3120 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3121 | { | |
cc2559bc KH |
3122 | int nid; |
3123 | ||
3124 | nid = __early_pfn_to_nid(pfn); | |
3125 | if (nid >= 0) | |
3126 | return nid; | |
3127 | /* just returns 0 */ | |
3128 | return 0; | |
f2dbcfa7 KH |
3129 | } |
3130 | ||
cc2559bc KH |
3131 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3132 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3133 | { | |
3134 | int nid; | |
3135 | ||
3136 | nid = __early_pfn_to_nid(pfn); | |
3137 | if (nid >= 0 && nid != node) | |
3138 | return false; | |
3139 | return true; | |
3140 | } | |
3141 | #endif | |
f2dbcfa7 | 3142 | |
c713216d MG |
3143 | /* Basic iterator support to walk early_node_map[] */ |
3144 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3145 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3146 | i = next_active_region_index_in_nid(i, nid)) | |
3147 | ||
3148 | /** | |
3149 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3150 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3151 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3152 | * |
3153 | * If an architecture guarantees that all ranges registered with | |
3154 | * add_active_ranges() contain no holes and may be freed, this | |
3155 | * this function may be used instead of calling free_bootmem() manually. | |
3156 | */ | |
3157 | void __init free_bootmem_with_active_regions(int nid, | |
3158 | unsigned long max_low_pfn) | |
3159 | { | |
3160 | int i; | |
3161 | ||
3162 | for_each_active_range_index_in_nid(i, nid) { | |
3163 | unsigned long size_pages = 0; | |
3164 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3165 | ||
3166 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3167 | continue; | |
3168 | ||
3169 | if (end_pfn > max_low_pfn) | |
3170 | end_pfn = max_low_pfn; | |
3171 | ||
3172 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3173 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3174 | PFN_PHYS(early_node_map[i].start_pfn), | |
3175 | size_pages << PAGE_SHIFT); | |
3176 | } | |
3177 | } | |
3178 | ||
b5bc6c0e YL |
3179 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
3180 | { | |
3181 | int i; | |
d52d53b8 | 3182 | int ret; |
b5bc6c0e | 3183 | |
d52d53b8 YL |
3184 | for_each_active_range_index_in_nid(i, nid) { |
3185 | ret = work_fn(early_node_map[i].start_pfn, | |
3186 | early_node_map[i].end_pfn, data); | |
3187 | if (ret) | |
3188 | break; | |
3189 | } | |
b5bc6c0e | 3190 | } |
c713216d MG |
3191 | /** |
3192 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3193 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3194 | * |
3195 | * If an architecture guarantees that all ranges registered with | |
3196 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3197 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3198 | */ |
3199 | void __init sparse_memory_present_with_active_regions(int nid) | |
3200 | { | |
3201 | int i; | |
3202 | ||
3203 | for_each_active_range_index_in_nid(i, nid) | |
3204 | memory_present(early_node_map[i].nid, | |
3205 | early_node_map[i].start_pfn, | |
3206 | early_node_map[i].end_pfn); | |
3207 | } | |
3208 | ||
3209 | /** | |
3210 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3211 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3212 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3213 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3214 | * |
3215 | * It returns the start and end page frame of a node based on information | |
3216 | * provided by an arch calling add_active_range(). If called for a node | |
3217 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3218 | * PFNs will be 0. |
c713216d | 3219 | */ |
a3142c8e | 3220 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3221 | unsigned long *start_pfn, unsigned long *end_pfn) |
3222 | { | |
3223 | int i; | |
3224 | *start_pfn = -1UL; | |
3225 | *end_pfn = 0; | |
3226 | ||
3227 | for_each_active_range_index_in_nid(i, nid) { | |
3228 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3229 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3230 | } | |
3231 | ||
633c0666 | 3232 | if (*start_pfn == -1UL) |
c713216d | 3233 | *start_pfn = 0; |
c713216d MG |
3234 | } |
3235 | ||
2a1e274a MG |
3236 | /* |
3237 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3238 | * assumption is made that zones within a node are ordered in monotonic | |
3239 | * increasing memory addresses so that the "highest" populated zone is used | |
3240 | */ | |
b69a7288 | 3241 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3242 | { |
3243 | int zone_index; | |
3244 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3245 | if (zone_index == ZONE_MOVABLE) | |
3246 | continue; | |
3247 | ||
3248 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3249 | arch_zone_lowest_possible_pfn[zone_index]) | |
3250 | break; | |
3251 | } | |
3252 | ||
3253 | VM_BUG_ON(zone_index == -1); | |
3254 | movable_zone = zone_index; | |
3255 | } | |
3256 | ||
3257 | /* | |
3258 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3259 | * because it is sized independant of architecture. Unlike the other zones, | |
3260 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3261 | * in each node depending on the size of each node and how evenly kernelcore | |
3262 | * is distributed. This helper function adjusts the zone ranges | |
3263 | * provided by the architecture for a given node by using the end of the | |
3264 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3265 | * zones within a node are in order of monotonic increases memory addresses | |
3266 | */ | |
b69a7288 | 3267 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
3268 | unsigned long zone_type, |
3269 | unsigned long node_start_pfn, | |
3270 | unsigned long node_end_pfn, | |
3271 | unsigned long *zone_start_pfn, | |
3272 | unsigned long *zone_end_pfn) | |
3273 | { | |
3274 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3275 | if (zone_movable_pfn[nid]) { | |
3276 | /* Size ZONE_MOVABLE */ | |
3277 | if (zone_type == ZONE_MOVABLE) { | |
3278 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3279 | *zone_end_pfn = min(node_end_pfn, | |
3280 | arch_zone_highest_possible_pfn[movable_zone]); | |
3281 | ||
3282 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3283 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3284 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3285 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3286 | ||
3287 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3288 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3289 | *zone_start_pfn = *zone_end_pfn; | |
3290 | } | |
3291 | } | |
3292 | ||
c713216d MG |
3293 | /* |
3294 | * Return the number of pages a zone spans in a node, including holes | |
3295 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3296 | */ | |
6ea6e688 | 3297 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3298 | unsigned long zone_type, |
3299 | unsigned long *ignored) | |
3300 | { | |
3301 | unsigned long node_start_pfn, node_end_pfn; | |
3302 | unsigned long zone_start_pfn, zone_end_pfn; | |
3303 | ||
3304 | /* Get the start and end of the node and zone */ | |
3305 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3306 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3307 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3308 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3309 | node_start_pfn, node_end_pfn, | |
3310 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3311 | |
3312 | /* Check that this node has pages within the zone's required range */ | |
3313 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3314 | return 0; | |
3315 | ||
3316 | /* Move the zone boundaries inside the node if necessary */ | |
3317 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3318 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3319 | ||
3320 | /* Return the spanned pages */ | |
3321 | return zone_end_pfn - zone_start_pfn; | |
3322 | } | |
3323 | ||
3324 | /* | |
3325 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3326 | * then all holes in the requested range will be accounted for. |
c713216d | 3327 | */ |
b69a7288 | 3328 | static unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3329 | unsigned long range_start_pfn, |
3330 | unsigned long range_end_pfn) | |
3331 | { | |
3332 | int i = 0; | |
3333 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3334 | unsigned long start_pfn; | |
3335 | ||
3336 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3337 | i = first_active_region_index_in_nid(nid); | |
3338 | if (i == -1) | |
3339 | return 0; | |
3340 | ||
b5445f95 MG |
3341 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3342 | ||
9c7cd687 MG |
3343 | /* Account for ranges before physical memory on this node */ |
3344 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3345 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3346 | |
3347 | /* Find all holes for the zone within the node */ | |
3348 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3349 | ||
3350 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3351 | if (prev_end_pfn >= range_end_pfn) | |
3352 | break; | |
3353 | ||
3354 | /* Make sure the end of the zone is not within the hole */ | |
3355 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3356 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3357 | ||
3358 | /* Update the hole size cound and move on */ | |
3359 | if (start_pfn > range_start_pfn) { | |
3360 | BUG_ON(prev_end_pfn > start_pfn); | |
3361 | hole_pages += start_pfn - prev_end_pfn; | |
3362 | } | |
3363 | prev_end_pfn = early_node_map[i].end_pfn; | |
3364 | } | |
3365 | ||
9c7cd687 MG |
3366 | /* Account for ranges past physical memory on this node */ |
3367 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3368 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3369 | max(range_start_pfn, prev_end_pfn); |
3370 | ||
c713216d MG |
3371 | return hole_pages; |
3372 | } | |
3373 | ||
3374 | /** | |
3375 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3376 | * @start_pfn: The start PFN to start searching for holes | |
3377 | * @end_pfn: The end PFN to stop searching for holes | |
3378 | * | |
88ca3b94 | 3379 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3380 | */ |
3381 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3382 | unsigned long end_pfn) | |
3383 | { | |
3384 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3385 | } | |
3386 | ||
3387 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3388 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3389 | unsigned long zone_type, |
3390 | unsigned long *ignored) | |
3391 | { | |
9c7cd687 MG |
3392 | unsigned long node_start_pfn, node_end_pfn; |
3393 | unsigned long zone_start_pfn, zone_end_pfn; | |
3394 | ||
3395 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3396 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3397 | node_start_pfn); | |
3398 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3399 | node_end_pfn); | |
3400 | ||
2a1e274a MG |
3401 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3402 | node_start_pfn, node_end_pfn, | |
3403 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3404 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3405 | } |
0e0b864e | 3406 | |
c713216d | 3407 | #else |
6ea6e688 | 3408 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3409 | unsigned long zone_type, |
3410 | unsigned long *zones_size) | |
3411 | { | |
3412 | return zones_size[zone_type]; | |
3413 | } | |
3414 | ||
6ea6e688 | 3415 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3416 | unsigned long zone_type, |
3417 | unsigned long *zholes_size) | |
3418 | { | |
3419 | if (!zholes_size) | |
3420 | return 0; | |
3421 | ||
3422 | return zholes_size[zone_type]; | |
3423 | } | |
0e0b864e | 3424 | |
c713216d MG |
3425 | #endif |
3426 | ||
a3142c8e | 3427 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3428 | unsigned long *zones_size, unsigned long *zholes_size) |
3429 | { | |
3430 | unsigned long realtotalpages, totalpages = 0; | |
3431 | enum zone_type i; | |
3432 | ||
3433 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3434 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3435 | zones_size); | |
3436 | pgdat->node_spanned_pages = totalpages; | |
3437 | ||
3438 | realtotalpages = totalpages; | |
3439 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3440 | realtotalpages -= | |
3441 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3442 | zholes_size); | |
3443 | pgdat->node_present_pages = realtotalpages; | |
3444 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3445 | realtotalpages); | |
3446 | } | |
3447 | ||
835c134e MG |
3448 | #ifndef CONFIG_SPARSEMEM |
3449 | /* | |
3450 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3451 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3452 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3453 | * round what is now in bits to nearest long in bits, then return it in |
3454 | * bytes. | |
3455 | */ | |
3456 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3457 | { | |
3458 | unsigned long usemapsize; | |
3459 | ||
d9c23400 MG |
3460 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3461 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3462 | usemapsize *= NR_PAGEBLOCK_BITS; |
3463 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3464 | ||
3465 | return usemapsize / 8; | |
3466 | } | |
3467 | ||
3468 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3469 | struct zone *zone, unsigned long zonesize) | |
3470 | { | |
3471 | unsigned long usemapsize = usemap_size(zonesize); | |
3472 | zone->pageblock_flags = NULL; | |
58a01a45 | 3473 | if (usemapsize) |
835c134e | 3474 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
835c134e MG |
3475 | } |
3476 | #else | |
3477 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3478 | struct zone *zone, unsigned long zonesize) {} | |
3479 | #endif /* CONFIG_SPARSEMEM */ | |
3480 | ||
d9c23400 | 3481 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3482 | |
3483 | /* Return a sensible default order for the pageblock size. */ | |
3484 | static inline int pageblock_default_order(void) | |
3485 | { | |
3486 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3487 | return HUGETLB_PAGE_ORDER; | |
3488 | ||
3489 | return MAX_ORDER-1; | |
3490 | } | |
3491 | ||
d9c23400 MG |
3492 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3493 | static inline void __init set_pageblock_order(unsigned int order) | |
3494 | { | |
3495 | /* Check that pageblock_nr_pages has not already been setup */ | |
3496 | if (pageblock_order) | |
3497 | return; | |
3498 | ||
3499 | /* | |
3500 | * Assume the largest contiguous order of interest is a huge page. | |
3501 | * This value may be variable depending on boot parameters on IA64 | |
3502 | */ | |
3503 | pageblock_order = order; | |
3504 | } | |
3505 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3506 | ||
ba72cb8c MG |
3507 | /* |
3508 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3509 | * and pageblock_default_order() are unused as pageblock_order is set | |
3510 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3511 | * pageblock_order based on the kernel config | |
3512 | */ | |
3513 | static inline int pageblock_default_order(unsigned int order) | |
3514 | { | |
3515 | return MAX_ORDER-1; | |
3516 | } | |
d9c23400 MG |
3517 | #define set_pageblock_order(x) do {} while (0) |
3518 | ||
3519 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3520 | ||
1da177e4 LT |
3521 | /* |
3522 | * Set up the zone data structures: | |
3523 | * - mark all pages reserved | |
3524 | * - mark all memory queues empty | |
3525 | * - clear the memory bitmaps | |
3526 | */ | |
b5a0e011 | 3527 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3528 | unsigned long *zones_size, unsigned long *zholes_size) |
3529 | { | |
2f1b6248 | 3530 | enum zone_type j; |
ed8ece2e | 3531 | int nid = pgdat->node_id; |
1da177e4 | 3532 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3533 | int ret; |
1da177e4 | 3534 | |
208d54e5 | 3535 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3536 | pgdat->nr_zones = 0; |
3537 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3538 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 3539 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
3540 | |
3541 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3542 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3543 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 3544 | enum lru_list l; |
1da177e4 | 3545 | |
c713216d MG |
3546 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3547 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3548 | zholes_size); | |
1da177e4 | 3549 | |
0e0b864e MG |
3550 | /* |
3551 | * Adjust realsize so that it accounts for how much memory | |
3552 | * is used by this zone for memmap. This affects the watermark | |
3553 | * and per-cpu initialisations | |
3554 | */ | |
f7232154 JW |
3555 | memmap_pages = |
3556 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
3557 | if (realsize >= memmap_pages) { |
3558 | realsize -= memmap_pages; | |
5594c8c8 YL |
3559 | if (memmap_pages) |
3560 | printk(KERN_DEBUG | |
3561 | " %s zone: %lu pages used for memmap\n", | |
3562 | zone_names[j], memmap_pages); | |
0e0b864e MG |
3563 | } else |
3564 | printk(KERN_WARNING | |
3565 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3566 | zone_names[j], memmap_pages, realsize); | |
3567 | ||
6267276f CL |
3568 | /* Account for reserved pages */ |
3569 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3570 | realsize -= dma_reserve; |
d903ef9f | 3571 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 3572 | zone_names[0], dma_reserve); |
0e0b864e MG |
3573 | } |
3574 | ||
98d2b0eb | 3575 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3576 | nr_kernel_pages += realsize; |
3577 | nr_all_pages += realsize; | |
3578 | ||
3579 | zone->spanned_pages = size; | |
3580 | zone->present_pages = realsize; | |
9614634f | 3581 | #ifdef CONFIG_NUMA |
d5f541ed | 3582 | zone->node = nid; |
8417bba4 | 3583 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3584 | / 100; |
0ff38490 | 3585 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3586 | #endif |
1da177e4 LT |
3587 | zone->name = zone_names[j]; |
3588 | spin_lock_init(&zone->lock); | |
3589 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3590 | zone_seqlock_init(zone); |
1da177e4 | 3591 | zone->zone_pgdat = pgdat; |
1da177e4 | 3592 | |
3bb1a852 | 3593 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3594 | |
ed8ece2e | 3595 | zone_pcp_init(zone); |
b69408e8 CL |
3596 | for_each_lru(l) { |
3597 | INIT_LIST_HEAD(&zone->lru[l].list); | |
3598 | zone->lru[l].nr_scan = 0; | |
3599 | } | |
6e901571 KM |
3600 | zone->reclaim_stat.recent_rotated[0] = 0; |
3601 | zone->reclaim_stat.recent_rotated[1] = 0; | |
3602 | zone->reclaim_stat.recent_scanned[0] = 0; | |
3603 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 3604 | zap_zone_vm_stats(zone); |
e815af95 | 3605 | zone->flags = 0; |
1da177e4 LT |
3606 | if (!size) |
3607 | continue; | |
3608 | ||
ba72cb8c | 3609 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3610 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3611 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3612 | size, MEMMAP_EARLY); | |
718127cc | 3613 | BUG_ON(ret); |
76cdd58e | 3614 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 3615 | zone_start_pfn += size; |
1da177e4 LT |
3616 | } |
3617 | } | |
3618 | ||
577a32f6 | 3619 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3620 | { |
1da177e4 LT |
3621 | /* Skip empty nodes */ |
3622 | if (!pgdat->node_spanned_pages) | |
3623 | return; | |
3624 | ||
d41dee36 | 3625 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3626 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3627 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3628 | unsigned long size, start, end; |
d41dee36 AW |
3629 | struct page *map; |
3630 | ||
e984bb43 BP |
3631 | /* |
3632 | * The zone's endpoints aren't required to be MAX_ORDER | |
3633 | * aligned but the node_mem_map endpoints must be in order | |
3634 | * for the buddy allocator to function correctly. | |
3635 | */ | |
3636 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3637 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3638 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3639 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3640 | map = alloc_remap(pgdat->node_id, size); |
3641 | if (!map) | |
3642 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3643 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3644 | } |
12d810c1 | 3645 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3646 | /* |
3647 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3648 | */ | |
c713216d | 3649 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3650 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3651 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3652 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3653 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3654 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3655 | } | |
1da177e4 | 3656 | #endif |
d41dee36 | 3657 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3658 | } |
3659 | ||
9109fb7b JW |
3660 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
3661 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 3662 | { |
9109fb7b JW |
3663 | pg_data_t *pgdat = NODE_DATA(nid); |
3664 | ||
1da177e4 LT |
3665 | pgdat->node_id = nid; |
3666 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3667 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3668 | |
3669 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
3670 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
3671 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3672 | nid, (unsigned long)pgdat, | |
3673 | (unsigned long)pgdat->node_mem_map); | |
3674 | #endif | |
1da177e4 LT |
3675 | |
3676 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3677 | } | |
3678 | ||
c713216d | 3679 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3680 | |
3681 | #if MAX_NUMNODES > 1 | |
3682 | /* | |
3683 | * Figure out the number of possible node ids. | |
3684 | */ | |
3685 | static void __init setup_nr_node_ids(void) | |
3686 | { | |
3687 | unsigned int node; | |
3688 | unsigned int highest = 0; | |
3689 | ||
3690 | for_each_node_mask(node, node_possible_map) | |
3691 | highest = node; | |
3692 | nr_node_ids = highest + 1; | |
3693 | } | |
3694 | #else | |
3695 | static inline void setup_nr_node_ids(void) | |
3696 | { | |
3697 | } | |
3698 | #endif | |
3699 | ||
c713216d MG |
3700 | /** |
3701 | * add_active_range - Register a range of PFNs backed by physical memory | |
3702 | * @nid: The node ID the range resides on | |
3703 | * @start_pfn: The start PFN of the available physical memory | |
3704 | * @end_pfn: The end PFN of the available physical memory | |
3705 | * | |
3706 | * These ranges are stored in an early_node_map[] and later used by | |
3707 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3708 | * range spans a memory hole, it is up to the architecture to ensure | |
3709 | * the memory is not freed by the bootmem allocator. If possible | |
3710 | * the range being registered will be merged with existing ranges. | |
3711 | */ | |
3712 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3713 | unsigned long end_pfn) | |
3714 | { | |
3715 | int i; | |
3716 | ||
6b74ab97 MG |
3717 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
3718 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3719 | "%d entries of %d used\n", | |
3720 | nid, start_pfn, end_pfn, | |
3721 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 3722 | |
2dbb51c4 MG |
3723 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
3724 | ||
c713216d MG |
3725 | /* Merge with existing active regions if possible */ |
3726 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3727 | if (early_node_map[i].nid != nid) | |
3728 | continue; | |
3729 | ||
3730 | /* Skip if an existing region covers this new one */ | |
3731 | if (start_pfn >= early_node_map[i].start_pfn && | |
3732 | end_pfn <= early_node_map[i].end_pfn) | |
3733 | return; | |
3734 | ||
3735 | /* Merge forward if suitable */ | |
3736 | if (start_pfn <= early_node_map[i].end_pfn && | |
3737 | end_pfn > early_node_map[i].end_pfn) { | |
3738 | early_node_map[i].end_pfn = end_pfn; | |
3739 | return; | |
3740 | } | |
3741 | ||
3742 | /* Merge backward if suitable */ | |
3743 | if (start_pfn < early_node_map[i].end_pfn && | |
3744 | end_pfn >= early_node_map[i].start_pfn) { | |
3745 | early_node_map[i].start_pfn = start_pfn; | |
3746 | return; | |
3747 | } | |
3748 | } | |
3749 | ||
3750 | /* Check that early_node_map is large enough */ | |
3751 | if (i >= MAX_ACTIVE_REGIONS) { | |
3752 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3753 | MAX_ACTIVE_REGIONS); | |
3754 | return; | |
3755 | } | |
3756 | ||
3757 | early_node_map[i].nid = nid; | |
3758 | early_node_map[i].start_pfn = start_pfn; | |
3759 | early_node_map[i].end_pfn = end_pfn; | |
3760 | nr_nodemap_entries = i + 1; | |
3761 | } | |
3762 | ||
3763 | /** | |
cc1050ba | 3764 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 3765 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
3766 | * @start_pfn: The new PFN of the range |
3767 | * @end_pfn: The new PFN of the range | |
c713216d MG |
3768 | * |
3769 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
3770 | * The map is kept near the end physical page range that has already been |
3771 | * registered. This function allows an arch to shrink an existing registered | |
3772 | * range. | |
c713216d | 3773 | */ |
cc1050ba YL |
3774 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
3775 | unsigned long end_pfn) | |
c713216d | 3776 | { |
cc1a9d86 YL |
3777 | int i, j; |
3778 | int removed = 0; | |
c713216d | 3779 | |
cc1050ba YL |
3780 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
3781 | nid, start_pfn, end_pfn); | |
3782 | ||
c713216d | 3783 | /* Find the old active region end and shrink */ |
cc1a9d86 | 3784 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
3785 | if (early_node_map[i].start_pfn >= start_pfn && |
3786 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 3787 | /* clear it */ |
cc1050ba | 3788 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
3789 | early_node_map[i].end_pfn = 0; |
3790 | removed = 1; | |
3791 | continue; | |
3792 | } | |
cc1050ba YL |
3793 | if (early_node_map[i].start_pfn < start_pfn && |
3794 | early_node_map[i].end_pfn > start_pfn) { | |
3795 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
3796 | early_node_map[i].end_pfn = start_pfn; | |
3797 | if (temp_end_pfn > end_pfn) | |
3798 | add_active_range(nid, end_pfn, temp_end_pfn); | |
3799 | continue; | |
3800 | } | |
3801 | if (early_node_map[i].start_pfn >= start_pfn && | |
3802 | early_node_map[i].end_pfn > end_pfn && | |
3803 | early_node_map[i].start_pfn < end_pfn) { | |
3804 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 3805 | continue; |
c713216d | 3806 | } |
cc1a9d86 YL |
3807 | } |
3808 | ||
3809 | if (!removed) | |
3810 | return; | |
3811 | ||
3812 | /* remove the blank ones */ | |
3813 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
3814 | if (early_node_map[i].nid != nid) | |
3815 | continue; | |
3816 | if (early_node_map[i].end_pfn) | |
3817 | continue; | |
3818 | /* we found it, get rid of it */ | |
3819 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
3820 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
3821 | sizeof(early_node_map[j])); | |
3822 | j = nr_nodemap_entries - 1; | |
3823 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
3824 | nr_nodemap_entries--; | |
3825 | } | |
c713216d MG |
3826 | } |
3827 | ||
3828 | /** | |
3829 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3830 | * |
c713216d MG |
3831 | * During discovery, it may be found that a table like SRAT is invalid |
3832 | * and an alternative discovery method must be used. This function removes | |
3833 | * all currently registered regions. | |
3834 | */ | |
88ca3b94 | 3835 | void __init remove_all_active_ranges(void) |
c713216d MG |
3836 | { |
3837 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3838 | nr_nodemap_entries = 0; | |
3839 | } | |
3840 | ||
3841 | /* Compare two active node_active_regions */ | |
3842 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3843 | { | |
3844 | struct node_active_region *arange = (struct node_active_region *)a; | |
3845 | struct node_active_region *brange = (struct node_active_region *)b; | |
3846 | ||
3847 | /* Done this way to avoid overflows */ | |
3848 | if (arange->start_pfn > brange->start_pfn) | |
3849 | return 1; | |
3850 | if (arange->start_pfn < brange->start_pfn) | |
3851 | return -1; | |
3852 | ||
3853 | return 0; | |
3854 | } | |
3855 | ||
3856 | /* sort the node_map by start_pfn */ | |
3857 | static void __init sort_node_map(void) | |
3858 | { | |
3859 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3860 | sizeof(struct node_active_region), | |
3861 | cmp_node_active_region, NULL); | |
3862 | } | |
3863 | ||
a6af2bc3 | 3864 | /* Find the lowest pfn for a node */ |
b69a7288 | 3865 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
3866 | { |
3867 | int i; | |
a6af2bc3 | 3868 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3869 | |
c713216d MG |
3870 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3871 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3872 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3873 | |
a6af2bc3 MG |
3874 | if (min_pfn == ULONG_MAX) { |
3875 | printk(KERN_WARNING | |
2bc0d261 | 3876 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
3877 | return 0; |
3878 | } | |
3879 | ||
3880 | return min_pfn; | |
c713216d MG |
3881 | } |
3882 | ||
3883 | /** | |
3884 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3885 | * | |
3886 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3887 | * add_active_range(). |
c713216d MG |
3888 | */ |
3889 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3890 | { | |
3891 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3892 | } | |
3893 | ||
37b07e41 LS |
3894 | /* |
3895 | * early_calculate_totalpages() | |
3896 | * Sum pages in active regions for movable zone. | |
3897 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3898 | */ | |
484f51f8 | 3899 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
3900 | { |
3901 | int i; | |
3902 | unsigned long totalpages = 0; | |
3903 | ||
37b07e41 LS |
3904 | for (i = 0; i < nr_nodemap_entries; i++) { |
3905 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3906 | early_node_map[i].start_pfn; |
37b07e41 LS |
3907 | totalpages += pages; |
3908 | if (pages) | |
3909 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3910 | } | |
3911 | return totalpages; | |
7e63efef MG |
3912 | } |
3913 | ||
2a1e274a MG |
3914 | /* |
3915 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3916 | * is spread evenly between nodes as long as the nodes have enough | |
3917 | * memory. When they don't, some nodes will have more kernelcore than | |
3918 | * others | |
3919 | */ | |
b69a7288 | 3920 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
3921 | { |
3922 | int i, nid; | |
3923 | unsigned long usable_startpfn; | |
3924 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3925 | unsigned long totalpages = early_calculate_totalpages(); |
3926 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3927 | |
7e63efef MG |
3928 | /* |
3929 | * If movablecore was specified, calculate what size of | |
3930 | * kernelcore that corresponds so that memory usable for | |
3931 | * any allocation type is evenly spread. If both kernelcore | |
3932 | * and movablecore are specified, then the value of kernelcore | |
3933 | * will be used for required_kernelcore if it's greater than | |
3934 | * what movablecore would have allowed. | |
3935 | */ | |
3936 | if (required_movablecore) { | |
7e63efef MG |
3937 | unsigned long corepages; |
3938 | ||
3939 | /* | |
3940 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3941 | * was requested by the user | |
3942 | */ | |
3943 | required_movablecore = | |
3944 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3945 | corepages = totalpages - required_movablecore; | |
3946 | ||
3947 | required_kernelcore = max(required_kernelcore, corepages); | |
3948 | } | |
3949 | ||
2a1e274a MG |
3950 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3951 | if (!required_kernelcore) | |
3952 | return; | |
3953 | ||
3954 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3955 | find_usable_zone_for_movable(); | |
3956 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3957 | ||
3958 | restart: | |
3959 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3960 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3961 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3962 | /* |
3963 | * Recalculate kernelcore_node if the division per node | |
3964 | * now exceeds what is necessary to satisfy the requested | |
3965 | * amount of memory for the kernel | |
3966 | */ | |
3967 | if (required_kernelcore < kernelcore_node) | |
3968 | kernelcore_node = required_kernelcore / usable_nodes; | |
3969 | ||
3970 | /* | |
3971 | * As the map is walked, we track how much memory is usable | |
3972 | * by the kernel using kernelcore_remaining. When it is | |
3973 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3974 | */ | |
3975 | kernelcore_remaining = kernelcore_node; | |
3976 | ||
3977 | /* Go through each range of PFNs within this node */ | |
3978 | for_each_active_range_index_in_nid(i, nid) { | |
3979 | unsigned long start_pfn, end_pfn; | |
3980 | unsigned long size_pages; | |
3981 | ||
3982 | start_pfn = max(early_node_map[i].start_pfn, | |
3983 | zone_movable_pfn[nid]); | |
3984 | end_pfn = early_node_map[i].end_pfn; | |
3985 | if (start_pfn >= end_pfn) | |
3986 | continue; | |
3987 | ||
3988 | /* Account for what is only usable for kernelcore */ | |
3989 | if (start_pfn < usable_startpfn) { | |
3990 | unsigned long kernel_pages; | |
3991 | kernel_pages = min(end_pfn, usable_startpfn) | |
3992 | - start_pfn; | |
3993 | ||
3994 | kernelcore_remaining -= min(kernel_pages, | |
3995 | kernelcore_remaining); | |
3996 | required_kernelcore -= min(kernel_pages, | |
3997 | required_kernelcore); | |
3998 | ||
3999 | /* Continue if range is now fully accounted */ | |
4000 | if (end_pfn <= usable_startpfn) { | |
4001 | ||
4002 | /* | |
4003 | * Push zone_movable_pfn to the end so | |
4004 | * that if we have to rebalance | |
4005 | * kernelcore across nodes, we will | |
4006 | * not double account here | |
4007 | */ | |
4008 | zone_movable_pfn[nid] = end_pfn; | |
4009 | continue; | |
4010 | } | |
4011 | start_pfn = usable_startpfn; | |
4012 | } | |
4013 | ||
4014 | /* | |
4015 | * The usable PFN range for ZONE_MOVABLE is from | |
4016 | * start_pfn->end_pfn. Calculate size_pages as the | |
4017 | * number of pages used as kernelcore | |
4018 | */ | |
4019 | size_pages = end_pfn - start_pfn; | |
4020 | if (size_pages > kernelcore_remaining) | |
4021 | size_pages = kernelcore_remaining; | |
4022 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4023 | ||
4024 | /* | |
4025 | * Some kernelcore has been met, update counts and | |
4026 | * break if the kernelcore for this node has been | |
4027 | * satisified | |
4028 | */ | |
4029 | required_kernelcore -= min(required_kernelcore, | |
4030 | size_pages); | |
4031 | kernelcore_remaining -= size_pages; | |
4032 | if (!kernelcore_remaining) | |
4033 | break; | |
4034 | } | |
4035 | } | |
4036 | ||
4037 | /* | |
4038 | * If there is still required_kernelcore, we do another pass with one | |
4039 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4040 | * along on the nodes that still have memory until kernelcore is | |
4041 | * satisified | |
4042 | */ | |
4043 | usable_nodes--; | |
4044 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4045 | goto restart; | |
4046 | ||
4047 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4048 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4049 | zone_movable_pfn[nid] = | |
4050 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
4051 | } | |
4052 | ||
37b07e41 LS |
4053 | /* Any regular memory on that node ? */ |
4054 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4055 | { | |
4056 | #ifdef CONFIG_HIGHMEM | |
4057 | enum zone_type zone_type; | |
4058 | ||
4059 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4060 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4061 | if (zone->present_pages) | |
4062 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4063 | } | |
4064 | #endif | |
4065 | } | |
4066 | ||
c713216d MG |
4067 | /** |
4068 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4069 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4070 | * |
4071 | * This will call free_area_init_node() for each active node in the system. | |
4072 | * Using the page ranges provided by add_active_range(), the size of each | |
4073 | * zone in each node and their holes is calculated. If the maximum PFN | |
4074 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4075 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4076 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4077 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4078 | * at arch_max_dma_pfn. | |
4079 | */ | |
4080 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4081 | { | |
4082 | unsigned long nid; | |
db99100d | 4083 | int i; |
c713216d | 4084 | |
a6af2bc3 MG |
4085 | /* Sort early_node_map as initialisation assumes it is sorted */ |
4086 | sort_node_map(); | |
4087 | ||
c713216d MG |
4088 | /* Record where the zone boundaries are */ |
4089 | memset(arch_zone_lowest_possible_pfn, 0, | |
4090 | sizeof(arch_zone_lowest_possible_pfn)); | |
4091 | memset(arch_zone_highest_possible_pfn, 0, | |
4092 | sizeof(arch_zone_highest_possible_pfn)); | |
4093 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4094 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4095 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4096 | if (i == ZONE_MOVABLE) |
4097 | continue; | |
c713216d MG |
4098 | arch_zone_lowest_possible_pfn[i] = |
4099 | arch_zone_highest_possible_pfn[i-1]; | |
4100 | arch_zone_highest_possible_pfn[i] = | |
4101 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4102 | } | |
2a1e274a MG |
4103 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4104 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4105 | ||
4106 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4107 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4108 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4109 | |
c713216d MG |
4110 | /* Print out the zone ranges */ |
4111 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4112 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4113 | if (i == ZONE_MOVABLE) | |
4114 | continue; | |
5dab8ec1 | 4115 | printk(" %-8s %0#10lx -> %0#10lx\n", |
c713216d MG |
4116 | zone_names[i], |
4117 | arch_zone_lowest_possible_pfn[i], | |
4118 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4119 | } |
4120 | ||
4121 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4122 | printk("Movable zone start PFN for each node\n"); | |
4123 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4124 | if (zone_movable_pfn[i]) | |
4125 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4126 | } | |
c713216d MG |
4127 | |
4128 | /* Print out the early_node_map[] */ | |
4129 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4130 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 4131 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
4132 | early_node_map[i].start_pfn, |
4133 | early_node_map[i].end_pfn); | |
4134 | ||
4135 | /* Initialise every node */ | |
708614e6 | 4136 | mminit_verify_pageflags_layout(); |
8ef82866 | 4137 | setup_nr_node_ids(); |
c713216d MG |
4138 | for_each_online_node(nid) { |
4139 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4140 | free_area_init_node(nid, NULL, |
c713216d | 4141 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4142 | |
4143 | /* Any memory on that node */ | |
4144 | if (pgdat->node_present_pages) | |
4145 | node_set_state(nid, N_HIGH_MEMORY); | |
4146 | check_for_regular_memory(pgdat); | |
c713216d MG |
4147 | } |
4148 | } | |
2a1e274a | 4149 | |
7e63efef | 4150 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4151 | { |
4152 | unsigned long long coremem; | |
4153 | if (!p) | |
4154 | return -EINVAL; | |
4155 | ||
4156 | coremem = memparse(p, &p); | |
7e63efef | 4157 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4158 | |
7e63efef | 4159 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4160 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4161 | ||
4162 | return 0; | |
4163 | } | |
ed7ed365 | 4164 | |
7e63efef MG |
4165 | /* |
4166 | * kernelcore=size sets the amount of memory for use for allocations that | |
4167 | * cannot be reclaimed or migrated. | |
4168 | */ | |
4169 | static int __init cmdline_parse_kernelcore(char *p) | |
4170 | { | |
4171 | return cmdline_parse_core(p, &required_kernelcore); | |
4172 | } | |
4173 | ||
4174 | /* | |
4175 | * movablecore=size sets the amount of memory for use for allocations that | |
4176 | * can be reclaimed or migrated. | |
4177 | */ | |
4178 | static int __init cmdline_parse_movablecore(char *p) | |
4179 | { | |
4180 | return cmdline_parse_core(p, &required_movablecore); | |
4181 | } | |
4182 | ||
ed7ed365 | 4183 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4184 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4185 | |
c713216d MG |
4186 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4187 | ||
0e0b864e | 4188 | /** |
88ca3b94 RD |
4189 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4190 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4191 | * |
4192 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4193 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4194 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4195 | * function may optionally be used to account for unfreeable pages in the |
4196 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4197 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4198 | */ |
4199 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4200 | { | |
4201 | dma_reserve = new_dma_reserve; | |
4202 | } | |
4203 | ||
93b7504e | 4204 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
52765583 | 4205 | struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] }; |
1da177e4 | 4206 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4207 | #endif |
1da177e4 LT |
4208 | |
4209 | void __init free_area_init(unsigned long *zones_size) | |
4210 | { | |
9109fb7b | 4211 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4212 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4213 | } | |
1da177e4 | 4214 | |
1da177e4 LT |
4215 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4216 | unsigned long action, void *hcpu) | |
4217 | { | |
4218 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4219 | |
8bb78442 | 4220 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4221 | drain_pages(cpu); |
4222 | ||
4223 | /* | |
4224 | * Spill the event counters of the dead processor | |
4225 | * into the current processors event counters. | |
4226 | * This artificially elevates the count of the current | |
4227 | * processor. | |
4228 | */ | |
f8891e5e | 4229 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4230 | |
4231 | /* | |
4232 | * Zero the differential counters of the dead processor | |
4233 | * so that the vm statistics are consistent. | |
4234 | * | |
4235 | * This is only okay since the processor is dead and cannot | |
4236 | * race with what we are doing. | |
4237 | */ | |
2244b95a | 4238 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4239 | } |
4240 | return NOTIFY_OK; | |
4241 | } | |
1da177e4 LT |
4242 | |
4243 | void __init page_alloc_init(void) | |
4244 | { | |
4245 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4246 | } | |
4247 | ||
cb45b0e9 HA |
4248 | /* |
4249 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4250 | * or min_free_kbytes changes. | |
4251 | */ | |
4252 | static void calculate_totalreserve_pages(void) | |
4253 | { | |
4254 | struct pglist_data *pgdat; | |
4255 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4256 | enum zone_type i, j; |
cb45b0e9 HA |
4257 | |
4258 | for_each_online_pgdat(pgdat) { | |
4259 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4260 | struct zone *zone = pgdat->node_zones + i; | |
4261 | unsigned long max = 0; | |
4262 | ||
4263 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4264 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4265 | if (zone->lowmem_reserve[j] > max) | |
4266 | max = zone->lowmem_reserve[j]; | |
4267 | } | |
4268 | ||
4269 | /* we treat pages_high as reserved pages. */ | |
4270 | max += zone->pages_high; | |
4271 | ||
4272 | if (max > zone->present_pages) | |
4273 | max = zone->present_pages; | |
4274 | reserve_pages += max; | |
4275 | } | |
4276 | } | |
4277 | totalreserve_pages = reserve_pages; | |
4278 | } | |
4279 | ||
1da177e4 LT |
4280 | /* |
4281 | * setup_per_zone_lowmem_reserve - called whenever | |
4282 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4283 | * has a correct pages reserved value, so an adequate number of | |
4284 | * pages are left in the zone after a successful __alloc_pages(). | |
4285 | */ | |
4286 | static void setup_per_zone_lowmem_reserve(void) | |
4287 | { | |
4288 | struct pglist_data *pgdat; | |
2f6726e5 | 4289 | enum zone_type j, idx; |
1da177e4 | 4290 | |
ec936fc5 | 4291 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4292 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4293 | struct zone *zone = pgdat->node_zones + j; | |
4294 | unsigned long present_pages = zone->present_pages; | |
4295 | ||
4296 | zone->lowmem_reserve[j] = 0; | |
4297 | ||
2f6726e5 CL |
4298 | idx = j; |
4299 | while (idx) { | |
1da177e4 LT |
4300 | struct zone *lower_zone; |
4301 | ||
2f6726e5 CL |
4302 | idx--; |
4303 | ||
1da177e4 LT |
4304 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4305 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4306 | ||
4307 | lower_zone = pgdat->node_zones + idx; | |
4308 | lower_zone->lowmem_reserve[j] = present_pages / | |
4309 | sysctl_lowmem_reserve_ratio[idx]; | |
4310 | present_pages += lower_zone->present_pages; | |
4311 | } | |
4312 | } | |
4313 | } | |
cb45b0e9 HA |
4314 | |
4315 | /* update totalreserve_pages */ | |
4316 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4317 | } |
4318 | ||
88ca3b94 RD |
4319 | /** |
4320 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
4321 | * | |
4322 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
4323 | * with respect to min_free_kbytes. | |
1da177e4 | 4324 | */ |
3947be19 | 4325 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
4326 | { |
4327 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4328 | unsigned long lowmem_pages = 0; | |
4329 | struct zone *zone; | |
4330 | unsigned long flags; | |
4331 | ||
4332 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4333 | for_each_zone(zone) { | |
4334 | if (!is_highmem(zone)) | |
4335 | lowmem_pages += zone->present_pages; | |
4336 | } | |
4337 | ||
4338 | for_each_zone(zone) { | |
ac924c60 AM |
4339 | u64 tmp; |
4340 | ||
1125b4e3 | 4341 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4342 | tmp = (u64)pages_min * zone->present_pages; |
4343 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4344 | if (is_highmem(zone)) { |
4345 | /* | |
669ed175 NP |
4346 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4347 | * need highmem pages, so cap pages_min to a small | |
4348 | * value here. | |
4349 | * | |
4350 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
4351 | * deltas controls asynch page reclaim, and so should | |
4352 | * not be capped for highmem. | |
1da177e4 LT |
4353 | */ |
4354 | int min_pages; | |
4355 | ||
4356 | min_pages = zone->present_pages / 1024; | |
4357 | if (min_pages < SWAP_CLUSTER_MAX) | |
4358 | min_pages = SWAP_CLUSTER_MAX; | |
4359 | if (min_pages > 128) | |
4360 | min_pages = 128; | |
4361 | zone->pages_min = min_pages; | |
4362 | } else { | |
669ed175 NP |
4363 | /* |
4364 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4365 | * proportionate to the zone's size. |
4366 | */ | |
669ed175 | 4367 | zone->pages_min = tmp; |
1da177e4 LT |
4368 | } |
4369 | ||
ac924c60 AM |
4370 | zone->pages_low = zone->pages_min + (tmp >> 2); |
4371 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
56fd56b8 | 4372 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4373 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4374 | } |
cb45b0e9 HA |
4375 | |
4376 | /* update totalreserve_pages */ | |
4377 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4378 | } |
4379 | ||
556adecb RR |
4380 | /** |
4381 | * setup_per_zone_inactive_ratio - called when min_free_kbytes changes. | |
4382 | * | |
4383 | * The inactive anon list should be small enough that the VM never has to | |
4384 | * do too much work, but large enough that each inactive page has a chance | |
4385 | * to be referenced again before it is swapped out. | |
4386 | * | |
4387 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4388 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4389 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4390 | * the anonymous pages are kept on the inactive list. | |
4391 | * | |
4392 | * total target max | |
4393 | * memory ratio inactive anon | |
4394 | * ------------------------------------- | |
4395 | * 10MB 1 5MB | |
4396 | * 100MB 1 50MB | |
4397 | * 1GB 3 250MB | |
4398 | * 10GB 10 0.9GB | |
4399 | * 100GB 31 3GB | |
4400 | * 1TB 101 10GB | |
4401 | * 10TB 320 32GB | |
4402 | */ | |
efab8186 | 4403 | static void setup_per_zone_inactive_ratio(void) |
556adecb RR |
4404 | { |
4405 | struct zone *zone; | |
4406 | ||
4407 | for_each_zone(zone) { | |
4408 | unsigned int gb, ratio; | |
4409 | ||
4410 | /* Zone size in gigabytes */ | |
4411 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
4412 | ratio = int_sqrt(10 * gb); | |
4413 | if (!ratio) | |
4414 | ratio = 1; | |
4415 | ||
4416 | zone->inactive_ratio = ratio; | |
4417 | } | |
4418 | } | |
4419 | ||
1da177e4 LT |
4420 | /* |
4421 | * Initialise min_free_kbytes. | |
4422 | * | |
4423 | * For small machines we want it small (128k min). For large machines | |
4424 | * we want it large (64MB max). But it is not linear, because network | |
4425 | * bandwidth does not increase linearly with machine size. We use | |
4426 | * | |
4427 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4428 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4429 | * | |
4430 | * which yields | |
4431 | * | |
4432 | * 16MB: 512k | |
4433 | * 32MB: 724k | |
4434 | * 64MB: 1024k | |
4435 | * 128MB: 1448k | |
4436 | * 256MB: 2048k | |
4437 | * 512MB: 2896k | |
4438 | * 1024MB: 4096k | |
4439 | * 2048MB: 5792k | |
4440 | * 4096MB: 8192k | |
4441 | * 8192MB: 11584k | |
4442 | * 16384MB: 16384k | |
4443 | */ | |
4444 | static int __init init_per_zone_pages_min(void) | |
4445 | { | |
4446 | unsigned long lowmem_kbytes; | |
4447 | ||
4448 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4449 | ||
4450 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4451 | if (min_free_kbytes < 128) | |
4452 | min_free_kbytes = 128; | |
4453 | if (min_free_kbytes > 65536) | |
4454 | min_free_kbytes = 65536; | |
4455 | setup_per_zone_pages_min(); | |
4456 | setup_per_zone_lowmem_reserve(); | |
556adecb | 4457 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
4458 | return 0; |
4459 | } | |
4460 | module_init(init_per_zone_pages_min) | |
4461 | ||
4462 | /* | |
4463 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4464 | * that we can call two helper functions whenever min_free_kbytes | |
4465 | * changes. | |
4466 | */ | |
4467 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4468 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4469 | { | |
4470 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
4471 | if (write) |
4472 | setup_per_zone_pages_min(); | |
1da177e4 LT |
4473 | return 0; |
4474 | } | |
4475 | ||
9614634f CL |
4476 | #ifdef CONFIG_NUMA |
4477 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4478 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4479 | { | |
4480 | struct zone *zone; | |
4481 | int rc; | |
4482 | ||
4483 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4484 | if (rc) | |
4485 | return rc; | |
4486 | ||
4487 | for_each_zone(zone) | |
8417bba4 | 4488 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4489 | sysctl_min_unmapped_ratio) / 100; |
4490 | return 0; | |
4491 | } | |
0ff38490 CL |
4492 | |
4493 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4494 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4495 | { | |
4496 | struct zone *zone; | |
4497 | int rc; | |
4498 | ||
4499 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4500 | if (rc) | |
4501 | return rc; | |
4502 | ||
4503 | for_each_zone(zone) | |
4504 | zone->min_slab_pages = (zone->present_pages * | |
4505 | sysctl_min_slab_ratio) / 100; | |
4506 | return 0; | |
4507 | } | |
9614634f CL |
4508 | #endif |
4509 | ||
1da177e4 LT |
4510 | /* |
4511 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4512 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4513 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4514 | * | |
4515 | * The reserve ratio obviously has absolutely no relation with the | |
4516 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
4517 | * if in function of the boot time zone sizes. | |
4518 | */ | |
4519 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4520 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4521 | { | |
4522 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4523 | setup_per_zone_lowmem_reserve(); | |
4524 | return 0; | |
4525 | } | |
4526 | ||
8ad4b1fb RS |
4527 | /* |
4528 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4529 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4530 | * can have before it gets flushed back to buddy allocator. | |
4531 | */ | |
4532 | ||
4533 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4534 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4535 | { | |
4536 | struct zone *zone; | |
4537 | unsigned int cpu; | |
4538 | int ret; | |
4539 | ||
4540 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4541 | if (!write || (ret == -EINVAL)) | |
4542 | return ret; | |
4543 | for_each_zone(zone) { | |
4544 | for_each_online_cpu(cpu) { | |
4545 | unsigned long high; | |
4546 | high = zone->present_pages / percpu_pagelist_fraction; | |
4547 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4548 | } | |
4549 | } | |
4550 | return 0; | |
4551 | } | |
4552 | ||
f034b5d4 | 4553 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4554 | |
4555 | #ifdef CONFIG_NUMA | |
4556 | static int __init set_hashdist(char *str) | |
4557 | { | |
4558 | if (!str) | |
4559 | return 0; | |
4560 | hashdist = simple_strtoul(str, &str, 0); | |
4561 | return 1; | |
4562 | } | |
4563 | __setup("hashdist=", set_hashdist); | |
4564 | #endif | |
4565 | ||
4566 | /* | |
4567 | * allocate a large system hash table from bootmem | |
4568 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4569 | * quantity of entries | |
4570 | * - limit is the number of hash buckets, not the total allocation size | |
4571 | */ | |
4572 | void *__init alloc_large_system_hash(const char *tablename, | |
4573 | unsigned long bucketsize, | |
4574 | unsigned long numentries, | |
4575 | int scale, | |
4576 | int flags, | |
4577 | unsigned int *_hash_shift, | |
4578 | unsigned int *_hash_mask, | |
4579 | unsigned long limit) | |
4580 | { | |
4581 | unsigned long long max = limit; | |
4582 | unsigned long log2qty, size; | |
4583 | void *table = NULL; | |
4584 | ||
4585 | /* allow the kernel cmdline to have a say */ | |
4586 | if (!numentries) { | |
4587 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4588 | numentries = nr_kernel_pages; |
1da177e4 LT |
4589 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4590 | numentries >>= 20 - PAGE_SHIFT; | |
4591 | numentries <<= 20 - PAGE_SHIFT; | |
4592 | ||
4593 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4594 | if (scale > PAGE_SHIFT) | |
4595 | numentries >>= (scale - PAGE_SHIFT); | |
4596 | else | |
4597 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4598 | |
4599 | /* Make sure we've got at least a 0-order allocation.. */ | |
4600 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4601 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4602 | } |
6e692ed3 | 4603 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4604 | |
4605 | /* limit allocation size to 1/16 total memory by default */ | |
4606 | if (max == 0) { | |
4607 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4608 | do_div(max, bucketsize); | |
4609 | } | |
4610 | ||
4611 | if (numentries > max) | |
4612 | numentries = max; | |
4613 | ||
f0d1b0b3 | 4614 | log2qty = ilog2(numentries); |
1da177e4 LT |
4615 | |
4616 | do { | |
4617 | size = bucketsize << log2qty; | |
4618 | if (flags & HASH_EARLY) | |
74768ed8 | 4619 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
4620 | else if (hashdist) |
4621 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4622 | else { | |
2309f9e6 | 4623 | unsigned long order = get_order(size); |
6c0db466 HD |
4624 | |
4625 | if (order < MAX_ORDER) | |
4626 | table = (void *)__get_free_pages(GFP_ATOMIC, | |
4627 | order); | |
1037b83b ED |
4628 | /* |
4629 | * If bucketsize is not a power-of-two, we may free | |
4630 | * some pages at the end of hash table. | |
4631 | */ | |
4632 | if (table) { | |
4633 | unsigned long alloc_end = (unsigned long)table + | |
4634 | (PAGE_SIZE << order); | |
4635 | unsigned long used = (unsigned long)table + | |
4636 | PAGE_ALIGN(size); | |
4637 | split_page(virt_to_page(table), order); | |
4638 | while (used < alloc_end) { | |
4639 | free_page(used); | |
4640 | used += PAGE_SIZE; | |
4641 | } | |
4642 | } | |
1da177e4 LT |
4643 | } |
4644 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4645 | ||
4646 | if (!table) | |
4647 | panic("Failed to allocate %s hash table\n", tablename); | |
4648 | ||
b49ad484 | 4649 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4650 | tablename, |
4651 | (1U << log2qty), | |
f0d1b0b3 | 4652 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4653 | size); |
4654 | ||
4655 | if (_hash_shift) | |
4656 | *_hash_shift = log2qty; | |
4657 | if (_hash_mask) | |
4658 | *_hash_mask = (1 << log2qty) - 1; | |
4659 | ||
dbb1f81c CM |
4660 | /* |
4661 | * If hashdist is set, the table allocation is done with __vmalloc() | |
4662 | * which invokes the kmemleak_alloc() callback. This function may also | |
4663 | * be called before the slab and kmemleak are initialised when | |
4664 | * kmemleak simply buffers the request to be executed later | |
4665 | * (GFP_ATOMIC flag ignored in this case). | |
4666 | */ | |
4667 | if (!hashdist) | |
4668 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); | |
4669 | ||
1da177e4 LT |
4670 | return table; |
4671 | } | |
a117e66e | 4672 | |
835c134e MG |
4673 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4674 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4675 | unsigned long pfn) | |
4676 | { | |
4677 | #ifdef CONFIG_SPARSEMEM | |
4678 | return __pfn_to_section(pfn)->pageblock_flags; | |
4679 | #else | |
4680 | return zone->pageblock_flags; | |
4681 | #endif /* CONFIG_SPARSEMEM */ | |
4682 | } | |
4683 | ||
4684 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4685 | { | |
4686 | #ifdef CONFIG_SPARSEMEM | |
4687 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4688 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4689 | #else |
4690 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4691 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4692 | #endif /* CONFIG_SPARSEMEM */ |
4693 | } | |
4694 | ||
4695 | /** | |
d9c23400 | 4696 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4697 | * @page: The page within the block of interest |
4698 | * @start_bitidx: The first bit of interest to retrieve | |
4699 | * @end_bitidx: The last bit of interest | |
4700 | * returns pageblock_bits flags | |
4701 | */ | |
4702 | unsigned long get_pageblock_flags_group(struct page *page, | |
4703 | int start_bitidx, int end_bitidx) | |
4704 | { | |
4705 | struct zone *zone; | |
4706 | unsigned long *bitmap; | |
4707 | unsigned long pfn, bitidx; | |
4708 | unsigned long flags = 0; | |
4709 | unsigned long value = 1; | |
4710 | ||
4711 | zone = page_zone(page); | |
4712 | pfn = page_to_pfn(page); | |
4713 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4714 | bitidx = pfn_to_bitidx(zone, pfn); | |
4715 | ||
4716 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4717 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4718 | flags |= value; | |
6220ec78 | 4719 | |
835c134e MG |
4720 | return flags; |
4721 | } | |
4722 | ||
4723 | /** | |
d9c23400 | 4724 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4725 | * @page: The page within the block of interest |
4726 | * @start_bitidx: The first bit of interest | |
4727 | * @end_bitidx: The last bit of interest | |
4728 | * @flags: The flags to set | |
4729 | */ | |
4730 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4731 | int start_bitidx, int end_bitidx) | |
4732 | { | |
4733 | struct zone *zone; | |
4734 | unsigned long *bitmap; | |
4735 | unsigned long pfn, bitidx; | |
4736 | unsigned long value = 1; | |
4737 | ||
4738 | zone = page_zone(page); | |
4739 | pfn = page_to_pfn(page); | |
4740 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4741 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
4742 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
4743 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
4744 | |
4745 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4746 | if (flags & value) | |
4747 | __set_bit(bitidx + start_bitidx, bitmap); | |
4748 | else | |
4749 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4750 | } | |
a5d76b54 KH |
4751 | |
4752 | /* | |
4753 | * This is designed as sub function...plz see page_isolation.c also. | |
4754 | * set/clear page block's type to be ISOLATE. | |
4755 | * page allocater never alloc memory from ISOLATE block. | |
4756 | */ | |
4757 | ||
4758 | int set_migratetype_isolate(struct page *page) | |
4759 | { | |
4760 | struct zone *zone; | |
4761 | unsigned long flags; | |
4762 | int ret = -EBUSY; | |
4763 | ||
4764 | zone = page_zone(page); | |
4765 | spin_lock_irqsave(&zone->lock, flags); | |
4766 | /* | |
4767 | * In future, more migrate types will be able to be isolation target. | |
4768 | */ | |
4769 | if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE) | |
4770 | goto out; | |
4771 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
4772 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
4773 | ret = 0; | |
4774 | out: | |
4775 | spin_unlock_irqrestore(&zone->lock, flags); | |
4776 | if (!ret) | |
9f8f2172 | 4777 | drain_all_pages(); |
a5d76b54 KH |
4778 | return ret; |
4779 | } | |
4780 | ||
4781 | void unset_migratetype_isolate(struct page *page) | |
4782 | { | |
4783 | struct zone *zone; | |
4784 | unsigned long flags; | |
4785 | zone = page_zone(page); | |
4786 | spin_lock_irqsave(&zone->lock, flags); | |
4787 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
4788 | goto out; | |
4789 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4790 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4791 | out: | |
4792 | spin_unlock_irqrestore(&zone->lock, flags); | |
4793 | } | |
0c0e6195 KH |
4794 | |
4795 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
4796 | /* | |
4797 | * All pages in the range must be isolated before calling this. | |
4798 | */ | |
4799 | void | |
4800 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
4801 | { | |
4802 | struct page *page; | |
4803 | struct zone *zone; | |
4804 | int order, i; | |
4805 | unsigned long pfn; | |
4806 | unsigned long flags; | |
4807 | /* find the first valid pfn */ | |
4808 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
4809 | if (pfn_valid(pfn)) | |
4810 | break; | |
4811 | if (pfn == end_pfn) | |
4812 | return; | |
4813 | zone = page_zone(pfn_to_page(pfn)); | |
4814 | spin_lock_irqsave(&zone->lock, flags); | |
4815 | pfn = start_pfn; | |
4816 | while (pfn < end_pfn) { | |
4817 | if (!pfn_valid(pfn)) { | |
4818 | pfn++; | |
4819 | continue; | |
4820 | } | |
4821 | page = pfn_to_page(pfn); | |
4822 | BUG_ON(page_count(page)); | |
4823 | BUG_ON(!PageBuddy(page)); | |
4824 | order = page_order(page); | |
4825 | #ifdef CONFIG_DEBUG_VM | |
4826 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
4827 | pfn, 1 << order, end_pfn); | |
4828 | #endif | |
4829 | list_del(&page->lru); | |
4830 | rmv_page_order(page); | |
4831 | zone->free_area[order].nr_free--; | |
4832 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
4833 | - (1UL << order)); | |
4834 | for (i = 0; i < (1 << order); i++) | |
4835 | SetPageReserved((page+i)); | |
4836 | pfn += (1 << order); | |
4837 | } | |
4838 | spin_unlock_irqrestore(&zone->lock, flags); | |
4839 | } | |
4840 | #endif |