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> |
b1eeab67 | 26 | #include <linux/kmemcheck.h> |
1da177e4 LT |
27 | #include <linux/module.h> |
28 | #include <linux/suspend.h> | |
29 | #include <linux/pagevec.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/slab.h> | |
5a3135c2 | 32 | #include <linux/oom.h> |
1da177e4 LT |
33 | #include <linux/notifier.h> |
34 | #include <linux/topology.h> | |
35 | #include <linux/sysctl.h> | |
36 | #include <linux/cpu.h> | |
37 | #include <linux/cpuset.h> | |
bdc8cb98 | 38 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
39 | #include <linux/nodemask.h> |
40 | #include <linux/vmalloc.h> | |
4be38e35 | 41 | #include <linux/mempolicy.h> |
6811378e | 42 | #include <linux/stop_machine.h> |
c713216d MG |
43 | #include <linux/sort.h> |
44 | #include <linux/pfn.h> | |
3fcfab16 | 45 | #include <linux/backing-dev.h> |
933e312e | 46 | #include <linux/fault-inject.h> |
a5d76b54 | 47 | #include <linux/page-isolation.h> |
52d4b9ac | 48 | #include <linux/page_cgroup.h> |
3ac7fe5a | 49 | #include <linux/debugobjects.h> |
dbb1f81c | 50 | #include <linux/kmemleak.h> |
1da177e4 LT |
51 | |
52 | #include <asm/tlbflush.h> | |
ac924c60 | 53 | #include <asm/div64.h> |
1da177e4 LT |
54 | #include "internal.h" |
55 | ||
56 | /* | |
13808910 | 57 | * Array of node states. |
1da177e4 | 58 | */ |
13808910 CL |
59 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
60 | [N_POSSIBLE] = NODE_MASK_ALL, | |
61 | [N_ONLINE] = { { [0] = 1UL } }, | |
62 | #ifndef CONFIG_NUMA | |
63 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
64 | #ifdef CONFIG_HIGHMEM | |
65 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
66 | #endif | |
67 | [N_CPU] = { { [0] = 1UL } }, | |
68 | #endif /* NUMA */ | |
69 | }; | |
70 | EXPORT_SYMBOL(node_states); | |
71 | ||
6c231b7b | 72 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 73 | unsigned long totalreserve_pages __read_mostly; |
22b31eec | 74 | unsigned long highest_memmap_pfn __read_mostly; |
8ad4b1fb | 75 | int percpu_pagelist_fraction; |
dcce284a | 76 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 77 | |
d9c23400 MG |
78 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
79 | int pageblock_order __read_mostly; | |
80 | #endif | |
81 | ||
d98c7a09 | 82 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 83 | |
1da177e4 LT |
84 | /* |
85 | * results with 256, 32 in the lowmem_reserve sysctl: | |
86 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
87 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
88 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
89 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
90 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
91 | * |
92 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
93 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 94 | */ |
2f1b6248 | 95 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 96 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 97 | 256, |
4b51d669 | 98 | #endif |
fb0e7942 | 99 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 100 | 256, |
fb0e7942 | 101 | #endif |
e53ef38d | 102 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 103 | 32, |
e53ef38d | 104 | #endif |
2a1e274a | 105 | 32, |
2f1b6248 | 106 | }; |
1da177e4 LT |
107 | |
108 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 109 | |
15ad7cdc | 110 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 111 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 112 | "DMA", |
4b51d669 | 113 | #endif |
fb0e7942 | 114 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 115 | "DMA32", |
fb0e7942 | 116 | #endif |
2f1b6248 | 117 | "Normal", |
e53ef38d | 118 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 119 | "HighMem", |
e53ef38d | 120 | #endif |
2a1e274a | 121 | "Movable", |
2f1b6248 CL |
122 | }; |
123 | ||
1da177e4 LT |
124 | int min_free_kbytes = 1024; |
125 | ||
86356ab1 YG |
126 | unsigned long __meminitdata nr_kernel_pages; |
127 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 128 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 129 | |
c713216d MG |
130 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
131 | /* | |
183ff22b | 132 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
133 | * ranges of memory (RAM) that may be registered with add_active_range(). |
134 | * Ranges passed to add_active_range() will be merged if possible | |
135 | * so the number of times add_active_range() can be called is | |
136 | * related to the number of nodes and the number of holes | |
137 | */ | |
138 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
139 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
140 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
141 | #else | |
142 | #if MAX_NUMNODES >= 32 | |
143 | /* If there can be many nodes, allow up to 50 holes per node */ | |
144 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
145 | #else | |
146 | /* By default, allow up to 256 distinct regions */ | |
147 | #define MAX_ACTIVE_REGIONS 256 | |
148 | #endif | |
149 | #endif | |
150 | ||
98011f56 JB |
151 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
152 | static int __meminitdata nr_nodemap_entries; | |
153 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
154 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
b69a7288 | 155 | static unsigned long __initdata required_kernelcore; |
484f51f8 | 156 | static unsigned long __initdata required_movablecore; |
b69a7288 | 157 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
158 | |
159 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
160 | int movable_zone; | |
161 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
162 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
163 | ||
418508c1 MS |
164 | #if MAX_NUMNODES > 1 |
165 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 166 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 167 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 168 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
169 | #endif |
170 | ||
9ef9acb0 MG |
171 | int page_group_by_mobility_disabled __read_mostly; |
172 | ||
b2a0ac88 MG |
173 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
174 | { | |
49255c61 MG |
175 | |
176 | if (unlikely(page_group_by_mobility_disabled)) | |
177 | migratetype = MIGRATE_UNMOVABLE; | |
178 | ||
b2a0ac88 MG |
179 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
180 | PB_migrate, PB_migrate_end); | |
181 | } | |
182 | ||
7f33d49a RW |
183 | bool oom_killer_disabled __read_mostly; |
184 | ||
13e7444b | 185 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 186 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 187 | { |
bdc8cb98 DH |
188 | int ret = 0; |
189 | unsigned seq; | |
190 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 191 | |
bdc8cb98 DH |
192 | do { |
193 | seq = zone_span_seqbegin(zone); | |
194 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
195 | ret = 1; | |
196 | else if (pfn < zone->zone_start_pfn) | |
197 | ret = 1; | |
198 | } while (zone_span_seqretry(zone, seq)); | |
199 | ||
200 | return ret; | |
c6a57e19 DH |
201 | } |
202 | ||
203 | static int page_is_consistent(struct zone *zone, struct page *page) | |
204 | { | |
14e07298 | 205 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 206 | return 0; |
1da177e4 | 207 | if (zone != page_zone(page)) |
c6a57e19 DH |
208 | return 0; |
209 | ||
210 | return 1; | |
211 | } | |
212 | /* | |
213 | * Temporary debugging check for pages not lying within a given zone. | |
214 | */ | |
215 | static int bad_range(struct zone *zone, struct page *page) | |
216 | { | |
217 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 218 | return 1; |
c6a57e19 DH |
219 | if (!page_is_consistent(zone, page)) |
220 | return 1; | |
221 | ||
1da177e4 LT |
222 | return 0; |
223 | } | |
13e7444b NP |
224 | #else |
225 | static inline int bad_range(struct zone *zone, struct page *page) | |
226 | { | |
227 | return 0; | |
228 | } | |
229 | #endif | |
230 | ||
224abf92 | 231 | static void bad_page(struct page *page) |
1da177e4 | 232 | { |
d936cf9b HD |
233 | static unsigned long resume; |
234 | static unsigned long nr_shown; | |
235 | static unsigned long nr_unshown; | |
236 | ||
237 | /* | |
238 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
239 | * or allow a steady drip of one report per second. | |
240 | */ | |
241 | if (nr_shown == 60) { | |
242 | if (time_before(jiffies, resume)) { | |
243 | nr_unshown++; | |
244 | goto out; | |
245 | } | |
246 | if (nr_unshown) { | |
1e9e6365 HD |
247 | printk(KERN_ALERT |
248 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
249 | nr_unshown); |
250 | nr_unshown = 0; | |
251 | } | |
252 | nr_shown = 0; | |
253 | } | |
254 | if (nr_shown++ == 0) | |
255 | resume = jiffies + 60 * HZ; | |
256 | ||
1e9e6365 | 257 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 258 | current->comm, page_to_pfn(page)); |
1e9e6365 | 259 | printk(KERN_ALERT |
3dc14741 HD |
260 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
261 | page, (void *)page->flags, page_count(page), | |
262 | page_mapcount(page), page->mapping, page->index); | |
3dc14741 | 263 | |
1da177e4 | 264 | dump_stack(); |
d936cf9b | 265 | out: |
8cc3b392 HD |
266 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
267 | __ClearPageBuddy(page); | |
9f158333 | 268 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
269 | } |
270 | ||
1da177e4 LT |
271 | /* |
272 | * Higher-order pages are called "compound pages". They are structured thusly: | |
273 | * | |
274 | * The first PAGE_SIZE page is called the "head page". | |
275 | * | |
276 | * The remaining PAGE_SIZE pages are called "tail pages". | |
277 | * | |
278 | * All pages have PG_compound set. All pages have their ->private pointing at | |
279 | * the head page (even the head page has this). | |
280 | * | |
41d78ba5 HD |
281 | * The first tail page's ->lru.next holds the address of the compound page's |
282 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
283 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 284 | */ |
d98c7a09 HD |
285 | |
286 | static void free_compound_page(struct page *page) | |
287 | { | |
d85f3385 | 288 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
289 | } |
290 | ||
01ad1c08 | 291 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
292 | { |
293 | int i; | |
294 | int nr_pages = 1 << order; | |
295 | ||
296 | set_compound_page_dtor(page, free_compound_page); | |
297 | set_compound_order(page, order); | |
298 | __SetPageHead(page); | |
299 | for (i = 1; i < nr_pages; i++) { | |
300 | struct page *p = page + i; | |
301 | ||
302 | __SetPageTail(p); | |
303 | p->first_page = page; | |
304 | } | |
305 | } | |
306 | ||
8cc3b392 | 307 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
308 | { |
309 | int i; | |
310 | int nr_pages = 1 << order; | |
8cc3b392 | 311 | int bad = 0; |
1da177e4 | 312 | |
8cc3b392 HD |
313 | if (unlikely(compound_order(page) != order) || |
314 | unlikely(!PageHead(page))) { | |
224abf92 | 315 | bad_page(page); |
8cc3b392 HD |
316 | bad++; |
317 | } | |
1da177e4 | 318 | |
6d777953 | 319 | __ClearPageHead(page); |
8cc3b392 | 320 | |
18229df5 AW |
321 | for (i = 1; i < nr_pages; i++) { |
322 | struct page *p = page + i; | |
1da177e4 | 323 | |
e713a21d | 324 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 325 | bad_page(page); |
8cc3b392 HD |
326 | bad++; |
327 | } | |
d85f3385 | 328 | __ClearPageTail(p); |
1da177e4 | 329 | } |
8cc3b392 HD |
330 | |
331 | return bad; | |
1da177e4 | 332 | } |
1da177e4 | 333 | |
17cf4406 NP |
334 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
335 | { | |
336 | int i; | |
337 | ||
6626c5d5 AM |
338 | /* |
339 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
340 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
341 | */ | |
725d704e | 342 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
343 | for (i = 0; i < (1 << order); i++) |
344 | clear_highpage(page + i); | |
345 | } | |
346 | ||
6aa3001b AM |
347 | static inline void set_page_order(struct page *page, int order) |
348 | { | |
4c21e2f2 | 349 | set_page_private(page, order); |
676165a8 | 350 | __SetPageBuddy(page); |
1da177e4 LT |
351 | } |
352 | ||
353 | static inline void rmv_page_order(struct page *page) | |
354 | { | |
676165a8 | 355 | __ClearPageBuddy(page); |
4c21e2f2 | 356 | set_page_private(page, 0); |
1da177e4 LT |
357 | } |
358 | ||
359 | /* | |
360 | * Locate the struct page for both the matching buddy in our | |
361 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
362 | * | |
363 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
364 | * the following equation: | |
365 | * B2 = B1 ^ (1 << O) | |
366 | * For example, if the starting buddy (buddy2) is #8 its order | |
367 | * 1 buddy is #10: | |
368 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
369 | * | |
370 | * 2) Any buddy B will have an order O+1 parent P which | |
371 | * satisfies the following equation: | |
372 | * P = B & ~(1 << O) | |
373 | * | |
d6e05edc | 374 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
375 | */ |
376 | static inline struct page * | |
377 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
378 | { | |
379 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
380 | ||
381 | return page + (buddy_idx - page_idx); | |
382 | } | |
383 | ||
384 | static inline unsigned long | |
385 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
386 | { | |
387 | return (page_idx & ~(1 << order)); | |
388 | } | |
389 | ||
390 | /* | |
391 | * This function checks whether a page is free && is the buddy | |
392 | * we can do coalesce a page and its buddy if | |
13e7444b | 393 | * (a) the buddy is not in a hole && |
676165a8 | 394 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
395 | * (c) a page and its buddy have the same order && |
396 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
397 | * |
398 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
399 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 400 | * |
676165a8 | 401 | * For recording page's order, we use page_private(page). |
1da177e4 | 402 | */ |
cb2b95e1 AW |
403 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
404 | int order) | |
1da177e4 | 405 | { |
14e07298 | 406 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 407 | return 0; |
13e7444b | 408 | |
cb2b95e1 AW |
409 | if (page_zone_id(page) != page_zone_id(buddy)) |
410 | return 0; | |
411 | ||
412 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
a3af9c38 | 413 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 414 | return 1; |
676165a8 | 415 | } |
6aa3001b | 416 | return 0; |
1da177e4 LT |
417 | } |
418 | ||
419 | /* | |
420 | * Freeing function for a buddy system allocator. | |
421 | * | |
422 | * The concept of a buddy system is to maintain direct-mapped table | |
423 | * (containing bit values) for memory blocks of various "orders". | |
424 | * The bottom level table contains the map for the smallest allocatable | |
425 | * units of memory (here, pages), and each level above it describes | |
426 | * pairs of units from the levels below, hence, "buddies". | |
427 | * At a high level, all that happens here is marking the table entry | |
428 | * at the bottom level available, and propagating the changes upward | |
429 | * as necessary, plus some accounting needed to play nicely with other | |
430 | * parts of the VM system. | |
431 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 432 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 433 | * order is recorded in page_private(page) field. |
1da177e4 LT |
434 | * So when we are allocating or freeing one, we can derive the state of the |
435 | * other. That is, if we allocate a small block, and both were | |
436 | * free, the remainder of the region must be split into blocks. | |
437 | * If a block is freed, and its buddy is also free, then this | |
438 | * triggers coalescing into a block of larger size. | |
439 | * | |
440 | * -- wli | |
441 | */ | |
442 | ||
48db57f8 | 443 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
444 | struct zone *zone, unsigned int order, |
445 | int migratetype) | |
1da177e4 LT |
446 | { |
447 | unsigned long page_idx; | |
1da177e4 | 448 | |
224abf92 | 449 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
450 | if (unlikely(destroy_compound_page(page, order))) |
451 | return; | |
1da177e4 | 452 | |
ed0ae21d MG |
453 | VM_BUG_ON(migratetype == -1); |
454 | ||
1da177e4 LT |
455 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
456 | ||
f2260e6b | 457 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 458 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 459 | |
1da177e4 LT |
460 | while (order < MAX_ORDER-1) { |
461 | unsigned long combined_idx; | |
1da177e4 LT |
462 | struct page *buddy; |
463 | ||
1da177e4 | 464 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 465 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 466 | break; |
13e7444b | 467 | |
3c82d0ce | 468 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 469 | list_del(&buddy->lru); |
b2a0ac88 | 470 | zone->free_area[order].nr_free--; |
1da177e4 | 471 | rmv_page_order(buddy); |
13e7444b | 472 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
473 | page = page + (combined_idx - page_idx); |
474 | page_idx = combined_idx; | |
475 | order++; | |
476 | } | |
477 | set_page_order(page, order); | |
b2a0ac88 MG |
478 | list_add(&page->lru, |
479 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
480 | zone->free_area[order].nr_free++; |
481 | } | |
482 | ||
092cead6 KM |
483 | #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT |
484 | /* | |
485 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
486 | * Page should not be on lru, so no need to fix that up. | |
487 | * free_pages_check() will verify... | |
488 | */ | |
489 | static inline void free_page_mlock(struct page *page) | |
490 | { | |
092cead6 KM |
491 | __dec_zone_page_state(page, NR_MLOCK); |
492 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
493 | } | |
494 | #else | |
495 | static void free_page_mlock(struct page *page) { } | |
496 | #endif | |
497 | ||
224abf92 | 498 | static inline int free_pages_check(struct page *page) |
1da177e4 | 499 | { |
92be2e33 NP |
500 | if (unlikely(page_mapcount(page) | |
501 | (page->mapping != NULL) | | |
a3af9c38 | 502 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 503 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
224abf92 | 504 | bad_page(page); |
79f4b7bf | 505 | return 1; |
8cc3b392 | 506 | } |
79f4b7bf HD |
507 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
508 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
509 | return 0; | |
1da177e4 LT |
510 | } |
511 | ||
512 | /* | |
513 | * Frees a list of pages. | |
514 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 515 | * count is the number of pages to free. |
1da177e4 LT |
516 | * |
517 | * If the zone was previously in an "all pages pinned" state then look to | |
518 | * see if this freeing clears that state. | |
519 | * | |
520 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
521 | * pinned" detection logic. | |
522 | */ | |
48db57f8 NP |
523 | static void free_pages_bulk(struct zone *zone, int count, |
524 | struct list_head *list, int order) | |
1da177e4 | 525 | { |
c54ad30c | 526 | spin_lock(&zone->lock); |
e815af95 | 527 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 528 | zone->pages_scanned = 0; |
f2260e6b MG |
529 | |
530 | __mod_zone_page_state(zone, NR_FREE_PAGES, count << order); | |
48db57f8 NP |
531 | while (count--) { |
532 | struct page *page; | |
533 | ||
725d704e | 534 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 535 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 536 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 537 | list_del(&page->lru); |
ed0ae21d | 538 | __free_one_page(page, zone, order, page_private(page)); |
1da177e4 | 539 | } |
c54ad30c | 540 | spin_unlock(&zone->lock); |
1da177e4 LT |
541 | } |
542 | ||
ed0ae21d MG |
543 | static void free_one_page(struct zone *zone, struct page *page, int order, |
544 | int migratetype) | |
1da177e4 | 545 | { |
006d22d9 | 546 | spin_lock(&zone->lock); |
e815af95 | 547 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 548 | zone->pages_scanned = 0; |
f2260e6b MG |
549 | |
550 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); | |
ed0ae21d | 551 | __free_one_page(page, zone, order, migratetype); |
006d22d9 | 552 | spin_unlock(&zone->lock); |
48db57f8 NP |
553 | } |
554 | ||
555 | static void __free_pages_ok(struct page *page, unsigned int order) | |
556 | { | |
557 | unsigned long flags; | |
1da177e4 | 558 | int i; |
8cc3b392 | 559 | int bad = 0; |
c277331d | 560 | int wasMlocked = TestClearPageMlocked(page); |
1da177e4 | 561 | |
b1eeab67 VN |
562 | kmemcheck_free_shadow(page, order); |
563 | ||
1da177e4 | 564 | for (i = 0 ; i < (1 << order) ; ++i) |
8cc3b392 HD |
565 | bad += free_pages_check(page + i); |
566 | if (bad) | |
689bcebf HD |
567 | return; |
568 | ||
3ac7fe5a | 569 | if (!PageHighMem(page)) { |
9858db50 | 570 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
571 | debug_check_no_obj_freed(page_address(page), |
572 | PAGE_SIZE << order); | |
573 | } | |
dafb1367 | 574 | arch_free_page(page, order); |
48db57f8 | 575 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 576 | |
c54ad30c | 577 | local_irq_save(flags); |
c277331d | 578 | if (unlikely(wasMlocked)) |
da456f14 | 579 | free_page_mlock(page); |
f8891e5e | 580 | __count_vm_events(PGFREE, 1 << order); |
ed0ae21d MG |
581 | free_one_page(page_zone(page), page, order, |
582 | get_pageblock_migratetype(page)); | |
c54ad30c | 583 | local_irq_restore(flags); |
1da177e4 LT |
584 | } |
585 | ||
a226f6c8 DH |
586 | /* |
587 | * permit the bootmem allocator to evade page validation on high-order frees | |
588 | */ | |
af370fb8 | 589 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
590 | { |
591 | if (order == 0) { | |
592 | __ClearPageReserved(page); | |
593 | set_page_count(page, 0); | |
7835e98b | 594 | set_page_refcounted(page); |
545b1ea9 | 595 | __free_page(page); |
a226f6c8 | 596 | } else { |
a226f6c8 DH |
597 | int loop; |
598 | ||
545b1ea9 | 599 | prefetchw(page); |
a226f6c8 DH |
600 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
601 | struct page *p = &page[loop]; | |
602 | ||
545b1ea9 NP |
603 | if (loop + 1 < BITS_PER_LONG) |
604 | prefetchw(p + 1); | |
a226f6c8 DH |
605 | __ClearPageReserved(p); |
606 | set_page_count(p, 0); | |
607 | } | |
608 | ||
7835e98b | 609 | set_page_refcounted(page); |
545b1ea9 | 610 | __free_pages(page, order); |
a226f6c8 DH |
611 | } |
612 | } | |
613 | ||
1da177e4 LT |
614 | |
615 | /* | |
616 | * The order of subdivision here is critical for the IO subsystem. | |
617 | * Please do not alter this order without good reasons and regression | |
618 | * testing. Specifically, as large blocks of memory are subdivided, | |
619 | * the order in which smaller blocks are delivered depends on the order | |
620 | * they're subdivided in this function. This is the primary factor | |
621 | * influencing the order in which pages are delivered to the IO | |
622 | * subsystem according to empirical testing, and this is also justified | |
623 | * by considering the behavior of a buddy system containing a single | |
624 | * large block of memory acted on by a series of small allocations. | |
625 | * This behavior is a critical factor in sglist merging's success. | |
626 | * | |
627 | * -- wli | |
628 | */ | |
085cc7d5 | 629 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
630 | int low, int high, struct free_area *area, |
631 | int migratetype) | |
1da177e4 LT |
632 | { |
633 | unsigned long size = 1 << high; | |
634 | ||
635 | while (high > low) { | |
636 | area--; | |
637 | high--; | |
638 | size >>= 1; | |
725d704e | 639 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 640 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
641 | area->nr_free++; |
642 | set_page_order(&page[size], high); | |
643 | } | |
1da177e4 LT |
644 | } |
645 | ||
1da177e4 LT |
646 | /* |
647 | * This page is about to be returned from the page allocator | |
648 | */ | |
17cf4406 | 649 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 650 | { |
92be2e33 NP |
651 | if (unlikely(page_mapcount(page) | |
652 | (page->mapping != NULL) | | |
a3af9c38 | 653 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 654 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
224abf92 | 655 | bad_page(page); |
689bcebf | 656 | return 1; |
8cc3b392 | 657 | } |
689bcebf | 658 | |
4c21e2f2 | 659 | set_page_private(page, 0); |
7835e98b | 660 | set_page_refcounted(page); |
cc102509 NP |
661 | |
662 | arch_alloc_page(page, order); | |
1da177e4 | 663 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
664 | |
665 | if (gfp_flags & __GFP_ZERO) | |
666 | prep_zero_page(page, order, gfp_flags); | |
667 | ||
668 | if (order && (gfp_flags & __GFP_COMP)) | |
669 | prep_compound_page(page, order); | |
670 | ||
689bcebf | 671 | return 0; |
1da177e4 LT |
672 | } |
673 | ||
56fd56b8 MG |
674 | /* |
675 | * Go through the free lists for the given migratetype and remove | |
676 | * the smallest available page from the freelists | |
677 | */ | |
728ec980 MG |
678 | static inline |
679 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
680 | int migratetype) |
681 | { | |
682 | unsigned int current_order; | |
683 | struct free_area * area; | |
684 | struct page *page; | |
685 | ||
686 | /* Find a page of the appropriate size in the preferred list */ | |
687 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
688 | area = &(zone->free_area[current_order]); | |
689 | if (list_empty(&area->free_list[migratetype])) | |
690 | continue; | |
691 | ||
692 | page = list_entry(area->free_list[migratetype].next, | |
693 | struct page, lru); | |
694 | list_del(&page->lru); | |
695 | rmv_page_order(page); | |
696 | area->nr_free--; | |
56fd56b8 MG |
697 | expand(zone, page, order, current_order, area, migratetype); |
698 | return page; | |
699 | } | |
700 | ||
701 | return NULL; | |
702 | } | |
703 | ||
704 | ||
b2a0ac88 MG |
705 | /* |
706 | * This array describes the order lists are fallen back to when | |
707 | * the free lists for the desirable migrate type are depleted | |
708 | */ | |
709 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
710 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
711 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
712 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
713 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
714 | }; |
715 | ||
c361be55 MG |
716 | /* |
717 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 718 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
719 | * boundary. If alignment is required, use move_freepages_block() |
720 | */ | |
b69a7288 AB |
721 | static int move_freepages(struct zone *zone, |
722 | struct page *start_page, struct page *end_page, | |
723 | int migratetype) | |
c361be55 MG |
724 | { |
725 | struct page *page; | |
726 | unsigned long order; | |
d100313f | 727 | int pages_moved = 0; |
c361be55 MG |
728 | |
729 | #ifndef CONFIG_HOLES_IN_ZONE | |
730 | /* | |
731 | * page_zone is not safe to call in this context when | |
732 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
733 | * anyway as we check zone boundaries in move_freepages_block(). | |
734 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 735 | * grouping pages by mobility |
c361be55 MG |
736 | */ |
737 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
738 | #endif | |
739 | ||
740 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
741 | /* Make sure we are not inadvertently changing nodes */ |
742 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
743 | ||
c361be55 MG |
744 | if (!pfn_valid_within(page_to_pfn(page))) { |
745 | page++; | |
746 | continue; | |
747 | } | |
748 | ||
749 | if (!PageBuddy(page)) { | |
750 | page++; | |
751 | continue; | |
752 | } | |
753 | ||
754 | order = page_order(page); | |
755 | list_del(&page->lru); | |
756 | list_add(&page->lru, | |
757 | &zone->free_area[order].free_list[migratetype]); | |
758 | page += 1 << order; | |
d100313f | 759 | pages_moved += 1 << order; |
c361be55 MG |
760 | } |
761 | ||
d100313f | 762 | return pages_moved; |
c361be55 MG |
763 | } |
764 | ||
b69a7288 AB |
765 | static int move_freepages_block(struct zone *zone, struct page *page, |
766 | int migratetype) | |
c361be55 MG |
767 | { |
768 | unsigned long start_pfn, end_pfn; | |
769 | struct page *start_page, *end_page; | |
770 | ||
771 | start_pfn = page_to_pfn(page); | |
d9c23400 | 772 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 773 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
774 | end_page = start_page + pageblock_nr_pages - 1; |
775 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
776 | |
777 | /* Do not cross zone boundaries */ | |
778 | if (start_pfn < zone->zone_start_pfn) | |
779 | start_page = page; | |
780 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
781 | return 0; | |
782 | ||
783 | return move_freepages(zone, start_page, end_page, migratetype); | |
784 | } | |
785 | ||
b2a0ac88 | 786 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
787 | static inline struct page * |
788 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
789 | { |
790 | struct free_area * area; | |
791 | int current_order; | |
792 | struct page *page; | |
793 | int migratetype, i; | |
794 | ||
795 | /* Find the largest possible block of pages in the other list */ | |
796 | for (current_order = MAX_ORDER-1; current_order >= order; | |
797 | --current_order) { | |
798 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
799 | migratetype = fallbacks[start_migratetype][i]; | |
800 | ||
56fd56b8 MG |
801 | /* MIGRATE_RESERVE handled later if necessary */ |
802 | if (migratetype == MIGRATE_RESERVE) | |
803 | continue; | |
e010487d | 804 | |
b2a0ac88 MG |
805 | area = &(zone->free_area[current_order]); |
806 | if (list_empty(&area->free_list[migratetype])) | |
807 | continue; | |
808 | ||
809 | page = list_entry(area->free_list[migratetype].next, | |
810 | struct page, lru); | |
811 | area->nr_free--; | |
812 | ||
813 | /* | |
c361be55 | 814 | * If breaking a large block of pages, move all free |
46dafbca MG |
815 | * pages to the preferred allocation list. If falling |
816 | * back for a reclaimable kernel allocation, be more | |
817 | * agressive about taking ownership of free pages | |
b2a0ac88 | 818 | */ |
d9c23400 | 819 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
46dafbca MG |
820 | start_migratetype == MIGRATE_RECLAIMABLE) { |
821 | unsigned long pages; | |
822 | pages = move_freepages_block(zone, page, | |
823 | start_migratetype); | |
824 | ||
825 | /* Claim the whole block if over half of it is free */ | |
d9c23400 | 826 | if (pages >= (1 << (pageblock_order-1))) |
46dafbca MG |
827 | set_pageblock_migratetype(page, |
828 | start_migratetype); | |
829 | ||
b2a0ac88 | 830 | migratetype = start_migratetype; |
c361be55 | 831 | } |
b2a0ac88 MG |
832 | |
833 | /* Remove the page from the freelists */ | |
834 | list_del(&page->lru); | |
835 | rmv_page_order(page); | |
b2a0ac88 | 836 | |
d9c23400 | 837 | if (current_order == pageblock_order) |
b2a0ac88 MG |
838 | set_pageblock_migratetype(page, |
839 | start_migratetype); | |
840 | ||
841 | expand(zone, page, order, current_order, area, migratetype); | |
842 | return page; | |
843 | } | |
844 | } | |
845 | ||
728ec980 | 846 | return NULL; |
b2a0ac88 MG |
847 | } |
848 | ||
56fd56b8 | 849 | /* |
1da177e4 LT |
850 | * Do the hard work of removing an element from the buddy allocator. |
851 | * Call me with the zone->lock already held. | |
852 | */ | |
b2a0ac88 MG |
853 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
854 | int migratetype) | |
1da177e4 | 855 | { |
1da177e4 LT |
856 | struct page *page; |
857 | ||
728ec980 | 858 | retry_reserve: |
56fd56b8 | 859 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 860 | |
728ec980 | 861 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 862 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 863 | |
728ec980 MG |
864 | /* |
865 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
866 | * is used because __rmqueue_smallest is an inline function | |
867 | * and we want just one call site | |
868 | */ | |
869 | if (!page) { | |
870 | migratetype = MIGRATE_RESERVE; | |
871 | goto retry_reserve; | |
872 | } | |
873 | } | |
874 | ||
b2a0ac88 | 875 | return page; |
1da177e4 LT |
876 | } |
877 | ||
878 | /* | |
879 | * Obtain a specified number of elements from the buddy allocator, all under | |
880 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
881 | * Returns the number of new pages which were placed at *list. | |
882 | */ | |
883 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 | 884 | unsigned long count, struct list_head *list, |
e084b2d9 | 885 | int migratetype, int cold) |
1da177e4 | 886 | { |
1da177e4 | 887 | int i; |
1da177e4 | 888 | |
c54ad30c | 889 | spin_lock(&zone->lock); |
1da177e4 | 890 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 891 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 892 | if (unlikely(page == NULL)) |
1da177e4 | 893 | break; |
81eabcbe MG |
894 | |
895 | /* | |
896 | * Split buddy pages returned by expand() are received here | |
897 | * in physical page order. The page is added to the callers and | |
898 | * list and the list head then moves forward. From the callers | |
899 | * perspective, the linked list is ordered by page number in | |
900 | * some conditions. This is useful for IO devices that can | |
901 | * merge IO requests if the physical pages are ordered | |
902 | * properly. | |
903 | */ | |
e084b2d9 MG |
904 | if (likely(cold == 0)) |
905 | list_add(&page->lru, list); | |
906 | else | |
907 | list_add_tail(&page->lru, list); | |
535131e6 | 908 | set_page_private(page, migratetype); |
81eabcbe | 909 | list = &page->lru; |
1da177e4 | 910 | } |
f2260e6b | 911 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 912 | spin_unlock(&zone->lock); |
085cc7d5 | 913 | return i; |
1da177e4 LT |
914 | } |
915 | ||
4ae7c039 | 916 | #ifdef CONFIG_NUMA |
8fce4d8e | 917 | /* |
4037d452 CL |
918 | * Called from the vmstat counter updater to drain pagesets of this |
919 | * currently executing processor on remote nodes after they have | |
920 | * expired. | |
921 | * | |
879336c3 CL |
922 | * Note that this function must be called with the thread pinned to |
923 | * a single processor. | |
8fce4d8e | 924 | */ |
4037d452 | 925 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 926 | { |
4ae7c039 | 927 | unsigned long flags; |
4037d452 | 928 | int to_drain; |
4ae7c039 | 929 | |
4037d452 CL |
930 | local_irq_save(flags); |
931 | if (pcp->count >= pcp->batch) | |
932 | to_drain = pcp->batch; | |
933 | else | |
934 | to_drain = pcp->count; | |
935 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
936 | pcp->count -= to_drain; | |
937 | local_irq_restore(flags); | |
4ae7c039 CL |
938 | } |
939 | #endif | |
940 | ||
9f8f2172 CL |
941 | /* |
942 | * Drain pages of the indicated processor. | |
943 | * | |
944 | * The processor must either be the current processor and the | |
945 | * thread pinned to the current processor or a processor that | |
946 | * is not online. | |
947 | */ | |
948 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 949 | { |
c54ad30c | 950 | unsigned long flags; |
1da177e4 | 951 | struct zone *zone; |
1da177e4 | 952 | |
ee99c71c | 953 | for_each_populated_zone(zone) { |
1da177e4 | 954 | struct per_cpu_pageset *pset; |
3dfa5721 | 955 | struct per_cpu_pages *pcp; |
1da177e4 | 956 | |
e7c8d5c9 | 957 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
958 | |
959 | pcp = &pset->pcp; | |
960 | local_irq_save(flags); | |
961 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
962 | pcp->count = 0; | |
963 | local_irq_restore(flags); | |
1da177e4 LT |
964 | } |
965 | } | |
1da177e4 | 966 | |
9f8f2172 CL |
967 | /* |
968 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
969 | */ | |
970 | void drain_local_pages(void *arg) | |
971 | { | |
972 | drain_pages(smp_processor_id()); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
977 | */ | |
978 | void drain_all_pages(void) | |
979 | { | |
15c8b6c1 | 980 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
981 | } |
982 | ||
296699de | 983 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
984 | |
985 | void mark_free_pages(struct zone *zone) | |
986 | { | |
f623f0db RW |
987 | unsigned long pfn, max_zone_pfn; |
988 | unsigned long flags; | |
b2a0ac88 | 989 | int order, t; |
1da177e4 LT |
990 | struct list_head *curr; |
991 | ||
992 | if (!zone->spanned_pages) | |
993 | return; | |
994 | ||
995 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
996 | |
997 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
998 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
999 | if (pfn_valid(pfn)) { | |
1000 | struct page *page = pfn_to_page(pfn); | |
1001 | ||
7be98234 RW |
1002 | if (!swsusp_page_is_forbidden(page)) |
1003 | swsusp_unset_page_free(page); | |
f623f0db | 1004 | } |
1da177e4 | 1005 | |
b2a0ac88 MG |
1006 | for_each_migratetype_order(order, t) { |
1007 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1008 | unsigned long i; |
1da177e4 | 1009 | |
f623f0db RW |
1010 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1011 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1012 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1013 | } |
b2a0ac88 | 1014 | } |
1da177e4 LT |
1015 | spin_unlock_irqrestore(&zone->lock, flags); |
1016 | } | |
e2c55dc8 | 1017 | #endif /* CONFIG_PM */ |
1da177e4 | 1018 | |
1da177e4 LT |
1019 | /* |
1020 | * Free a 0-order page | |
1021 | */ | |
920c7a5d | 1022 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1023 | { |
1024 | struct zone *zone = page_zone(page); | |
1025 | struct per_cpu_pages *pcp; | |
1026 | unsigned long flags; | |
c277331d | 1027 | int wasMlocked = TestClearPageMlocked(page); |
1da177e4 | 1028 | |
b1eeab67 VN |
1029 | kmemcheck_free_shadow(page, 0); |
1030 | ||
1da177e4 LT |
1031 | if (PageAnon(page)) |
1032 | page->mapping = NULL; | |
224abf92 | 1033 | if (free_pages_check(page)) |
689bcebf HD |
1034 | return; |
1035 | ||
3ac7fe5a | 1036 | if (!PageHighMem(page)) { |
9858db50 | 1037 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); |
3ac7fe5a TG |
1038 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); |
1039 | } | |
dafb1367 | 1040 | arch_free_page(page, 0); |
689bcebf HD |
1041 | kernel_map_pages(page, 1, 0); |
1042 | ||
3dfa5721 | 1043 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
974709bd | 1044 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 1045 | local_irq_save(flags); |
c277331d | 1046 | if (unlikely(wasMlocked)) |
da456f14 | 1047 | free_page_mlock(page); |
f8891e5e | 1048 | __count_vm_event(PGFREE); |
da456f14 | 1049 | |
3dfa5721 CL |
1050 | if (cold) |
1051 | list_add_tail(&page->lru, &pcp->list); | |
1052 | else | |
1053 | list_add(&page->lru, &pcp->list); | |
1da177e4 | 1054 | pcp->count++; |
48db57f8 NP |
1055 | if (pcp->count >= pcp->high) { |
1056 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
1057 | pcp->count -= pcp->batch; | |
1058 | } | |
1da177e4 LT |
1059 | local_irq_restore(flags); |
1060 | put_cpu(); | |
1061 | } | |
1062 | ||
920c7a5d | 1063 | void free_hot_page(struct page *page) |
1da177e4 LT |
1064 | { |
1065 | free_hot_cold_page(page, 0); | |
1066 | } | |
1067 | ||
920c7a5d | 1068 | void free_cold_page(struct page *page) |
1da177e4 LT |
1069 | { |
1070 | free_hot_cold_page(page, 1); | |
1071 | } | |
1072 | ||
8dfcc9ba NP |
1073 | /* |
1074 | * split_page takes a non-compound higher-order page, and splits it into | |
1075 | * n (1<<order) sub-pages: page[0..n] | |
1076 | * Each sub-page must be freed individually. | |
1077 | * | |
1078 | * Note: this is probably too low level an operation for use in drivers. | |
1079 | * Please consult with lkml before using this in your driver. | |
1080 | */ | |
1081 | void split_page(struct page *page, unsigned int order) | |
1082 | { | |
1083 | int i; | |
1084 | ||
725d704e NP |
1085 | VM_BUG_ON(PageCompound(page)); |
1086 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1087 | |
1088 | #ifdef CONFIG_KMEMCHECK | |
1089 | /* | |
1090 | * Split shadow pages too, because free(page[0]) would | |
1091 | * otherwise free the whole shadow. | |
1092 | */ | |
1093 | if (kmemcheck_page_is_tracked(page)) | |
1094 | split_page(virt_to_page(page[0].shadow), order); | |
1095 | #endif | |
1096 | ||
7835e98b NP |
1097 | for (i = 1; i < (1 << order); i++) |
1098 | set_page_refcounted(page + i); | |
8dfcc9ba | 1099 | } |
8dfcc9ba | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1103 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1104 | * or two. | |
1105 | */ | |
0a15c3e9 MG |
1106 | static inline |
1107 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1108 | struct zone *zone, int order, gfp_t gfp_flags, |
1109 | int migratetype) | |
1da177e4 LT |
1110 | { |
1111 | unsigned long flags; | |
689bcebf | 1112 | struct page *page; |
1da177e4 | 1113 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1114 | int cpu; |
1da177e4 | 1115 | |
689bcebf | 1116 | again: |
a74609fa | 1117 | cpu = get_cpu(); |
48db57f8 | 1118 | if (likely(order == 0)) { |
1da177e4 LT |
1119 | struct per_cpu_pages *pcp; |
1120 | ||
3dfa5721 | 1121 | pcp = &zone_pcp(zone, cpu)->pcp; |
1da177e4 | 1122 | local_irq_save(flags); |
a74609fa | 1123 | if (!pcp->count) { |
941c7105 | 1124 | pcp->count = rmqueue_bulk(zone, 0, |
e084b2d9 MG |
1125 | pcp->batch, &pcp->list, |
1126 | migratetype, cold); | |
a74609fa NP |
1127 | if (unlikely(!pcp->count)) |
1128 | goto failed; | |
1da177e4 | 1129 | } |
b92a6edd | 1130 | |
535131e6 | 1131 | /* Find a page of the appropriate migrate type */ |
3dfa5721 CL |
1132 | if (cold) { |
1133 | list_for_each_entry_reverse(page, &pcp->list, lru) | |
1134 | if (page_private(page) == migratetype) | |
1135 | break; | |
1136 | } else { | |
1137 | list_for_each_entry(page, &pcp->list, lru) | |
1138 | if (page_private(page) == migratetype) | |
1139 | break; | |
1140 | } | |
535131e6 | 1141 | |
b92a6edd MG |
1142 | /* Allocate more to the pcp list if necessary */ |
1143 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 | 1144 | pcp->count += rmqueue_bulk(zone, 0, |
e084b2d9 MG |
1145 | pcp->batch, &pcp->list, |
1146 | migratetype, cold); | |
535131e6 | 1147 | page = list_entry(pcp->list.next, struct page, lru); |
535131e6 | 1148 | } |
b92a6edd MG |
1149 | |
1150 | list_del(&page->lru); | |
1151 | pcp->count--; | |
7fb1d9fc | 1152 | } else { |
dab48dab AM |
1153 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1154 | /* | |
1155 | * __GFP_NOFAIL is not to be used in new code. | |
1156 | * | |
1157 | * All __GFP_NOFAIL callers should be fixed so that they | |
1158 | * properly detect and handle allocation failures. | |
1159 | * | |
1160 | * We most definitely don't want callers attempting to | |
4923abf9 | 1161 | * allocate greater than order-1 page units with |
dab48dab AM |
1162 | * __GFP_NOFAIL. |
1163 | */ | |
4923abf9 | 1164 | WARN_ON_ONCE(order > 1); |
dab48dab | 1165 | } |
1da177e4 | 1166 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1167 | page = __rmqueue(zone, order, migratetype); |
f2260e6b | 1168 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
a74609fa NP |
1169 | spin_unlock(&zone->lock); |
1170 | if (!page) | |
1171 | goto failed; | |
1da177e4 LT |
1172 | } |
1173 | ||
f8891e5e | 1174 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1175 | zone_statistics(preferred_zone, zone); |
a74609fa NP |
1176 | local_irq_restore(flags); |
1177 | put_cpu(); | |
1da177e4 | 1178 | |
725d704e | 1179 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1180 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1181 | goto again; |
1da177e4 | 1182 | return page; |
a74609fa NP |
1183 | |
1184 | failed: | |
1185 | local_irq_restore(flags); | |
1186 | put_cpu(); | |
1187 | return NULL; | |
1da177e4 LT |
1188 | } |
1189 | ||
41858966 MG |
1190 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
1191 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1192 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1193 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1194 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1195 | ||
1196 | /* Mask to get the watermark bits */ | |
1197 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1198 | ||
3148890b NP |
1199 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
1200 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1201 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1202 | |
933e312e AM |
1203 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1204 | ||
1205 | static struct fail_page_alloc_attr { | |
1206 | struct fault_attr attr; | |
1207 | ||
1208 | u32 ignore_gfp_highmem; | |
1209 | u32 ignore_gfp_wait; | |
54114994 | 1210 | u32 min_order; |
933e312e AM |
1211 | |
1212 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1213 | ||
1214 | struct dentry *ignore_gfp_highmem_file; | |
1215 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1216 | struct dentry *min_order_file; |
933e312e AM |
1217 | |
1218 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1219 | ||
1220 | } fail_page_alloc = { | |
1221 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1222 | .ignore_gfp_wait = 1, |
1223 | .ignore_gfp_highmem = 1, | |
54114994 | 1224 | .min_order = 1, |
933e312e AM |
1225 | }; |
1226 | ||
1227 | static int __init setup_fail_page_alloc(char *str) | |
1228 | { | |
1229 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1230 | } | |
1231 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1232 | ||
1233 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1234 | { | |
54114994 AM |
1235 | if (order < fail_page_alloc.min_order) |
1236 | return 0; | |
933e312e AM |
1237 | if (gfp_mask & __GFP_NOFAIL) |
1238 | return 0; | |
1239 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1240 | return 0; | |
1241 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1242 | return 0; | |
1243 | ||
1244 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1245 | } | |
1246 | ||
1247 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1248 | ||
1249 | static int __init fail_page_alloc_debugfs(void) | |
1250 | { | |
1251 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1252 | struct dentry *dir; | |
1253 | int err; | |
1254 | ||
1255 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1256 | "fail_page_alloc"); | |
1257 | if (err) | |
1258 | return err; | |
1259 | dir = fail_page_alloc.attr.dentries.dir; | |
1260 | ||
1261 | fail_page_alloc.ignore_gfp_wait_file = | |
1262 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1263 | &fail_page_alloc.ignore_gfp_wait); | |
1264 | ||
1265 | fail_page_alloc.ignore_gfp_highmem_file = | |
1266 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1267 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1268 | fail_page_alloc.min_order_file = |
1269 | debugfs_create_u32("min-order", mode, dir, | |
1270 | &fail_page_alloc.min_order); | |
933e312e AM |
1271 | |
1272 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1273 | !fail_page_alloc.ignore_gfp_highmem_file || |
1274 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1275 | err = -ENOMEM; |
1276 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1277 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1278 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1279 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1280 | } | |
1281 | ||
1282 | return err; | |
1283 | } | |
1284 | ||
1285 | late_initcall(fail_page_alloc_debugfs); | |
1286 | ||
1287 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1288 | ||
1289 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1290 | ||
1291 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1292 | { | |
1293 | return 0; | |
1294 | } | |
1295 | ||
1296 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1297 | ||
1da177e4 LT |
1298 | /* |
1299 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1300 | * of the allocation. | |
1301 | */ | |
1302 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1303 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1304 | { |
1305 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1306 | long min = mark; |
1307 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1308 | int o; |
1309 | ||
7fb1d9fc | 1310 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1311 | min -= min / 2; |
7fb1d9fc | 1312 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1313 | min -= min / 4; |
1314 | ||
1315 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1316 | return 0; | |
1317 | for (o = 0; o < order; o++) { | |
1318 | /* At the next order, this order's pages become unavailable */ | |
1319 | free_pages -= z->free_area[o].nr_free << o; | |
1320 | ||
1321 | /* Require fewer higher order pages to be free */ | |
1322 | min >>= 1; | |
1323 | ||
1324 | if (free_pages <= min) | |
1325 | return 0; | |
1326 | } | |
1327 | return 1; | |
1328 | } | |
1329 | ||
9276b1bc PJ |
1330 | #ifdef CONFIG_NUMA |
1331 | /* | |
1332 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1333 | * skip over zones that are not allowed by the cpuset, or that have | |
1334 | * been recently (in last second) found to be nearly full. See further | |
1335 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1336 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1337 | * |
1338 | * If the zonelist cache is present in the passed in zonelist, then | |
1339 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1340 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1341 | * |
1342 | * If the zonelist cache is not available for this zonelist, does | |
1343 | * nothing and returns NULL. | |
1344 | * | |
1345 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1346 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1347 | * | |
1348 | * We hold off even calling zlc_setup, until after we've checked the | |
1349 | * first zone in the zonelist, on the theory that most allocations will | |
1350 | * be satisfied from that first zone, so best to examine that zone as | |
1351 | * quickly as we can. | |
1352 | */ | |
1353 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1354 | { | |
1355 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1356 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1357 | ||
1358 | zlc = zonelist->zlcache_ptr; | |
1359 | if (!zlc) | |
1360 | return NULL; | |
1361 | ||
f05111f5 | 1362 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1363 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1364 | zlc->last_full_zap = jiffies; | |
1365 | } | |
1366 | ||
1367 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1368 | &cpuset_current_mems_allowed : | |
37b07e41 | 1369 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1370 | return allowednodes; |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1375 | * if it is worth looking at further for free memory: | |
1376 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1377 | * bit set in the zonelist_cache fullzones BITMAP). | |
1378 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1379 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1380 | * Return true (non-zero) if zone is worth looking at further, or | |
1381 | * else return false (zero) if it is not. | |
1382 | * | |
1383 | * This check -ignores- the distinction between various watermarks, | |
1384 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1385 | * found to be full for any variation of these watermarks, it will | |
1386 | * be considered full for up to one second by all requests, unless | |
1387 | * we are so low on memory on all allowed nodes that we are forced | |
1388 | * into the second scan of the zonelist. | |
1389 | * | |
1390 | * In the second scan we ignore this zonelist cache and exactly | |
1391 | * apply the watermarks to all zones, even it is slower to do so. | |
1392 | * We are low on memory in the second scan, and should leave no stone | |
1393 | * unturned looking for a free page. | |
1394 | */ | |
dd1a239f | 1395 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1396 | nodemask_t *allowednodes) |
1397 | { | |
1398 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1399 | int i; /* index of *z in zonelist zones */ | |
1400 | int n; /* node that zone *z is on */ | |
1401 | ||
1402 | zlc = zonelist->zlcache_ptr; | |
1403 | if (!zlc) | |
1404 | return 1; | |
1405 | ||
dd1a239f | 1406 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1407 | n = zlc->z_to_n[i]; |
1408 | ||
1409 | /* This zone is worth trying if it is allowed but not full */ | |
1410 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1415 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1416 | * from that zone don't waste time re-examining it. | |
1417 | */ | |
dd1a239f | 1418 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1419 | { |
1420 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1421 | int i; /* index of *z in zonelist zones */ | |
1422 | ||
1423 | zlc = zonelist->zlcache_ptr; | |
1424 | if (!zlc) | |
1425 | return; | |
1426 | ||
dd1a239f | 1427 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1428 | |
1429 | set_bit(i, zlc->fullzones); | |
1430 | } | |
1431 | ||
1432 | #else /* CONFIG_NUMA */ | |
1433 | ||
1434 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1435 | { | |
1436 | return NULL; | |
1437 | } | |
1438 | ||
dd1a239f | 1439 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1440 | nodemask_t *allowednodes) |
1441 | { | |
1442 | return 1; | |
1443 | } | |
1444 | ||
dd1a239f | 1445 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1446 | { |
1447 | } | |
1448 | #endif /* CONFIG_NUMA */ | |
1449 | ||
7fb1d9fc | 1450 | /* |
0798e519 | 1451 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1452 | * a page. |
1453 | */ | |
1454 | static struct page * | |
19770b32 | 1455 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1456 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1457 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1458 | { |
dd1a239f | 1459 | struct zoneref *z; |
7fb1d9fc | 1460 | struct page *page = NULL; |
54a6eb5c | 1461 | int classzone_idx; |
5117f45d | 1462 | struct zone *zone; |
9276b1bc PJ |
1463 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1464 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1465 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1466 | |
19770b32 | 1467 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1468 | zonelist_scan: |
7fb1d9fc | 1469 | /* |
9276b1bc | 1470 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1471 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1472 | */ | |
19770b32 MG |
1473 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1474 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1475 | if (NUMA_BUILD && zlc_active && |
1476 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1477 | continue; | |
7fb1d9fc | 1478 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1479 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1480 | goto try_next_zone; |
7fb1d9fc | 1481 | |
41858966 | 1482 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1483 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1484 | unsigned long mark; |
fa5e084e MG |
1485 | int ret; |
1486 | ||
41858966 | 1487 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1488 | if (zone_watermark_ok(zone, order, mark, |
1489 | classzone_idx, alloc_flags)) | |
1490 | goto try_this_zone; | |
1491 | ||
1492 | if (zone_reclaim_mode == 0) | |
1493 | goto this_zone_full; | |
1494 | ||
1495 | ret = zone_reclaim(zone, gfp_mask, order); | |
1496 | switch (ret) { | |
1497 | case ZONE_RECLAIM_NOSCAN: | |
1498 | /* did not scan */ | |
1499 | goto try_next_zone; | |
1500 | case ZONE_RECLAIM_FULL: | |
1501 | /* scanned but unreclaimable */ | |
1502 | goto this_zone_full; | |
1503 | default: | |
1504 | /* did we reclaim enough */ | |
1505 | if (!zone_watermark_ok(zone, order, mark, | |
1506 | classzone_idx, alloc_flags)) | |
9276b1bc | 1507 | goto this_zone_full; |
0798e519 | 1508 | } |
7fb1d9fc RS |
1509 | } |
1510 | ||
fa5e084e | 1511 | try_this_zone: |
3dd28266 MG |
1512 | page = buffered_rmqueue(preferred_zone, zone, order, |
1513 | gfp_mask, migratetype); | |
0798e519 | 1514 | if (page) |
7fb1d9fc | 1515 | break; |
9276b1bc PJ |
1516 | this_zone_full: |
1517 | if (NUMA_BUILD) | |
1518 | zlc_mark_zone_full(zonelist, z); | |
1519 | try_next_zone: | |
62bc62a8 | 1520 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
d395b734 MG |
1521 | /* |
1522 | * we do zlc_setup after the first zone is tried but only | |
1523 | * if there are multiple nodes make it worthwhile | |
1524 | */ | |
9276b1bc PJ |
1525 | allowednodes = zlc_setup(zonelist, alloc_flags); |
1526 | zlc_active = 1; | |
1527 | did_zlc_setup = 1; | |
1528 | } | |
54a6eb5c | 1529 | } |
9276b1bc PJ |
1530 | |
1531 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1532 | /* Disable zlc cache for second zonelist scan */ | |
1533 | zlc_active = 0; | |
1534 | goto zonelist_scan; | |
1535 | } | |
7fb1d9fc | 1536 | return page; |
753ee728 MH |
1537 | } |
1538 | ||
11e33f6a MG |
1539 | static inline int |
1540 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1541 | unsigned long pages_reclaimed) | |
1da177e4 | 1542 | { |
11e33f6a MG |
1543 | /* Do not loop if specifically requested */ |
1544 | if (gfp_mask & __GFP_NORETRY) | |
1545 | return 0; | |
1da177e4 | 1546 | |
11e33f6a MG |
1547 | /* |
1548 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1549 | * means __GFP_NOFAIL, but that may not be true in other | |
1550 | * implementations. | |
1551 | */ | |
1552 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1553 | return 1; | |
1554 | ||
1555 | /* | |
1556 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1557 | * specified, then we retry until we no longer reclaim any pages | |
1558 | * (above), or we've reclaimed an order of pages at least as | |
1559 | * large as the allocation's order. In both cases, if the | |
1560 | * allocation still fails, we stop retrying. | |
1561 | */ | |
1562 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1563 | return 1; | |
cf40bd16 | 1564 | |
11e33f6a MG |
1565 | /* |
1566 | * Don't let big-order allocations loop unless the caller | |
1567 | * explicitly requests that. | |
1568 | */ | |
1569 | if (gfp_mask & __GFP_NOFAIL) | |
1570 | return 1; | |
1da177e4 | 1571 | |
11e33f6a MG |
1572 | return 0; |
1573 | } | |
933e312e | 1574 | |
11e33f6a MG |
1575 | static inline struct page * |
1576 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1577 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1578 | nodemask_t *nodemask, struct zone *preferred_zone, |
1579 | int migratetype) | |
11e33f6a MG |
1580 | { |
1581 | struct page *page; | |
1582 | ||
1583 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
1584 | if (!try_set_zone_oom(zonelist, gfp_mask)) { | |
1585 | schedule_timeout_uninterruptible(1); | |
1da177e4 LT |
1586 | return NULL; |
1587 | } | |
6b1de916 | 1588 | |
11e33f6a MG |
1589 | /* |
1590 | * Go through the zonelist yet one more time, keep very high watermark | |
1591 | * here, this is only to catch a parallel oom killing, we must fail if | |
1592 | * we're still under heavy pressure. | |
1593 | */ | |
1594 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1595 | order, zonelist, high_zoneidx, | |
5117f45d | 1596 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 1597 | preferred_zone, migratetype); |
7fb1d9fc | 1598 | if (page) |
11e33f6a MG |
1599 | goto out; |
1600 | ||
1601 | /* The OOM killer will not help higher order allocs */ | |
82553a93 | 1602 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL)) |
11e33f6a MG |
1603 | goto out; |
1604 | ||
1605 | /* Exhausted what can be done so it's blamo time */ | |
1606 | out_of_memory(zonelist, gfp_mask, order); | |
1607 | ||
1608 | out: | |
1609 | clear_zonelist_oom(zonelist, gfp_mask); | |
1610 | return page; | |
1611 | } | |
1612 | ||
1613 | /* The really slow allocator path where we enter direct reclaim */ | |
1614 | static inline struct page * | |
1615 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1616 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
5117f45d | 1617 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
3dd28266 | 1618 | int migratetype, unsigned long *did_some_progress) |
11e33f6a MG |
1619 | { |
1620 | struct page *page = NULL; | |
1621 | struct reclaim_state reclaim_state; | |
1622 | struct task_struct *p = current; | |
1623 | ||
1624 | cond_resched(); | |
1625 | ||
1626 | /* We now go into synchronous reclaim */ | |
1627 | cpuset_memory_pressure_bump(); | |
1628 | ||
1629 | /* | |
1630 | * The task's cpuset might have expanded its set of allowable nodes | |
1631 | */ | |
1632 | p->flags |= PF_MEMALLOC; | |
1633 | lockdep_set_current_reclaim_state(gfp_mask); | |
1634 | reclaim_state.reclaimed_slab = 0; | |
1635 | p->reclaim_state = &reclaim_state; | |
1636 | ||
1637 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1638 | ||
1639 | p->reclaim_state = NULL; | |
1640 | lockdep_clear_current_reclaim_state(); | |
1641 | p->flags &= ~PF_MEMALLOC; | |
1642 | ||
1643 | cond_resched(); | |
1644 | ||
1645 | if (order != 0) | |
1646 | drain_all_pages(); | |
1647 | ||
1648 | if (likely(*did_some_progress)) | |
1649 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1650 | zonelist, high_zoneidx, |
3dd28266 MG |
1651 | alloc_flags, preferred_zone, |
1652 | migratetype); | |
11e33f6a MG |
1653 | return page; |
1654 | } | |
1655 | ||
1da177e4 | 1656 | /* |
11e33f6a MG |
1657 | * This is called in the allocator slow-path if the allocation request is of |
1658 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 1659 | */ |
11e33f6a MG |
1660 | static inline struct page * |
1661 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1662 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1663 | nodemask_t *nodemask, struct zone *preferred_zone, |
1664 | int migratetype) | |
11e33f6a MG |
1665 | { |
1666 | struct page *page; | |
1667 | ||
1668 | do { | |
1669 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1670 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 1671 | preferred_zone, migratetype); |
11e33f6a MG |
1672 | |
1673 | if (!page && gfp_mask & __GFP_NOFAIL) | |
8aa7e847 | 1674 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
1675 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
1676 | ||
1677 | return page; | |
1678 | } | |
1679 | ||
1680 | static inline | |
1681 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
1682 | enum zone_type high_zoneidx) | |
1da177e4 | 1683 | { |
dd1a239f MG |
1684 | struct zoneref *z; |
1685 | struct zone *zone; | |
1da177e4 | 1686 | |
11e33f6a MG |
1687 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
1688 | wakeup_kswapd(zone, order); | |
1689 | } | |
cf40bd16 | 1690 | |
341ce06f PZ |
1691 | static inline int |
1692 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
1693 | { | |
1694 | struct task_struct *p = current; | |
1695 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; | |
1696 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 1697 | |
a56f57ff MG |
1698 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
1699 | BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH); | |
933e312e | 1700 | |
341ce06f PZ |
1701 | /* |
1702 | * The caller may dip into page reserves a bit more if the caller | |
1703 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1704 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1705 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1706 | */ | |
a56f57ff | 1707 | alloc_flags |= (gfp_mask & __GFP_HIGH); |
1da177e4 | 1708 | |
341ce06f PZ |
1709 | if (!wait) { |
1710 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 1711 | /* |
341ce06f PZ |
1712 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
1713 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 1714 | */ |
341ce06f PZ |
1715 | alloc_flags &= ~ALLOC_CPUSET; |
1716 | } else if (unlikely(rt_task(p))) | |
1717 | alloc_flags |= ALLOC_HARDER; | |
1718 | ||
1719 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
1720 | if (!in_interrupt() && | |
1721 | ((p->flags & PF_MEMALLOC) || | |
1722 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
1723 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1da177e4 | 1724 | } |
6b1de916 | 1725 | |
341ce06f PZ |
1726 | return alloc_flags; |
1727 | } | |
1728 | ||
11e33f6a MG |
1729 | static inline struct page * |
1730 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
1731 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1732 | nodemask_t *nodemask, struct zone *preferred_zone, |
1733 | int migratetype) | |
11e33f6a MG |
1734 | { |
1735 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1736 | struct page *page = NULL; | |
1737 | int alloc_flags; | |
1738 | unsigned long pages_reclaimed = 0; | |
1739 | unsigned long did_some_progress; | |
1740 | struct task_struct *p = current; | |
1da177e4 | 1741 | |
72807a74 MG |
1742 | /* |
1743 | * In the slowpath, we sanity check order to avoid ever trying to | |
1744 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
1745 | * be using allocators in order of preference for an area that is | |
1746 | * too large. | |
1747 | */ | |
1fc28b70 MG |
1748 | if (order >= MAX_ORDER) { |
1749 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 1750 | return NULL; |
1fc28b70 | 1751 | } |
1da177e4 | 1752 | |
952f3b51 CL |
1753 | /* |
1754 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1755 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1756 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1757 | * using a larger set of nodes after it has established that the | |
1758 | * allowed per node queues are empty and that nodes are | |
1759 | * over allocated. | |
1760 | */ | |
1761 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1762 | goto nopage; | |
1763 | ||
11e33f6a | 1764 | wake_all_kswapd(order, zonelist, high_zoneidx); |
1da177e4 | 1765 | |
9bf2229f | 1766 | /* |
7fb1d9fc RS |
1767 | * OK, we're below the kswapd watermark and have kicked background |
1768 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1769 | * to how we want to proceed. | |
9bf2229f | 1770 | */ |
341ce06f | 1771 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 1772 | |
11e33f6a | 1773 | restart: |
341ce06f | 1774 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 1775 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
1776 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
1777 | preferred_zone, migratetype); | |
7fb1d9fc RS |
1778 | if (page) |
1779 | goto got_pg; | |
1da177e4 | 1780 | |
b43a57bb | 1781 | rebalance: |
11e33f6a | 1782 | /* Allocate without watermarks if the context allows */ |
341ce06f PZ |
1783 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
1784 | page = __alloc_pages_high_priority(gfp_mask, order, | |
1785 | zonelist, high_zoneidx, nodemask, | |
1786 | preferred_zone, migratetype); | |
1787 | if (page) | |
1788 | goto got_pg; | |
1da177e4 LT |
1789 | } |
1790 | ||
1791 | /* Atomic allocations - we can't balance anything */ | |
1792 | if (!wait) | |
1793 | goto nopage; | |
1794 | ||
341ce06f PZ |
1795 | /* Avoid recursion of direct reclaim */ |
1796 | if (p->flags & PF_MEMALLOC) | |
1797 | goto nopage; | |
1798 | ||
6583bb64 DR |
1799 | /* Avoid allocations with no watermarks from looping endlessly */ |
1800 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
1801 | goto nopage; | |
1802 | ||
11e33f6a MG |
1803 | /* Try direct reclaim and then allocating */ |
1804 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
1805 | zonelist, high_zoneidx, | |
1806 | nodemask, | |
5117f45d | 1807 | alloc_flags, preferred_zone, |
3dd28266 | 1808 | migratetype, &did_some_progress); |
11e33f6a MG |
1809 | if (page) |
1810 | goto got_pg; | |
1da177e4 | 1811 | |
e33c3b5e | 1812 | /* |
11e33f6a MG |
1813 | * If we failed to make any progress reclaiming, then we are |
1814 | * running out of options and have to consider going OOM | |
e33c3b5e | 1815 | */ |
11e33f6a MG |
1816 | if (!did_some_progress) { |
1817 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
1818 | if (oom_killer_disabled) |
1819 | goto nopage; | |
11e33f6a MG |
1820 | page = __alloc_pages_may_oom(gfp_mask, order, |
1821 | zonelist, high_zoneidx, | |
3dd28266 MG |
1822 | nodemask, preferred_zone, |
1823 | migratetype); | |
11e33f6a MG |
1824 | if (page) |
1825 | goto got_pg; | |
1da177e4 | 1826 | |
11e33f6a | 1827 | /* |
82553a93 DR |
1828 | * The OOM killer does not trigger for high-order |
1829 | * ~__GFP_NOFAIL allocations so if no progress is being | |
1830 | * made, there are no other options and retrying is | |
1831 | * unlikely to help. | |
11e33f6a | 1832 | */ |
82553a93 DR |
1833 | if (order > PAGE_ALLOC_COSTLY_ORDER && |
1834 | !(gfp_mask & __GFP_NOFAIL)) | |
11e33f6a | 1835 | goto nopage; |
e2c55dc8 | 1836 | |
ff0ceb9d DR |
1837 | goto restart; |
1838 | } | |
1da177e4 LT |
1839 | } |
1840 | ||
11e33f6a | 1841 | /* Check if we should retry the allocation */ |
a41f24ea | 1842 | pages_reclaimed += did_some_progress; |
11e33f6a MG |
1843 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
1844 | /* Wait for some write requests to complete then retry */ | |
8aa7e847 | 1845 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
1846 | goto rebalance; |
1847 | } | |
1848 | ||
1849 | nopage: | |
1850 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1851 | printk(KERN_WARNING "%s: page allocation failure." | |
1852 | " order:%d, mode:0x%x\n", | |
1853 | p->comm, order, gfp_mask); | |
1854 | dump_stack(); | |
578c2fd6 | 1855 | show_mem(); |
1da177e4 | 1856 | } |
b1eeab67 | 1857 | return page; |
1da177e4 | 1858 | got_pg: |
b1eeab67 VN |
1859 | if (kmemcheck_enabled) |
1860 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1da177e4 | 1861 | return page; |
11e33f6a | 1862 | |
1da177e4 | 1863 | } |
11e33f6a MG |
1864 | |
1865 | /* | |
1866 | * This is the 'heart' of the zoned buddy allocator. | |
1867 | */ | |
1868 | struct page * | |
1869 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
1870 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1871 | { | |
1872 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 1873 | struct zone *preferred_zone; |
11e33f6a | 1874 | struct page *page; |
3dd28266 | 1875 | int migratetype = allocflags_to_migratetype(gfp_mask); |
11e33f6a | 1876 | |
dcce284a BH |
1877 | gfp_mask &= gfp_allowed_mask; |
1878 | ||
11e33f6a MG |
1879 | lockdep_trace_alloc(gfp_mask); |
1880 | ||
1881 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1882 | ||
1883 | if (should_fail_alloc_page(gfp_mask, order)) | |
1884 | return NULL; | |
1885 | ||
1886 | /* | |
1887 | * Check the zones suitable for the gfp_mask contain at least one | |
1888 | * valid zone. It's possible to have an empty zonelist as a result | |
1889 | * of GFP_THISNODE and a memoryless node | |
1890 | */ | |
1891 | if (unlikely(!zonelist->_zonerefs->zone)) | |
1892 | return NULL; | |
1893 | ||
5117f45d MG |
1894 | /* The preferred zone is used for statistics later */ |
1895 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); | |
1896 | if (!preferred_zone) | |
1897 | return NULL; | |
1898 | ||
1899 | /* First allocation attempt */ | |
11e33f6a | 1900 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
5117f45d | 1901 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
3dd28266 | 1902 | preferred_zone, migratetype); |
11e33f6a MG |
1903 | if (unlikely(!page)) |
1904 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 1905 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 1906 | preferred_zone, migratetype); |
11e33f6a MG |
1907 | |
1908 | return page; | |
1da177e4 | 1909 | } |
d239171e | 1910 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
1911 | |
1912 | /* | |
1913 | * Common helper functions. | |
1914 | */ | |
920c7a5d | 1915 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1916 | { |
1917 | struct page * page; | |
1918 | page = alloc_pages(gfp_mask, order); | |
1919 | if (!page) | |
1920 | return 0; | |
1921 | return (unsigned long) page_address(page); | |
1922 | } | |
1923 | ||
1924 | EXPORT_SYMBOL(__get_free_pages); | |
1925 | ||
920c7a5d | 1926 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1927 | { |
1928 | struct page * page; | |
1929 | ||
1930 | /* | |
1931 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1932 | * a highmem page | |
1933 | */ | |
725d704e | 1934 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1935 | |
1936 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1937 | if (page) | |
1938 | return (unsigned long) page_address(page); | |
1939 | return 0; | |
1940 | } | |
1941 | ||
1942 | EXPORT_SYMBOL(get_zeroed_page); | |
1943 | ||
1944 | void __pagevec_free(struct pagevec *pvec) | |
1945 | { | |
1946 | int i = pagevec_count(pvec); | |
1947 | ||
1948 | while (--i >= 0) | |
1949 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1950 | } | |
1951 | ||
920c7a5d | 1952 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 1953 | { |
b5810039 | 1954 | if (put_page_testzero(page)) { |
1da177e4 LT |
1955 | if (order == 0) |
1956 | free_hot_page(page); | |
1957 | else | |
1958 | __free_pages_ok(page, order); | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | EXPORT_SYMBOL(__free_pages); | |
1963 | ||
920c7a5d | 1964 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
1965 | { |
1966 | if (addr != 0) { | |
725d704e | 1967 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1968 | __free_pages(virt_to_page((void *)addr), order); |
1969 | } | |
1970 | } | |
1971 | ||
1972 | EXPORT_SYMBOL(free_pages); | |
1973 | ||
2be0ffe2 TT |
1974 | /** |
1975 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
1976 | * @size: the number of bytes to allocate | |
1977 | * @gfp_mask: GFP flags for the allocation | |
1978 | * | |
1979 | * This function is similar to alloc_pages(), except that it allocates the | |
1980 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
1981 | * allocate memory in power-of-two pages. | |
1982 | * | |
1983 | * This function is also limited by MAX_ORDER. | |
1984 | * | |
1985 | * Memory allocated by this function must be released by free_pages_exact(). | |
1986 | */ | |
1987 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
1988 | { | |
1989 | unsigned int order = get_order(size); | |
1990 | unsigned long addr; | |
1991 | ||
1992 | addr = __get_free_pages(gfp_mask, order); | |
1993 | if (addr) { | |
1994 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
1995 | unsigned long used = addr + PAGE_ALIGN(size); | |
1996 | ||
5bfd7560 | 1997 | split_page(virt_to_page((void *)addr), order); |
2be0ffe2 TT |
1998 | while (used < alloc_end) { |
1999 | free_page(used); | |
2000 | used += PAGE_SIZE; | |
2001 | } | |
2002 | } | |
2003 | ||
2004 | return (void *)addr; | |
2005 | } | |
2006 | EXPORT_SYMBOL(alloc_pages_exact); | |
2007 | ||
2008 | /** | |
2009 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2010 | * @virt: the value returned by alloc_pages_exact. | |
2011 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2012 | * | |
2013 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2014 | */ | |
2015 | void free_pages_exact(void *virt, size_t size) | |
2016 | { | |
2017 | unsigned long addr = (unsigned long)virt; | |
2018 | unsigned long end = addr + PAGE_ALIGN(size); | |
2019 | ||
2020 | while (addr < end) { | |
2021 | free_page(addr); | |
2022 | addr += PAGE_SIZE; | |
2023 | } | |
2024 | } | |
2025 | EXPORT_SYMBOL(free_pages_exact); | |
2026 | ||
1da177e4 LT |
2027 | static unsigned int nr_free_zone_pages(int offset) |
2028 | { | |
dd1a239f | 2029 | struct zoneref *z; |
54a6eb5c MG |
2030 | struct zone *zone; |
2031 | ||
e310fd43 | 2032 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2033 | unsigned int sum = 0; |
2034 | ||
0e88460d | 2035 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2036 | |
54a6eb5c | 2037 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2038 | unsigned long size = zone->present_pages; |
41858966 | 2039 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2040 | if (size > high) |
2041 | sum += size - high; | |
1da177e4 LT |
2042 | } |
2043 | ||
2044 | return sum; | |
2045 | } | |
2046 | ||
2047 | /* | |
2048 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2049 | */ | |
2050 | unsigned int nr_free_buffer_pages(void) | |
2051 | { | |
af4ca457 | 2052 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2053 | } |
c2f1a551 | 2054 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2055 | |
2056 | /* | |
2057 | * Amount of free RAM allocatable within all zones | |
2058 | */ | |
2059 | unsigned int nr_free_pagecache_pages(void) | |
2060 | { | |
2a1e274a | 2061 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2062 | } |
08e0f6a9 CL |
2063 | |
2064 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2065 | { |
08e0f6a9 | 2066 | if (NUMA_BUILD) |
25ba77c1 | 2067 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2068 | } |
1da177e4 | 2069 | |
1da177e4 LT |
2070 | void si_meminfo(struct sysinfo *val) |
2071 | { | |
2072 | val->totalram = totalram_pages; | |
2073 | val->sharedram = 0; | |
d23ad423 | 2074 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2075 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2076 | val->totalhigh = totalhigh_pages; |
2077 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2078 | val->mem_unit = PAGE_SIZE; |
2079 | } | |
2080 | ||
2081 | EXPORT_SYMBOL(si_meminfo); | |
2082 | ||
2083 | #ifdef CONFIG_NUMA | |
2084 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2085 | { | |
2086 | pg_data_t *pgdat = NODE_DATA(nid); | |
2087 | ||
2088 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2089 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2090 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2091 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2092 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2093 | NR_FREE_PAGES); | |
98d2b0eb CL |
2094 | #else |
2095 | val->totalhigh = 0; | |
2096 | val->freehigh = 0; | |
2097 | #endif | |
1da177e4 LT |
2098 | val->mem_unit = PAGE_SIZE; |
2099 | } | |
2100 | #endif | |
2101 | ||
2102 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
2103 | ||
2104 | /* | |
2105 | * Show free area list (used inside shift_scroll-lock stuff) | |
2106 | * We also calculate the percentage fragmentation. We do this by counting the | |
2107 | * memory on each free list with the exception of the first item on the list. | |
2108 | */ | |
2109 | void show_free_areas(void) | |
2110 | { | |
c7241913 | 2111 | int cpu; |
1da177e4 LT |
2112 | struct zone *zone; |
2113 | ||
ee99c71c | 2114 | for_each_populated_zone(zone) { |
c7241913 JS |
2115 | show_node(zone); |
2116 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2117 | |
6b482c67 | 2118 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2119 | struct per_cpu_pageset *pageset; |
2120 | ||
e7c8d5c9 | 2121 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 2122 | |
3dfa5721 CL |
2123 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2124 | cpu, pageset->pcp.high, | |
2125 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2126 | } |
2127 | } | |
2128 | ||
7b854121 LS |
2129 | printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n" |
2130 | " inactive_file:%lu" | |
7b854121 | 2131 | " unevictable:%lu" |
7b854121 | 2132 | " dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 2133 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe RR |
2134 | global_page_state(NR_ACTIVE_ANON), |
2135 | global_page_state(NR_ACTIVE_FILE), | |
2136 | global_page_state(NR_INACTIVE_ANON), | |
2137 | global_page_state(NR_INACTIVE_FILE), | |
7b854121 | 2138 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2139 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2140 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2141 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2142 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
2143 | global_page_state(NR_SLAB_RECLAIMABLE) + |
2144 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2145 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
2146 | global_page_state(NR_PAGETABLE), |
2147 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2148 | |
ee99c71c | 2149 | for_each_populated_zone(zone) { |
1da177e4 LT |
2150 | int i; |
2151 | ||
2152 | show_node(zone); | |
2153 | printk("%s" | |
2154 | " free:%lukB" | |
2155 | " min:%lukB" | |
2156 | " low:%lukB" | |
2157 | " high:%lukB" | |
4f98a2fe RR |
2158 | " active_anon:%lukB" |
2159 | " inactive_anon:%lukB" | |
2160 | " active_file:%lukB" | |
2161 | " inactive_file:%lukB" | |
7b854121 | 2162 | " unevictable:%lukB" |
1da177e4 LT |
2163 | " present:%lukB" |
2164 | " pages_scanned:%lu" | |
2165 | " all_unreclaimable? %s" | |
2166 | "\n", | |
2167 | zone->name, | |
d23ad423 | 2168 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
2169 | K(min_wmark_pages(zone)), |
2170 | K(low_wmark_pages(zone)), | |
2171 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
2172 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2173 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2174 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2175 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 2176 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
1da177e4 LT |
2177 | K(zone->present_pages), |
2178 | zone->pages_scanned, | |
e815af95 | 2179 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
2180 | ); |
2181 | printk("lowmem_reserve[]:"); | |
2182 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2183 | printk(" %lu", zone->lowmem_reserve[i]); | |
2184 | printk("\n"); | |
2185 | } | |
2186 | ||
ee99c71c | 2187 | for_each_populated_zone(zone) { |
8f9de51a | 2188 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
2189 | |
2190 | show_node(zone); | |
2191 | printk("%s: ", zone->name); | |
1da177e4 LT |
2192 | |
2193 | spin_lock_irqsave(&zone->lock, flags); | |
2194 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2195 | nr[order] = zone->free_area[order].nr_free; |
2196 | total += nr[order] << order; | |
1da177e4 LT |
2197 | } |
2198 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2199 | for (order = 0; order < MAX_ORDER; order++) |
2200 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2201 | printk("= %lukB\n", K(total)); |
2202 | } | |
2203 | ||
e6f3602d LW |
2204 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2205 | ||
1da177e4 LT |
2206 | show_swap_cache_info(); |
2207 | } | |
2208 | ||
19770b32 MG |
2209 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2210 | { | |
2211 | zoneref->zone = zone; | |
2212 | zoneref->zone_idx = zone_idx(zone); | |
2213 | } | |
2214 | ||
1da177e4 LT |
2215 | /* |
2216 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2217 | * |
2218 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2219 | */ |
f0c0b2b8 KH |
2220 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2221 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2222 | { |
1a93205b CL |
2223 | struct zone *zone; |
2224 | ||
98d2b0eb | 2225 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2226 | zone_type++; |
02a68a5e CL |
2227 | |
2228 | do { | |
2f6726e5 | 2229 | zone_type--; |
070f8032 | 2230 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2231 | if (populated_zone(zone)) { |
dd1a239f MG |
2232 | zoneref_set_zone(zone, |
2233 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2234 | check_highest_zone(zone_type); |
1da177e4 | 2235 | } |
02a68a5e | 2236 | |
2f6726e5 | 2237 | } while (zone_type); |
070f8032 | 2238 | return nr_zones; |
1da177e4 LT |
2239 | } |
2240 | ||
f0c0b2b8 KH |
2241 | |
2242 | /* | |
2243 | * zonelist_order: | |
2244 | * 0 = automatic detection of better ordering. | |
2245 | * 1 = order by ([node] distance, -zonetype) | |
2246 | * 2 = order by (-zonetype, [node] distance) | |
2247 | * | |
2248 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2249 | * the same zonelist. So only NUMA can configure this param. | |
2250 | */ | |
2251 | #define ZONELIST_ORDER_DEFAULT 0 | |
2252 | #define ZONELIST_ORDER_NODE 1 | |
2253 | #define ZONELIST_ORDER_ZONE 2 | |
2254 | ||
2255 | /* zonelist order in the kernel. | |
2256 | * set_zonelist_order() will set this to NODE or ZONE. | |
2257 | */ | |
2258 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2259 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2260 | ||
2261 | ||
1da177e4 | 2262 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2263 | /* The value user specified ....changed by config */ |
2264 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2265 | /* string for sysctl */ | |
2266 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2267 | char numa_zonelist_order[16] = "default"; | |
2268 | ||
2269 | /* | |
2270 | * interface for configure zonelist ordering. | |
2271 | * command line option "numa_zonelist_order" | |
2272 | * = "[dD]efault - default, automatic configuration. | |
2273 | * = "[nN]ode - order by node locality, then by zone within node | |
2274 | * = "[zZ]one - order by zone, then by locality within zone | |
2275 | */ | |
2276 | ||
2277 | static int __parse_numa_zonelist_order(char *s) | |
2278 | { | |
2279 | if (*s == 'd' || *s == 'D') { | |
2280 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2281 | } else if (*s == 'n' || *s == 'N') { | |
2282 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2283 | } else if (*s == 'z' || *s == 'Z') { | |
2284 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2285 | } else { | |
2286 | printk(KERN_WARNING | |
2287 | "Ignoring invalid numa_zonelist_order value: " | |
2288 | "%s\n", s); | |
2289 | return -EINVAL; | |
2290 | } | |
2291 | return 0; | |
2292 | } | |
2293 | ||
2294 | static __init int setup_numa_zonelist_order(char *s) | |
2295 | { | |
2296 | if (s) | |
2297 | return __parse_numa_zonelist_order(s); | |
2298 | return 0; | |
2299 | } | |
2300 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2301 | ||
2302 | /* | |
2303 | * sysctl handler for numa_zonelist_order | |
2304 | */ | |
2305 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
2306 | struct file *file, void __user *buffer, size_t *length, | |
2307 | loff_t *ppos) | |
2308 | { | |
2309 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2310 | int ret; | |
2311 | ||
2312 | if (write) | |
2313 | strncpy(saved_string, (char*)table->data, | |
2314 | NUMA_ZONELIST_ORDER_LEN); | |
2315 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
2316 | if (ret) | |
2317 | return ret; | |
2318 | if (write) { | |
2319 | int oldval = user_zonelist_order; | |
2320 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2321 | /* | |
2322 | * bogus value. restore saved string | |
2323 | */ | |
2324 | strncpy((char*)table->data, saved_string, | |
2325 | NUMA_ZONELIST_ORDER_LEN); | |
2326 | user_zonelist_order = oldval; | |
2327 | } else if (oldval != user_zonelist_order) | |
2328 | build_all_zonelists(); | |
2329 | } | |
2330 | return 0; | |
2331 | } | |
2332 | ||
2333 | ||
62bc62a8 | 2334 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
2335 | static int node_load[MAX_NUMNODES]; |
2336 | ||
1da177e4 | 2337 | /** |
4dc3b16b | 2338 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2339 | * @node: node whose fallback list we're appending |
2340 | * @used_node_mask: nodemask_t of already used nodes | |
2341 | * | |
2342 | * We use a number of factors to determine which is the next node that should | |
2343 | * appear on a given node's fallback list. The node should not have appeared | |
2344 | * already in @node's fallback list, and it should be the next closest node | |
2345 | * according to the distance array (which contains arbitrary distance values | |
2346 | * from each node to each node in the system), and should also prefer nodes | |
2347 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2348 | * on them otherwise. | |
2349 | * It returns -1 if no node is found. | |
2350 | */ | |
f0c0b2b8 | 2351 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2352 | { |
4cf808eb | 2353 | int n, val; |
1da177e4 LT |
2354 | int min_val = INT_MAX; |
2355 | int best_node = -1; | |
a70f7302 | 2356 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2357 | |
4cf808eb LT |
2358 | /* Use the local node if we haven't already */ |
2359 | if (!node_isset(node, *used_node_mask)) { | |
2360 | node_set(node, *used_node_mask); | |
2361 | return node; | |
2362 | } | |
1da177e4 | 2363 | |
37b07e41 | 2364 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2365 | |
2366 | /* Don't want a node to appear more than once */ | |
2367 | if (node_isset(n, *used_node_mask)) | |
2368 | continue; | |
2369 | ||
1da177e4 LT |
2370 | /* Use the distance array to find the distance */ |
2371 | val = node_distance(node, n); | |
2372 | ||
4cf808eb LT |
2373 | /* Penalize nodes under us ("prefer the next node") */ |
2374 | val += (n < node); | |
2375 | ||
1da177e4 | 2376 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2377 | tmp = cpumask_of_node(n); |
2378 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2379 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2380 | ||
2381 | /* Slight preference for less loaded node */ | |
2382 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2383 | val += node_load[n]; | |
2384 | ||
2385 | if (val < min_val) { | |
2386 | min_val = val; | |
2387 | best_node = n; | |
2388 | } | |
2389 | } | |
2390 | ||
2391 | if (best_node >= 0) | |
2392 | node_set(best_node, *used_node_mask); | |
2393 | ||
2394 | return best_node; | |
2395 | } | |
2396 | ||
f0c0b2b8 KH |
2397 | |
2398 | /* | |
2399 | * Build zonelists ordered by node and zones within node. | |
2400 | * This results in maximum locality--normal zone overflows into local | |
2401 | * DMA zone, if any--but risks exhausting DMA zone. | |
2402 | */ | |
2403 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2404 | { |
f0c0b2b8 | 2405 | int j; |
1da177e4 | 2406 | struct zonelist *zonelist; |
f0c0b2b8 | 2407 | |
54a6eb5c | 2408 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2409 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2410 | ; |
2411 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2412 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2413 | zonelist->_zonerefs[j].zone = NULL; |
2414 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2415 | } |
2416 | ||
523b9458 CL |
2417 | /* |
2418 | * Build gfp_thisnode zonelists | |
2419 | */ | |
2420 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2421 | { | |
523b9458 CL |
2422 | int j; |
2423 | struct zonelist *zonelist; | |
2424 | ||
54a6eb5c MG |
2425 | zonelist = &pgdat->node_zonelists[1]; |
2426 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2427 | zonelist->_zonerefs[j].zone = NULL; |
2428 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2429 | } |
2430 | ||
f0c0b2b8 KH |
2431 | /* |
2432 | * Build zonelists ordered by zone and nodes within zones. | |
2433 | * This results in conserving DMA zone[s] until all Normal memory is | |
2434 | * exhausted, but results in overflowing to remote node while memory | |
2435 | * may still exist in local DMA zone. | |
2436 | */ | |
2437 | static int node_order[MAX_NUMNODES]; | |
2438 | ||
2439 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2440 | { | |
f0c0b2b8 KH |
2441 | int pos, j, node; |
2442 | int zone_type; /* needs to be signed */ | |
2443 | struct zone *z; | |
2444 | struct zonelist *zonelist; | |
2445 | ||
54a6eb5c MG |
2446 | zonelist = &pgdat->node_zonelists[0]; |
2447 | pos = 0; | |
2448 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2449 | for (j = 0; j < nr_nodes; j++) { | |
2450 | node = node_order[j]; | |
2451 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2452 | if (populated_zone(z)) { | |
dd1a239f MG |
2453 | zoneref_set_zone(z, |
2454 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2455 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2456 | } |
2457 | } | |
f0c0b2b8 | 2458 | } |
dd1a239f MG |
2459 | zonelist->_zonerefs[pos].zone = NULL; |
2460 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2461 | } |
2462 | ||
2463 | static int default_zonelist_order(void) | |
2464 | { | |
2465 | int nid, zone_type; | |
2466 | unsigned long low_kmem_size,total_size; | |
2467 | struct zone *z; | |
2468 | int average_size; | |
2469 | /* | |
2470 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2471 | * If they are really small and used heavily, the system can fall | |
2472 | * into OOM very easily. | |
2473 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2474 | */ | |
2475 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2476 | low_kmem_size = 0; | |
2477 | total_size = 0; | |
2478 | for_each_online_node(nid) { | |
2479 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2480 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2481 | if (populated_zone(z)) { | |
2482 | if (zone_type < ZONE_NORMAL) | |
2483 | low_kmem_size += z->present_pages; | |
2484 | total_size += z->present_pages; | |
2485 | } | |
2486 | } | |
2487 | } | |
2488 | if (!low_kmem_size || /* there are no DMA area. */ | |
2489 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2490 | return ZONELIST_ORDER_NODE; | |
2491 | /* | |
2492 | * look into each node's config. | |
2493 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2494 | * local memory, NODE_ORDER may be suitable. | |
2495 | */ | |
37b07e41 LS |
2496 | average_size = total_size / |
2497 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2498 | for_each_online_node(nid) { |
2499 | low_kmem_size = 0; | |
2500 | total_size = 0; | |
2501 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2502 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2503 | if (populated_zone(z)) { | |
2504 | if (zone_type < ZONE_NORMAL) | |
2505 | low_kmem_size += z->present_pages; | |
2506 | total_size += z->present_pages; | |
2507 | } | |
2508 | } | |
2509 | if (low_kmem_size && | |
2510 | total_size > average_size && /* ignore small node */ | |
2511 | low_kmem_size > total_size * 70/100) | |
2512 | return ZONELIST_ORDER_NODE; | |
2513 | } | |
2514 | return ZONELIST_ORDER_ZONE; | |
2515 | } | |
2516 | ||
2517 | static void set_zonelist_order(void) | |
2518 | { | |
2519 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2520 | current_zonelist_order = default_zonelist_order(); | |
2521 | else | |
2522 | current_zonelist_order = user_zonelist_order; | |
2523 | } | |
2524 | ||
2525 | static void build_zonelists(pg_data_t *pgdat) | |
2526 | { | |
2527 | int j, node, load; | |
2528 | enum zone_type i; | |
1da177e4 | 2529 | nodemask_t used_mask; |
f0c0b2b8 KH |
2530 | int local_node, prev_node; |
2531 | struct zonelist *zonelist; | |
2532 | int order = current_zonelist_order; | |
1da177e4 LT |
2533 | |
2534 | /* initialize zonelists */ | |
523b9458 | 2535 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2536 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2537 | zonelist->_zonerefs[0].zone = NULL; |
2538 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2539 | } |
2540 | ||
2541 | /* NUMA-aware ordering of nodes */ | |
2542 | local_node = pgdat->node_id; | |
62bc62a8 | 2543 | load = nr_online_nodes; |
1da177e4 LT |
2544 | prev_node = local_node; |
2545 | nodes_clear(used_mask); | |
f0c0b2b8 | 2546 | |
f0c0b2b8 KH |
2547 | memset(node_order, 0, sizeof(node_order)); |
2548 | j = 0; | |
2549 | ||
1da177e4 | 2550 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2551 | int distance = node_distance(local_node, node); |
2552 | ||
2553 | /* | |
2554 | * If another node is sufficiently far away then it is better | |
2555 | * to reclaim pages in a zone before going off node. | |
2556 | */ | |
2557 | if (distance > RECLAIM_DISTANCE) | |
2558 | zone_reclaim_mode = 1; | |
2559 | ||
1da177e4 LT |
2560 | /* |
2561 | * We don't want to pressure a particular node. | |
2562 | * So adding penalty to the first node in same | |
2563 | * distance group to make it round-robin. | |
2564 | */ | |
9eeff239 | 2565 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2566 | node_load[node] = load; |
2567 | ||
1da177e4 LT |
2568 | prev_node = node; |
2569 | load--; | |
f0c0b2b8 KH |
2570 | if (order == ZONELIST_ORDER_NODE) |
2571 | build_zonelists_in_node_order(pgdat, node); | |
2572 | else | |
2573 | node_order[j++] = node; /* remember order */ | |
2574 | } | |
1da177e4 | 2575 | |
f0c0b2b8 KH |
2576 | if (order == ZONELIST_ORDER_ZONE) { |
2577 | /* calculate node order -- i.e., DMA last! */ | |
2578 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2579 | } |
523b9458 CL |
2580 | |
2581 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2582 | } |
2583 | ||
9276b1bc | 2584 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2585 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2586 | { |
54a6eb5c MG |
2587 | struct zonelist *zonelist; |
2588 | struct zonelist_cache *zlc; | |
dd1a239f | 2589 | struct zoneref *z; |
9276b1bc | 2590 | |
54a6eb5c MG |
2591 | zonelist = &pgdat->node_zonelists[0]; |
2592 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2593 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2594 | for (z = zonelist->_zonerefs; z->zone; z++) |
2595 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2596 | } |
2597 | ||
f0c0b2b8 | 2598 | |
1da177e4 LT |
2599 | #else /* CONFIG_NUMA */ |
2600 | ||
f0c0b2b8 KH |
2601 | static void set_zonelist_order(void) |
2602 | { | |
2603 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2604 | } | |
2605 | ||
2606 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2607 | { |
19655d34 | 2608 | int node, local_node; |
54a6eb5c MG |
2609 | enum zone_type j; |
2610 | struct zonelist *zonelist; | |
1da177e4 LT |
2611 | |
2612 | local_node = pgdat->node_id; | |
1da177e4 | 2613 | |
54a6eb5c MG |
2614 | zonelist = &pgdat->node_zonelists[0]; |
2615 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2616 | |
54a6eb5c MG |
2617 | /* |
2618 | * Now we build the zonelist so that it contains the zones | |
2619 | * of all the other nodes. | |
2620 | * We don't want to pressure a particular node, so when | |
2621 | * building the zones for node N, we make sure that the | |
2622 | * zones coming right after the local ones are those from | |
2623 | * node N+1 (modulo N) | |
2624 | */ | |
2625 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2626 | if (!node_online(node)) | |
2627 | continue; | |
2628 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2629 | MAX_NR_ZONES - 1); | |
1da177e4 | 2630 | } |
54a6eb5c MG |
2631 | for (node = 0; node < local_node; node++) { |
2632 | if (!node_online(node)) | |
2633 | continue; | |
2634 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2635 | MAX_NR_ZONES - 1); | |
2636 | } | |
2637 | ||
dd1a239f MG |
2638 | zonelist->_zonerefs[j].zone = NULL; |
2639 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
2640 | } |
2641 | ||
9276b1bc | 2642 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2643 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2644 | { |
54a6eb5c | 2645 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
2646 | } |
2647 | ||
1da177e4 LT |
2648 | #endif /* CONFIG_NUMA */ |
2649 | ||
9b1a4d38 | 2650 | /* return values int ....just for stop_machine() */ |
f0c0b2b8 | 2651 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2652 | { |
6811378e | 2653 | int nid; |
9276b1bc | 2654 | |
7f9cfb31 BL |
2655 | #ifdef CONFIG_NUMA |
2656 | memset(node_load, 0, sizeof(node_load)); | |
2657 | #endif | |
9276b1bc | 2658 | for_each_online_node(nid) { |
7ea1530a CL |
2659 | pg_data_t *pgdat = NODE_DATA(nid); |
2660 | ||
2661 | build_zonelists(pgdat); | |
2662 | build_zonelist_cache(pgdat); | |
9276b1bc | 2663 | } |
6811378e YG |
2664 | return 0; |
2665 | } | |
2666 | ||
f0c0b2b8 | 2667 | void build_all_zonelists(void) |
6811378e | 2668 | { |
f0c0b2b8 KH |
2669 | set_zonelist_order(); |
2670 | ||
6811378e | 2671 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2672 | __build_all_zonelists(NULL); |
68ad8df4 | 2673 | mminit_verify_zonelist(); |
6811378e YG |
2674 | cpuset_init_current_mems_allowed(); |
2675 | } else { | |
183ff22b | 2676 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 2677 | of zonelist */ |
9b1a4d38 | 2678 | stop_machine(__build_all_zonelists, NULL, NULL); |
6811378e YG |
2679 | /* cpuset refresh routine should be here */ |
2680 | } | |
bd1e22b8 | 2681 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2682 | /* |
2683 | * Disable grouping by mobility if the number of pages in the | |
2684 | * system is too low to allow the mechanism to work. It would be | |
2685 | * more accurate, but expensive to check per-zone. This check is | |
2686 | * made on memory-hotadd so a system can start with mobility | |
2687 | * disabled and enable it later | |
2688 | */ | |
d9c23400 | 2689 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2690 | page_group_by_mobility_disabled = 1; |
2691 | else | |
2692 | page_group_by_mobility_disabled = 0; | |
2693 | ||
2694 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2695 | "Total pages: %ld\n", | |
62bc62a8 | 2696 | nr_online_nodes, |
f0c0b2b8 | 2697 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 2698 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2699 | vm_total_pages); |
2700 | #ifdef CONFIG_NUMA | |
2701 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2702 | #endif | |
1da177e4 LT |
2703 | } |
2704 | ||
2705 | /* | |
2706 | * Helper functions to size the waitqueue hash table. | |
2707 | * Essentially these want to choose hash table sizes sufficiently | |
2708 | * large so that collisions trying to wait on pages are rare. | |
2709 | * But in fact, the number of active page waitqueues on typical | |
2710 | * systems is ridiculously low, less than 200. So this is even | |
2711 | * conservative, even though it seems large. | |
2712 | * | |
2713 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2714 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2715 | */ | |
2716 | #define PAGES_PER_WAITQUEUE 256 | |
2717 | ||
cca448fe | 2718 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2719 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2720 | { |
2721 | unsigned long size = 1; | |
2722 | ||
2723 | pages /= PAGES_PER_WAITQUEUE; | |
2724 | ||
2725 | while (size < pages) | |
2726 | size <<= 1; | |
2727 | ||
2728 | /* | |
2729 | * Once we have dozens or even hundreds of threads sleeping | |
2730 | * on IO we've got bigger problems than wait queue collision. | |
2731 | * Limit the size of the wait table to a reasonable size. | |
2732 | */ | |
2733 | size = min(size, 4096UL); | |
2734 | ||
2735 | return max(size, 4UL); | |
2736 | } | |
cca448fe YG |
2737 | #else |
2738 | /* | |
2739 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2740 | * a suitable size for its wait_table. So we use the maximum size now. | |
2741 | * | |
2742 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2743 | * | |
2744 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2745 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2746 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2747 | * | |
2748 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2749 | * or more by the traditional way. (See above). It equals: | |
2750 | * | |
2751 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2752 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2753 | * powerpc (64K page size) : = (32G +16M)byte. | |
2754 | */ | |
2755 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2756 | { | |
2757 | return 4096UL; | |
2758 | } | |
2759 | #endif | |
1da177e4 LT |
2760 | |
2761 | /* | |
2762 | * This is an integer logarithm so that shifts can be used later | |
2763 | * to extract the more random high bits from the multiplicative | |
2764 | * hash function before the remainder is taken. | |
2765 | */ | |
2766 | static inline unsigned long wait_table_bits(unsigned long size) | |
2767 | { | |
2768 | return ffz(~size); | |
2769 | } | |
2770 | ||
2771 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2772 | ||
56fd56b8 | 2773 | /* |
d9c23400 | 2774 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
2775 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
2776 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
2777 | * higher will lead to a bigger reserve which will get freed as contiguous |
2778 | * blocks as reclaim kicks in | |
2779 | */ | |
2780 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2781 | { | |
2782 | unsigned long start_pfn, pfn, end_pfn; | |
2783 | struct page *page; | |
2784 | unsigned long reserve, block_migratetype; | |
2785 | ||
2786 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2787 | start_pfn = zone->zone_start_pfn; | |
2788 | end_pfn = start_pfn + zone->spanned_pages; | |
41858966 | 2789 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 2790 | pageblock_order; |
56fd56b8 | 2791 | |
d9c23400 | 2792 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2793 | if (!pfn_valid(pfn)) |
2794 | continue; | |
2795 | page = pfn_to_page(pfn); | |
2796 | ||
344c790e AL |
2797 | /* Watch out for overlapping nodes */ |
2798 | if (page_to_nid(page) != zone_to_nid(zone)) | |
2799 | continue; | |
2800 | ||
56fd56b8 MG |
2801 | /* Blocks with reserved pages will never free, skip them. */ |
2802 | if (PageReserved(page)) | |
2803 | continue; | |
2804 | ||
2805 | block_migratetype = get_pageblock_migratetype(page); | |
2806 | ||
2807 | /* If this block is reserved, account for it */ | |
2808 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2809 | reserve--; | |
2810 | continue; | |
2811 | } | |
2812 | ||
2813 | /* Suitable for reserving if this block is movable */ | |
2814 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2815 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2816 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2817 | reserve--; | |
2818 | continue; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * If the reserve is met and this is a previous reserved block, | |
2823 | * take it back | |
2824 | */ | |
2825 | if (block_migratetype == MIGRATE_RESERVE) { | |
2826 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2827 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2828 | } | |
2829 | } | |
2830 | } | |
ac0e5b7a | 2831 | |
1da177e4 LT |
2832 | /* |
2833 | * Initially all pages are reserved - free ones are freed | |
2834 | * up by free_all_bootmem() once the early boot process is | |
2835 | * done. Non-atomic initialization, single-pass. | |
2836 | */ | |
c09b4240 | 2837 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2838 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2839 | { |
1da177e4 | 2840 | struct page *page; |
29751f69 AW |
2841 | unsigned long end_pfn = start_pfn + size; |
2842 | unsigned long pfn; | |
86051ca5 | 2843 | struct zone *z; |
1da177e4 | 2844 | |
22b31eec HD |
2845 | if (highest_memmap_pfn < end_pfn - 1) |
2846 | highest_memmap_pfn = end_pfn - 1; | |
2847 | ||
86051ca5 | 2848 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 2849 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2850 | /* |
2851 | * There can be holes in boot-time mem_map[]s | |
2852 | * handed to this function. They do not | |
2853 | * exist on hotplugged memory. | |
2854 | */ | |
2855 | if (context == MEMMAP_EARLY) { | |
2856 | if (!early_pfn_valid(pfn)) | |
2857 | continue; | |
2858 | if (!early_pfn_in_nid(pfn, nid)) | |
2859 | continue; | |
2860 | } | |
d41dee36 AW |
2861 | page = pfn_to_page(pfn); |
2862 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 2863 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 2864 | init_page_count(page); |
1da177e4 LT |
2865 | reset_page_mapcount(page); |
2866 | SetPageReserved(page); | |
b2a0ac88 MG |
2867 | /* |
2868 | * Mark the block movable so that blocks are reserved for | |
2869 | * movable at startup. This will force kernel allocations | |
2870 | * to reserve their blocks rather than leaking throughout | |
2871 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2872 | * kernel allocations are made. Later some blocks near |
2873 | * the start are marked MIGRATE_RESERVE by | |
2874 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
2875 | * |
2876 | * bitmap is created for zone's valid pfn range. but memmap | |
2877 | * can be created for invalid pages (for alignment) | |
2878 | * check here not to call set_pageblock_migratetype() against | |
2879 | * pfn out of zone. | |
b2a0ac88 | 2880 | */ |
86051ca5 KH |
2881 | if ((z->zone_start_pfn <= pfn) |
2882 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
2883 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 2884 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2885 | |
1da177e4 LT |
2886 | INIT_LIST_HEAD(&page->lru); |
2887 | #ifdef WANT_PAGE_VIRTUAL | |
2888 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2889 | if (!is_highmem_idx(zone)) | |
3212c6be | 2890 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2891 | #endif |
1da177e4 LT |
2892 | } |
2893 | } | |
2894 | ||
1e548deb | 2895 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 2896 | { |
b2a0ac88 MG |
2897 | int order, t; |
2898 | for_each_migratetype_order(order, t) { | |
2899 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2900 | zone->free_area[order].nr_free = 0; |
2901 | } | |
2902 | } | |
2903 | ||
2904 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2905 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2906 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2907 | #endif |
2908 | ||
1d6f4e60 | 2909 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 2910 | { |
3a6be87f | 2911 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
2912 | int batch; |
2913 | ||
2914 | /* | |
2915 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2916 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2917 | * |
2918 | * OK, so we don't know how big the cache is. So guess. | |
2919 | */ | |
2920 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2921 | if (batch * PAGE_SIZE > 512 * 1024) |
2922 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2923 | batch /= 4; /* We effectively *= 4 below */ |
2924 | if (batch < 1) | |
2925 | batch = 1; | |
2926 | ||
2927 | /* | |
0ceaacc9 NP |
2928 | * Clamp the batch to a 2^n - 1 value. Having a power |
2929 | * of 2 value was found to be more likely to have | |
2930 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2931 | * |
0ceaacc9 NP |
2932 | * For example if 2 tasks are alternately allocating |
2933 | * batches of pages, one task can end up with a lot | |
2934 | * of pages of one half of the possible page colors | |
2935 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2936 | */ |
9155203a | 2937 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 2938 | |
e7c8d5c9 | 2939 | return batch; |
3a6be87f DH |
2940 | |
2941 | #else | |
2942 | /* The deferral and batching of frees should be suppressed under NOMMU | |
2943 | * conditions. | |
2944 | * | |
2945 | * The problem is that NOMMU needs to be able to allocate large chunks | |
2946 | * of contiguous memory as there's no hardware page translation to | |
2947 | * assemble apparent contiguous memory from discontiguous pages. | |
2948 | * | |
2949 | * Queueing large contiguous runs of pages for batching, however, | |
2950 | * causes the pages to actually be freed in smaller chunks. As there | |
2951 | * can be a significant delay between the individual batches being | |
2952 | * recycled, this leads to the once large chunks of space being | |
2953 | * fragmented and becoming unavailable for high-order allocations. | |
2954 | */ | |
2955 | return 0; | |
2956 | #endif | |
e7c8d5c9 CL |
2957 | } |
2958 | ||
b69a7288 | 2959 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
2960 | { |
2961 | struct per_cpu_pages *pcp; | |
2962 | ||
1c6fe946 MD |
2963 | memset(p, 0, sizeof(*p)); |
2964 | ||
3dfa5721 | 2965 | pcp = &p->pcp; |
2caaad41 | 2966 | pcp->count = 0; |
2caaad41 CL |
2967 | pcp->high = 6 * batch; |
2968 | pcp->batch = max(1UL, 1 * batch); | |
2969 | INIT_LIST_HEAD(&pcp->list); | |
2caaad41 CL |
2970 | } |
2971 | ||
8ad4b1fb RS |
2972 | /* |
2973 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2974 | * to the value high for the pageset p. | |
2975 | */ | |
2976 | ||
2977 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2978 | unsigned long high) | |
2979 | { | |
2980 | struct per_cpu_pages *pcp; | |
2981 | ||
3dfa5721 | 2982 | pcp = &p->pcp; |
8ad4b1fb RS |
2983 | pcp->high = high; |
2984 | pcp->batch = max(1UL, high/4); | |
2985 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2986 | pcp->batch = PAGE_SHIFT * 8; | |
2987 | } | |
2988 | ||
2989 | ||
e7c8d5c9 CL |
2990 | #ifdef CONFIG_NUMA |
2991 | /* | |
2caaad41 CL |
2992 | * Boot pageset table. One per cpu which is going to be used for all |
2993 | * zones and all nodes. The parameters will be set in such a way | |
2994 | * that an item put on a list will immediately be handed over to | |
2995 | * the buddy list. This is safe since pageset manipulation is done | |
2996 | * with interrupts disabled. | |
2997 | * | |
2998 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2999 | * |
3000 | * The boot_pagesets must be kept even after bootup is complete for | |
3001 | * unused processors and/or zones. They do play a role for bootstrapping | |
3002 | * hotplugged processors. | |
3003 | * | |
3004 | * zoneinfo_show() and maybe other functions do | |
3005 | * not check if the processor is online before following the pageset pointer. | |
3006 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 3007 | */ |
88a2a4ac | 3008 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
3009 | |
3010 | /* | |
3011 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
3012 | * per cpu pageset array in struct zone. |
3013 | */ | |
6292d9aa | 3014 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
3015 | { |
3016 | struct zone *zone, *dzone; | |
37c0708d CL |
3017 | int node = cpu_to_node(cpu); |
3018 | ||
3019 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 | 3020 | |
ee99c71c | 3021 | for_each_populated_zone(zone) { |
23316bc8 | 3022 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 3023 | GFP_KERNEL, node); |
23316bc8 | 3024 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 3025 | goto bad; |
e7c8d5c9 | 3026 | |
23316bc8 | 3027 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
3028 | |
3029 | if (percpu_pagelist_fraction) | |
3030 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
3031 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
3032 | } |
3033 | ||
3034 | return 0; | |
3035 | bad: | |
3036 | for_each_zone(dzone) { | |
64191688 AM |
3037 | if (!populated_zone(dzone)) |
3038 | continue; | |
e7c8d5c9 CL |
3039 | if (dzone == zone) |
3040 | break; | |
23316bc8 | 3041 | kfree(zone_pcp(dzone, cpu)); |
364df0eb | 3042 | zone_pcp(dzone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 CL |
3043 | } |
3044 | return -ENOMEM; | |
3045 | } | |
3046 | ||
3047 | static inline void free_zone_pagesets(int cpu) | |
3048 | { | |
e7c8d5c9 CL |
3049 | struct zone *zone; |
3050 | ||
3051 | for_each_zone(zone) { | |
3052 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
3053 | ||
f3ef9ead DR |
3054 | /* Free per_cpu_pageset if it is slab allocated */ |
3055 | if (pset != &boot_pageset[cpu]) | |
3056 | kfree(pset); | |
364df0eb | 3057 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 | 3058 | } |
e7c8d5c9 CL |
3059 | } |
3060 | ||
9c7b216d | 3061 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
3062 | unsigned long action, |
3063 | void *hcpu) | |
3064 | { | |
3065 | int cpu = (long)hcpu; | |
3066 | int ret = NOTIFY_OK; | |
3067 | ||
3068 | switch (action) { | |
ce421c79 | 3069 | case CPU_UP_PREPARE: |
8bb78442 | 3070 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
3071 | if (process_zones(cpu)) |
3072 | ret = NOTIFY_BAD; | |
3073 | break; | |
3074 | case CPU_UP_CANCELED: | |
8bb78442 | 3075 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 3076 | case CPU_DEAD: |
8bb78442 | 3077 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
3078 | free_zone_pagesets(cpu); |
3079 | break; | |
3080 | default: | |
3081 | break; | |
e7c8d5c9 CL |
3082 | } |
3083 | return ret; | |
3084 | } | |
3085 | ||
74b85f37 | 3086 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
3087 | { &pageset_cpuup_callback, NULL, 0 }; |
3088 | ||
78d9955b | 3089 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
3090 | { |
3091 | int err; | |
3092 | ||
3093 | /* Initialize per_cpu_pageset for cpu 0. | |
3094 | * A cpuup callback will do this for every cpu | |
3095 | * as it comes online | |
3096 | */ | |
3097 | err = process_zones(smp_processor_id()); | |
3098 | BUG_ON(err); | |
3099 | register_cpu_notifier(&pageset_notifier); | |
3100 | } | |
3101 | ||
3102 | #endif | |
3103 | ||
577a32f6 | 3104 | static noinline __init_refok |
cca448fe | 3105 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
3106 | { |
3107 | int i; | |
3108 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 3109 | size_t alloc_size; |
ed8ece2e DH |
3110 | |
3111 | /* | |
3112 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3113 | * per zone. | |
3114 | */ | |
02b694de YG |
3115 | zone->wait_table_hash_nr_entries = |
3116 | wait_table_hash_nr_entries(zone_size_pages); | |
3117 | zone->wait_table_bits = | |
3118 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
3119 | alloc_size = zone->wait_table_hash_nr_entries |
3120 | * sizeof(wait_queue_head_t); | |
3121 | ||
cd94b9db | 3122 | if (!slab_is_available()) { |
cca448fe YG |
3123 | zone->wait_table = (wait_queue_head_t *) |
3124 | alloc_bootmem_node(pgdat, alloc_size); | |
3125 | } else { | |
3126 | /* | |
3127 | * This case means that a zone whose size was 0 gets new memory | |
3128 | * via memory hot-add. | |
3129 | * But it may be the case that a new node was hot-added. In | |
3130 | * this case vmalloc() will not be able to use this new node's | |
3131 | * memory - this wait_table must be initialized to use this new | |
3132 | * node itself as well. | |
3133 | * To use this new node's memory, further consideration will be | |
3134 | * necessary. | |
3135 | */ | |
8691f3a7 | 3136 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3137 | } |
3138 | if (!zone->wait_table) | |
3139 | return -ENOMEM; | |
ed8ece2e | 3140 | |
02b694de | 3141 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3142 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3143 | |
3144 | return 0; | |
ed8ece2e DH |
3145 | } |
3146 | ||
c09b4240 | 3147 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
3148 | { |
3149 | int cpu; | |
3150 | unsigned long batch = zone_batchsize(zone); | |
3151 | ||
3152 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3153 | #ifdef CONFIG_NUMA | |
3154 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 3155 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
3156 | setup_pageset(&boot_pageset[cpu],0); |
3157 | #else | |
3158 | setup_pageset(zone_pcp(zone,cpu), batch); | |
3159 | #endif | |
3160 | } | |
f5335c0f AB |
3161 | if (zone->present_pages) |
3162 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
3163 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
3164 | } |
3165 | ||
718127cc YG |
3166 | __meminit int init_currently_empty_zone(struct zone *zone, |
3167 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3168 | unsigned long size, |
3169 | enum memmap_context context) | |
ed8ece2e DH |
3170 | { |
3171 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3172 | int ret; |
3173 | ret = zone_wait_table_init(zone, size); | |
3174 | if (ret) | |
3175 | return ret; | |
ed8ece2e DH |
3176 | pgdat->nr_zones = zone_idx(zone) + 1; |
3177 | ||
ed8ece2e DH |
3178 | zone->zone_start_pfn = zone_start_pfn; |
3179 | ||
708614e6 MG |
3180 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3181 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3182 | pgdat->node_id, | |
3183 | (unsigned long)zone_idx(zone), | |
3184 | zone_start_pfn, (zone_start_pfn + size)); | |
3185 | ||
1e548deb | 3186 | zone_init_free_lists(zone); |
718127cc YG |
3187 | |
3188 | return 0; | |
ed8ece2e DH |
3189 | } |
3190 | ||
c713216d MG |
3191 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3192 | /* | |
3193 | * Basic iterator support. Return the first range of PFNs for a node | |
3194 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3195 | */ | |
a3142c8e | 3196 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
3197 | { |
3198 | int i; | |
3199 | ||
3200 | for (i = 0; i < nr_nodemap_entries; i++) | |
3201 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3202 | return i; | |
3203 | ||
3204 | return -1; | |
3205 | } | |
3206 | ||
3207 | /* | |
3208 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 3209 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 3210 | */ |
a3142c8e | 3211 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
3212 | { |
3213 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3214 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3215 | return index; | |
3216 | ||
3217 | return -1; | |
3218 | } | |
3219 | ||
3220 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3221 | /* | |
3222 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3223 | * Architectures may implement their own version but if add_active_range() | |
3224 | * was used and there are no special requirements, this is a convenient | |
3225 | * alternative | |
3226 | */ | |
f2dbcfa7 | 3227 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
3228 | { |
3229 | int i; | |
3230 | ||
3231 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3232 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3233 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3234 | ||
3235 | if (start_pfn <= pfn && pfn < end_pfn) | |
3236 | return early_node_map[i].nid; | |
3237 | } | |
cc2559bc KH |
3238 | /* This is a memory hole */ |
3239 | return -1; | |
c713216d MG |
3240 | } |
3241 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3242 | ||
f2dbcfa7 KH |
3243 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3244 | { | |
cc2559bc KH |
3245 | int nid; |
3246 | ||
3247 | nid = __early_pfn_to_nid(pfn); | |
3248 | if (nid >= 0) | |
3249 | return nid; | |
3250 | /* just returns 0 */ | |
3251 | return 0; | |
f2dbcfa7 KH |
3252 | } |
3253 | ||
cc2559bc KH |
3254 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3255 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3256 | { | |
3257 | int nid; | |
3258 | ||
3259 | nid = __early_pfn_to_nid(pfn); | |
3260 | if (nid >= 0 && nid != node) | |
3261 | return false; | |
3262 | return true; | |
3263 | } | |
3264 | #endif | |
f2dbcfa7 | 3265 | |
c713216d MG |
3266 | /* Basic iterator support to walk early_node_map[] */ |
3267 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3268 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3269 | i = next_active_region_index_in_nid(i, nid)) | |
3270 | ||
3271 | /** | |
3272 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3273 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3274 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3275 | * |
3276 | * If an architecture guarantees that all ranges registered with | |
3277 | * add_active_ranges() contain no holes and may be freed, this | |
3278 | * this function may be used instead of calling free_bootmem() manually. | |
3279 | */ | |
3280 | void __init free_bootmem_with_active_regions(int nid, | |
3281 | unsigned long max_low_pfn) | |
3282 | { | |
3283 | int i; | |
3284 | ||
3285 | for_each_active_range_index_in_nid(i, nid) { | |
3286 | unsigned long size_pages = 0; | |
3287 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3288 | ||
3289 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3290 | continue; | |
3291 | ||
3292 | if (end_pfn > max_low_pfn) | |
3293 | end_pfn = max_low_pfn; | |
3294 | ||
3295 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3296 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3297 | PFN_PHYS(early_node_map[i].start_pfn), | |
3298 | size_pages << PAGE_SHIFT); | |
3299 | } | |
3300 | } | |
3301 | ||
b5bc6c0e YL |
3302 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
3303 | { | |
3304 | int i; | |
d52d53b8 | 3305 | int ret; |
b5bc6c0e | 3306 | |
d52d53b8 YL |
3307 | for_each_active_range_index_in_nid(i, nid) { |
3308 | ret = work_fn(early_node_map[i].start_pfn, | |
3309 | early_node_map[i].end_pfn, data); | |
3310 | if (ret) | |
3311 | break; | |
3312 | } | |
b5bc6c0e | 3313 | } |
c713216d MG |
3314 | /** |
3315 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3316 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3317 | * |
3318 | * If an architecture guarantees that all ranges registered with | |
3319 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3320 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3321 | */ |
3322 | void __init sparse_memory_present_with_active_regions(int nid) | |
3323 | { | |
3324 | int i; | |
3325 | ||
3326 | for_each_active_range_index_in_nid(i, nid) | |
3327 | memory_present(early_node_map[i].nid, | |
3328 | early_node_map[i].start_pfn, | |
3329 | early_node_map[i].end_pfn); | |
3330 | } | |
3331 | ||
3332 | /** | |
3333 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3334 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3335 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3336 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3337 | * |
3338 | * It returns the start and end page frame of a node based on information | |
3339 | * provided by an arch calling add_active_range(). If called for a node | |
3340 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3341 | * PFNs will be 0. |
c713216d | 3342 | */ |
a3142c8e | 3343 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3344 | unsigned long *start_pfn, unsigned long *end_pfn) |
3345 | { | |
3346 | int i; | |
3347 | *start_pfn = -1UL; | |
3348 | *end_pfn = 0; | |
3349 | ||
3350 | for_each_active_range_index_in_nid(i, nid) { | |
3351 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3352 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3353 | } | |
3354 | ||
633c0666 | 3355 | if (*start_pfn == -1UL) |
c713216d | 3356 | *start_pfn = 0; |
c713216d MG |
3357 | } |
3358 | ||
2a1e274a MG |
3359 | /* |
3360 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3361 | * assumption is made that zones within a node are ordered in monotonic | |
3362 | * increasing memory addresses so that the "highest" populated zone is used | |
3363 | */ | |
b69a7288 | 3364 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3365 | { |
3366 | int zone_index; | |
3367 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3368 | if (zone_index == ZONE_MOVABLE) | |
3369 | continue; | |
3370 | ||
3371 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3372 | arch_zone_lowest_possible_pfn[zone_index]) | |
3373 | break; | |
3374 | } | |
3375 | ||
3376 | VM_BUG_ON(zone_index == -1); | |
3377 | movable_zone = zone_index; | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3382 | * because it is sized independant of architecture. Unlike the other zones, | |
3383 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3384 | * in each node depending on the size of each node and how evenly kernelcore | |
3385 | * is distributed. This helper function adjusts the zone ranges | |
3386 | * provided by the architecture for a given node by using the end of the | |
3387 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3388 | * zones within a node are in order of monotonic increases memory addresses | |
3389 | */ | |
b69a7288 | 3390 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
3391 | unsigned long zone_type, |
3392 | unsigned long node_start_pfn, | |
3393 | unsigned long node_end_pfn, | |
3394 | unsigned long *zone_start_pfn, | |
3395 | unsigned long *zone_end_pfn) | |
3396 | { | |
3397 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3398 | if (zone_movable_pfn[nid]) { | |
3399 | /* Size ZONE_MOVABLE */ | |
3400 | if (zone_type == ZONE_MOVABLE) { | |
3401 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3402 | *zone_end_pfn = min(node_end_pfn, | |
3403 | arch_zone_highest_possible_pfn[movable_zone]); | |
3404 | ||
3405 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3406 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3407 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3408 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3409 | ||
3410 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3411 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3412 | *zone_start_pfn = *zone_end_pfn; | |
3413 | } | |
3414 | } | |
3415 | ||
c713216d MG |
3416 | /* |
3417 | * Return the number of pages a zone spans in a node, including holes | |
3418 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3419 | */ | |
6ea6e688 | 3420 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3421 | unsigned long zone_type, |
3422 | unsigned long *ignored) | |
3423 | { | |
3424 | unsigned long node_start_pfn, node_end_pfn; | |
3425 | unsigned long zone_start_pfn, zone_end_pfn; | |
3426 | ||
3427 | /* Get the start and end of the node and zone */ | |
3428 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3429 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3430 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3431 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3432 | node_start_pfn, node_end_pfn, | |
3433 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3434 | |
3435 | /* Check that this node has pages within the zone's required range */ | |
3436 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3437 | return 0; | |
3438 | ||
3439 | /* Move the zone boundaries inside the node if necessary */ | |
3440 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3441 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3442 | ||
3443 | /* Return the spanned pages */ | |
3444 | return zone_end_pfn - zone_start_pfn; | |
3445 | } | |
3446 | ||
3447 | /* | |
3448 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3449 | * then all holes in the requested range will be accounted for. |
c713216d | 3450 | */ |
b69a7288 | 3451 | static unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3452 | unsigned long range_start_pfn, |
3453 | unsigned long range_end_pfn) | |
3454 | { | |
3455 | int i = 0; | |
3456 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3457 | unsigned long start_pfn; | |
3458 | ||
3459 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3460 | i = first_active_region_index_in_nid(nid); | |
3461 | if (i == -1) | |
3462 | return 0; | |
3463 | ||
b5445f95 MG |
3464 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3465 | ||
9c7cd687 MG |
3466 | /* Account for ranges before physical memory on this node */ |
3467 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3468 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3469 | |
3470 | /* Find all holes for the zone within the node */ | |
3471 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3472 | ||
3473 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3474 | if (prev_end_pfn >= range_end_pfn) | |
3475 | break; | |
3476 | ||
3477 | /* Make sure the end of the zone is not within the hole */ | |
3478 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3479 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3480 | ||
3481 | /* Update the hole size cound and move on */ | |
3482 | if (start_pfn > range_start_pfn) { | |
3483 | BUG_ON(prev_end_pfn > start_pfn); | |
3484 | hole_pages += start_pfn - prev_end_pfn; | |
3485 | } | |
3486 | prev_end_pfn = early_node_map[i].end_pfn; | |
3487 | } | |
3488 | ||
9c7cd687 MG |
3489 | /* Account for ranges past physical memory on this node */ |
3490 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3491 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3492 | max(range_start_pfn, prev_end_pfn); |
3493 | ||
c713216d MG |
3494 | return hole_pages; |
3495 | } | |
3496 | ||
3497 | /** | |
3498 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3499 | * @start_pfn: The start PFN to start searching for holes | |
3500 | * @end_pfn: The end PFN to stop searching for holes | |
3501 | * | |
88ca3b94 | 3502 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3503 | */ |
3504 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3505 | unsigned long end_pfn) | |
3506 | { | |
3507 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3508 | } | |
3509 | ||
3510 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3511 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3512 | unsigned long zone_type, |
3513 | unsigned long *ignored) | |
3514 | { | |
9c7cd687 MG |
3515 | unsigned long node_start_pfn, node_end_pfn; |
3516 | unsigned long zone_start_pfn, zone_end_pfn; | |
3517 | ||
3518 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3519 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3520 | node_start_pfn); | |
3521 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3522 | node_end_pfn); | |
3523 | ||
2a1e274a MG |
3524 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3525 | node_start_pfn, node_end_pfn, | |
3526 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3527 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3528 | } |
0e0b864e | 3529 | |
c713216d | 3530 | #else |
6ea6e688 | 3531 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3532 | unsigned long zone_type, |
3533 | unsigned long *zones_size) | |
3534 | { | |
3535 | return zones_size[zone_type]; | |
3536 | } | |
3537 | ||
6ea6e688 | 3538 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3539 | unsigned long zone_type, |
3540 | unsigned long *zholes_size) | |
3541 | { | |
3542 | if (!zholes_size) | |
3543 | return 0; | |
3544 | ||
3545 | return zholes_size[zone_type]; | |
3546 | } | |
0e0b864e | 3547 | |
c713216d MG |
3548 | #endif |
3549 | ||
a3142c8e | 3550 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3551 | unsigned long *zones_size, unsigned long *zholes_size) |
3552 | { | |
3553 | unsigned long realtotalpages, totalpages = 0; | |
3554 | enum zone_type i; | |
3555 | ||
3556 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3557 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3558 | zones_size); | |
3559 | pgdat->node_spanned_pages = totalpages; | |
3560 | ||
3561 | realtotalpages = totalpages; | |
3562 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3563 | realtotalpages -= | |
3564 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3565 | zholes_size); | |
3566 | pgdat->node_present_pages = realtotalpages; | |
3567 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3568 | realtotalpages); | |
3569 | } | |
3570 | ||
835c134e MG |
3571 | #ifndef CONFIG_SPARSEMEM |
3572 | /* | |
3573 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3574 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3575 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3576 | * round what is now in bits to nearest long in bits, then return it in |
3577 | * bytes. | |
3578 | */ | |
3579 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3580 | { | |
3581 | unsigned long usemapsize; | |
3582 | ||
d9c23400 MG |
3583 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3584 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3585 | usemapsize *= NR_PAGEBLOCK_BITS; |
3586 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3587 | ||
3588 | return usemapsize / 8; | |
3589 | } | |
3590 | ||
3591 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3592 | struct zone *zone, unsigned long zonesize) | |
3593 | { | |
3594 | unsigned long usemapsize = usemap_size(zonesize); | |
3595 | zone->pageblock_flags = NULL; | |
58a01a45 | 3596 | if (usemapsize) |
835c134e | 3597 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
835c134e MG |
3598 | } |
3599 | #else | |
3600 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3601 | struct zone *zone, unsigned long zonesize) {} | |
3602 | #endif /* CONFIG_SPARSEMEM */ | |
3603 | ||
d9c23400 | 3604 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3605 | |
3606 | /* Return a sensible default order for the pageblock size. */ | |
3607 | static inline int pageblock_default_order(void) | |
3608 | { | |
3609 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3610 | return HUGETLB_PAGE_ORDER; | |
3611 | ||
3612 | return MAX_ORDER-1; | |
3613 | } | |
3614 | ||
d9c23400 MG |
3615 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3616 | static inline void __init set_pageblock_order(unsigned int order) | |
3617 | { | |
3618 | /* Check that pageblock_nr_pages has not already been setup */ | |
3619 | if (pageblock_order) | |
3620 | return; | |
3621 | ||
3622 | /* | |
3623 | * Assume the largest contiguous order of interest is a huge page. | |
3624 | * This value may be variable depending on boot parameters on IA64 | |
3625 | */ | |
3626 | pageblock_order = order; | |
3627 | } | |
3628 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3629 | ||
ba72cb8c MG |
3630 | /* |
3631 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3632 | * and pageblock_default_order() are unused as pageblock_order is set | |
3633 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3634 | * pageblock_order based on the kernel config | |
3635 | */ | |
3636 | static inline int pageblock_default_order(unsigned int order) | |
3637 | { | |
3638 | return MAX_ORDER-1; | |
3639 | } | |
d9c23400 MG |
3640 | #define set_pageblock_order(x) do {} while (0) |
3641 | ||
3642 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3643 | ||
1da177e4 LT |
3644 | /* |
3645 | * Set up the zone data structures: | |
3646 | * - mark all pages reserved | |
3647 | * - mark all memory queues empty | |
3648 | * - clear the memory bitmaps | |
3649 | */ | |
b5a0e011 | 3650 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3651 | unsigned long *zones_size, unsigned long *zholes_size) |
3652 | { | |
2f1b6248 | 3653 | enum zone_type j; |
ed8ece2e | 3654 | int nid = pgdat->node_id; |
1da177e4 | 3655 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3656 | int ret; |
1da177e4 | 3657 | |
208d54e5 | 3658 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3659 | pgdat->nr_zones = 0; |
3660 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3661 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 3662 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
3663 | |
3664 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3665 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3666 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 3667 | enum lru_list l; |
1da177e4 | 3668 | |
c713216d MG |
3669 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3670 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3671 | zholes_size); | |
1da177e4 | 3672 | |
0e0b864e MG |
3673 | /* |
3674 | * Adjust realsize so that it accounts for how much memory | |
3675 | * is used by this zone for memmap. This affects the watermark | |
3676 | * and per-cpu initialisations | |
3677 | */ | |
f7232154 JW |
3678 | memmap_pages = |
3679 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
3680 | if (realsize >= memmap_pages) { |
3681 | realsize -= memmap_pages; | |
5594c8c8 YL |
3682 | if (memmap_pages) |
3683 | printk(KERN_DEBUG | |
3684 | " %s zone: %lu pages used for memmap\n", | |
3685 | zone_names[j], memmap_pages); | |
0e0b864e MG |
3686 | } else |
3687 | printk(KERN_WARNING | |
3688 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3689 | zone_names[j], memmap_pages, realsize); | |
3690 | ||
6267276f CL |
3691 | /* Account for reserved pages */ |
3692 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3693 | realsize -= dma_reserve; |
d903ef9f | 3694 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 3695 | zone_names[0], dma_reserve); |
0e0b864e MG |
3696 | } |
3697 | ||
98d2b0eb | 3698 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3699 | nr_kernel_pages += realsize; |
3700 | nr_all_pages += realsize; | |
3701 | ||
3702 | zone->spanned_pages = size; | |
3703 | zone->present_pages = realsize; | |
9614634f | 3704 | #ifdef CONFIG_NUMA |
d5f541ed | 3705 | zone->node = nid; |
8417bba4 | 3706 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3707 | / 100; |
0ff38490 | 3708 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3709 | #endif |
1da177e4 LT |
3710 | zone->name = zone_names[j]; |
3711 | spin_lock_init(&zone->lock); | |
3712 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3713 | zone_seqlock_init(zone); |
1da177e4 | 3714 | zone->zone_pgdat = pgdat; |
1da177e4 | 3715 | |
3bb1a852 | 3716 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3717 | |
ed8ece2e | 3718 | zone_pcp_init(zone); |
b69408e8 CL |
3719 | for_each_lru(l) { |
3720 | INIT_LIST_HEAD(&zone->lru[l].list); | |
6e08a369 | 3721 | zone->lru[l].nr_saved_scan = 0; |
b69408e8 | 3722 | } |
6e901571 KM |
3723 | zone->reclaim_stat.recent_rotated[0] = 0; |
3724 | zone->reclaim_stat.recent_rotated[1] = 0; | |
3725 | zone->reclaim_stat.recent_scanned[0] = 0; | |
3726 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 3727 | zap_zone_vm_stats(zone); |
e815af95 | 3728 | zone->flags = 0; |
1da177e4 LT |
3729 | if (!size) |
3730 | continue; | |
3731 | ||
ba72cb8c | 3732 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3733 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3734 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3735 | size, MEMMAP_EARLY); | |
718127cc | 3736 | BUG_ON(ret); |
76cdd58e | 3737 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 3738 | zone_start_pfn += size; |
1da177e4 LT |
3739 | } |
3740 | } | |
3741 | ||
577a32f6 | 3742 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3743 | { |
1da177e4 LT |
3744 | /* Skip empty nodes */ |
3745 | if (!pgdat->node_spanned_pages) | |
3746 | return; | |
3747 | ||
d41dee36 | 3748 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3749 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3750 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3751 | unsigned long size, start, end; |
d41dee36 AW |
3752 | struct page *map; |
3753 | ||
e984bb43 BP |
3754 | /* |
3755 | * The zone's endpoints aren't required to be MAX_ORDER | |
3756 | * aligned but the node_mem_map endpoints must be in order | |
3757 | * for the buddy allocator to function correctly. | |
3758 | */ | |
3759 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3760 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3761 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3762 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3763 | map = alloc_remap(pgdat->node_id, size); |
3764 | if (!map) | |
3765 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3766 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3767 | } |
12d810c1 | 3768 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3769 | /* |
3770 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3771 | */ | |
c713216d | 3772 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3773 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3774 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3775 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3776 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3777 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3778 | } | |
1da177e4 | 3779 | #endif |
d41dee36 | 3780 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3781 | } |
3782 | ||
9109fb7b JW |
3783 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
3784 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 3785 | { |
9109fb7b JW |
3786 | pg_data_t *pgdat = NODE_DATA(nid); |
3787 | ||
1da177e4 LT |
3788 | pgdat->node_id = nid; |
3789 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3790 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3791 | |
3792 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
3793 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
3794 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3795 | nid, (unsigned long)pgdat, | |
3796 | (unsigned long)pgdat->node_mem_map); | |
3797 | #endif | |
1da177e4 LT |
3798 | |
3799 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3800 | } | |
3801 | ||
c713216d | 3802 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3803 | |
3804 | #if MAX_NUMNODES > 1 | |
3805 | /* | |
3806 | * Figure out the number of possible node ids. | |
3807 | */ | |
3808 | static void __init setup_nr_node_ids(void) | |
3809 | { | |
3810 | unsigned int node; | |
3811 | unsigned int highest = 0; | |
3812 | ||
3813 | for_each_node_mask(node, node_possible_map) | |
3814 | highest = node; | |
3815 | nr_node_ids = highest + 1; | |
3816 | } | |
3817 | #else | |
3818 | static inline void setup_nr_node_ids(void) | |
3819 | { | |
3820 | } | |
3821 | #endif | |
3822 | ||
c713216d MG |
3823 | /** |
3824 | * add_active_range - Register a range of PFNs backed by physical memory | |
3825 | * @nid: The node ID the range resides on | |
3826 | * @start_pfn: The start PFN of the available physical memory | |
3827 | * @end_pfn: The end PFN of the available physical memory | |
3828 | * | |
3829 | * These ranges are stored in an early_node_map[] and later used by | |
3830 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3831 | * range spans a memory hole, it is up to the architecture to ensure | |
3832 | * the memory is not freed by the bootmem allocator. If possible | |
3833 | * the range being registered will be merged with existing ranges. | |
3834 | */ | |
3835 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3836 | unsigned long end_pfn) | |
3837 | { | |
3838 | int i; | |
3839 | ||
6b74ab97 MG |
3840 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
3841 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3842 | "%d entries of %d used\n", | |
3843 | nid, start_pfn, end_pfn, | |
3844 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 3845 | |
2dbb51c4 MG |
3846 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
3847 | ||
c713216d MG |
3848 | /* Merge with existing active regions if possible */ |
3849 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3850 | if (early_node_map[i].nid != nid) | |
3851 | continue; | |
3852 | ||
3853 | /* Skip if an existing region covers this new one */ | |
3854 | if (start_pfn >= early_node_map[i].start_pfn && | |
3855 | end_pfn <= early_node_map[i].end_pfn) | |
3856 | return; | |
3857 | ||
3858 | /* Merge forward if suitable */ | |
3859 | if (start_pfn <= early_node_map[i].end_pfn && | |
3860 | end_pfn > early_node_map[i].end_pfn) { | |
3861 | early_node_map[i].end_pfn = end_pfn; | |
3862 | return; | |
3863 | } | |
3864 | ||
3865 | /* Merge backward if suitable */ | |
3866 | if (start_pfn < early_node_map[i].end_pfn && | |
3867 | end_pfn >= early_node_map[i].start_pfn) { | |
3868 | early_node_map[i].start_pfn = start_pfn; | |
3869 | return; | |
3870 | } | |
3871 | } | |
3872 | ||
3873 | /* Check that early_node_map is large enough */ | |
3874 | if (i >= MAX_ACTIVE_REGIONS) { | |
3875 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3876 | MAX_ACTIVE_REGIONS); | |
3877 | return; | |
3878 | } | |
3879 | ||
3880 | early_node_map[i].nid = nid; | |
3881 | early_node_map[i].start_pfn = start_pfn; | |
3882 | early_node_map[i].end_pfn = end_pfn; | |
3883 | nr_nodemap_entries = i + 1; | |
3884 | } | |
3885 | ||
3886 | /** | |
cc1050ba | 3887 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 3888 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
3889 | * @start_pfn: The new PFN of the range |
3890 | * @end_pfn: The new PFN of the range | |
c713216d MG |
3891 | * |
3892 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
3893 | * The map is kept near the end physical page range that has already been |
3894 | * registered. This function allows an arch to shrink an existing registered | |
3895 | * range. | |
c713216d | 3896 | */ |
cc1050ba YL |
3897 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
3898 | unsigned long end_pfn) | |
c713216d | 3899 | { |
cc1a9d86 YL |
3900 | int i, j; |
3901 | int removed = 0; | |
c713216d | 3902 | |
cc1050ba YL |
3903 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
3904 | nid, start_pfn, end_pfn); | |
3905 | ||
c713216d | 3906 | /* Find the old active region end and shrink */ |
cc1a9d86 | 3907 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
3908 | if (early_node_map[i].start_pfn >= start_pfn && |
3909 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 3910 | /* clear it */ |
cc1050ba | 3911 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
3912 | early_node_map[i].end_pfn = 0; |
3913 | removed = 1; | |
3914 | continue; | |
3915 | } | |
cc1050ba YL |
3916 | if (early_node_map[i].start_pfn < start_pfn && |
3917 | early_node_map[i].end_pfn > start_pfn) { | |
3918 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
3919 | early_node_map[i].end_pfn = start_pfn; | |
3920 | if (temp_end_pfn > end_pfn) | |
3921 | add_active_range(nid, end_pfn, temp_end_pfn); | |
3922 | continue; | |
3923 | } | |
3924 | if (early_node_map[i].start_pfn >= start_pfn && | |
3925 | early_node_map[i].end_pfn > end_pfn && | |
3926 | early_node_map[i].start_pfn < end_pfn) { | |
3927 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 3928 | continue; |
c713216d | 3929 | } |
cc1a9d86 YL |
3930 | } |
3931 | ||
3932 | if (!removed) | |
3933 | return; | |
3934 | ||
3935 | /* remove the blank ones */ | |
3936 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
3937 | if (early_node_map[i].nid != nid) | |
3938 | continue; | |
3939 | if (early_node_map[i].end_pfn) | |
3940 | continue; | |
3941 | /* we found it, get rid of it */ | |
3942 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
3943 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
3944 | sizeof(early_node_map[j])); | |
3945 | j = nr_nodemap_entries - 1; | |
3946 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
3947 | nr_nodemap_entries--; | |
3948 | } | |
c713216d MG |
3949 | } |
3950 | ||
3951 | /** | |
3952 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3953 | * |
c713216d MG |
3954 | * During discovery, it may be found that a table like SRAT is invalid |
3955 | * and an alternative discovery method must be used. This function removes | |
3956 | * all currently registered regions. | |
3957 | */ | |
88ca3b94 | 3958 | void __init remove_all_active_ranges(void) |
c713216d MG |
3959 | { |
3960 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3961 | nr_nodemap_entries = 0; | |
3962 | } | |
3963 | ||
3964 | /* Compare two active node_active_regions */ | |
3965 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3966 | { | |
3967 | struct node_active_region *arange = (struct node_active_region *)a; | |
3968 | struct node_active_region *brange = (struct node_active_region *)b; | |
3969 | ||
3970 | /* Done this way to avoid overflows */ | |
3971 | if (arange->start_pfn > brange->start_pfn) | |
3972 | return 1; | |
3973 | if (arange->start_pfn < brange->start_pfn) | |
3974 | return -1; | |
3975 | ||
3976 | return 0; | |
3977 | } | |
3978 | ||
3979 | /* sort the node_map by start_pfn */ | |
3980 | static void __init sort_node_map(void) | |
3981 | { | |
3982 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3983 | sizeof(struct node_active_region), | |
3984 | cmp_node_active_region, NULL); | |
3985 | } | |
3986 | ||
a6af2bc3 | 3987 | /* Find the lowest pfn for a node */ |
b69a7288 | 3988 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
3989 | { |
3990 | int i; | |
a6af2bc3 | 3991 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3992 | |
c713216d MG |
3993 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3994 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3995 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3996 | |
a6af2bc3 MG |
3997 | if (min_pfn == ULONG_MAX) { |
3998 | printk(KERN_WARNING | |
2bc0d261 | 3999 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4000 | return 0; |
4001 | } | |
4002 | ||
4003 | return min_pfn; | |
c713216d MG |
4004 | } |
4005 | ||
4006 | /** | |
4007 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4008 | * | |
4009 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4010 | * add_active_range(). |
c713216d MG |
4011 | */ |
4012 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4013 | { | |
4014 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4015 | } | |
4016 | ||
37b07e41 LS |
4017 | /* |
4018 | * early_calculate_totalpages() | |
4019 | * Sum pages in active regions for movable zone. | |
4020 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4021 | */ | |
484f51f8 | 4022 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
4023 | { |
4024 | int i; | |
4025 | unsigned long totalpages = 0; | |
4026 | ||
37b07e41 LS |
4027 | for (i = 0; i < nr_nodemap_entries; i++) { |
4028 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 4029 | early_node_map[i].start_pfn; |
37b07e41 LS |
4030 | totalpages += pages; |
4031 | if (pages) | |
4032 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
4033 | } | |
4034 | return totalpages; | |
7e63efef MG |
4035 | } |
4036 | ||
2a1e274a MG |
4037 | /* |
4038 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4039 | * is spread evenly between nodes as long as the nodes have enough | |
4040 | * memory. When they don't, some nodes will have more kernelcore than | |
4041 | * others | |
4042 | */ | |
b69a7288 | 4043 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
4044 | { |
4045 | int i, nid; | |
4046 | unsigned long usable_startpfn; | |
4047 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4048 | /* save the state before borrow the nodemask */ |
4049 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4050 | unsigned long totalpages = early_calculate_totalpages(); |
4051 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4052 | |
7e63efef MG |
4053 | /* |
4054 | * If movablecore was specified, calculate what size of | |
4055 | * kernelcore that corresponds so that memory usable for | |
4056 | * any allocation type is evenly spread. If both kernelcore | |
4057 | * and movablecore are specified, then the value of kernelcore | |
4058 | * will be used for required_kernelcore if it's greater than | |
4059 | * what movablecore would have allowed. | |
4060 | */ | |
4061 | if (required_movablecore) { | |
7e63efef MG |
4062 | unsigned long corepages; |
4063 | ||
4064 | /* | |
4065 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4066 | * was requested by the user | |
4067 | */ | |
4068 | required_movablecore = | |
4069 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4070 | corepages = totalpages - required_movablecore; | |
4071 | ||
4072 | required_kernelcore = max(required_kernelcore, corepages); | |
4073 | } | |
4074 | ||
2a1e274a MG |
4075 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4076 | if (!required_kernelcore) | |
66918dcd | 4077 | goto out; |
2a1e274a MG |
4078 | |
4079 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4080 | find_usable_zone_for_movable(); | |
4081 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4082 | ||
4083 | restart: | |
4084 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4085 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4086 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
4087 | /* |
4088 | * Recalculate kernelcore_node if the division per node | |
4089 | * now exceeds what is necessary to satisfy the requested | |
4090 | * amount of memory for the kernel | |
4091 | */ | |
4092 | if (required_kernelcore < kernelcore_node) | |
4093 | kernelcore_node = required_kernelcore / usable_nodes; | |
4094 | ||
4095 | /* | |
4096 | * As the map is walked, we track how much memory is usable | |
4097 | * by the kernel using kernelcore_remaining. When it is | |
4098 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4099 | */ | |
4100 | kernelcore_remaining = kernelcore_node; | |
4101 | ||
4102 | /* Go through each range of PFNs within this node */ | |
4103 | for_each_active_range_index_in_nid(i, nid) { | |
4104 | unsigned long start_pfn, end_pfn; | |
4105 | unsigned long size_pages; | |
4106 | ||
4107 | start_pfn = max(early_node_map[i].start_pfn, | |
4108 | zone_movable_pfn[nid]); | |
4109 | end_pfn = early_node_map[i].end_pfn; | |
4110 | if (start_pfn >= end_pfn) | |
4111 | continue; | |
4112 | ||
4113 | /* Account for what is only usable for kernelcore */ | |
4114 | if (start_pfn < usable_startpfn) { | |
4115 | unsigned long kernel_pages; | |
4116 | kernel_pages = min(end_pfn, usable_startpfn) | |
4117 | - start_pfn; | |
4118 | ||
4119 | kernelcore_remaining -= min(kernel_pages, | |
4120 | kernelcore_remaining); | |
4121 | required_kernelcore -= min(kernel_pages, | |
4122 | required_kernelcore); | |
4123 | ||
4124 | /* Continue if range is now fully accounted */ | |
4125 | if (end_pfn <= usable_startpfn) { | |
4126 | ||
4127 | /* | |
4128 | * Push zone_movable_pfn to the end so | |
4129 | * that if we have to rebalance | |
4130 | * kernelcore across nodes, we will | |
4131 | * not double account here | |
4132 | */ | |
4133 | zone_movable_pfn[nid] = end_pfn; | |
4134 | continue; | |
4135 | } | |
4136 | start_pfn = usable_startpfn; | |
4137 | } | |
4138 | ||
4139 | /* | |
4140 | * The usable PFN range for ZONE_MOVABLE is from | |
4141 | * start_pfn->end_pfn. Calculate size_pages as the | |
4142 | * number of pages used as kernelcore | |
4143 | */ | |
4144 | size_pages = end_pfn - start_pfn; | |
4145 | if (size_pages > kernelcore_remaining) | |
4146 | size_pages = kernelcore_remaining; | |
4147 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4148 | ||
4149 | /* | |
4150 | * Some kernelcore has been met, update counts and | |
4151 | * break if the kernelcore for this node has been | |
4152 | * satisified | |
4153 | */ | |
4154 | required_kernelcore -= min(required_kernelcore, | |
4155 | size_pages); | |
4156 | kernelcore_remaining -= size_pages; | |
4157 | if (!kernelcore_remaining) | |
4158 | break; | |
4159 | } | |
4160 | } | |
4161 | ||
4162 | /* | |
4163 | * If there is still required_kernelcore, we do another pass with one | |
4164 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4165 | * along on the nodes that still have memory until kernelcore is | |
4166 | * satisified | |
4167 | */ | |
4168 | usable_nodes--; | |
4169 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4170 | goto restart; | |
4171 | ||
4172 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4173 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4174 | zone_movable_pfn[nid] = | |
4175 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4176 | |
4177 | out: | |
4178 | /* restore the node_state */ | |
4179 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4180 | } |
4181 | ||
37b07e41 LS |
4182 | /* Any regular memory on that node ? */ |
4183 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4184 | { | |
4185 | #ifdef CONFIG_HIGHMEM | |
4186 | enum zone_type zone_type; | |
4187 | ||
4188 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4189 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4190 | if (zone->present_pages) | |
4191 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4192 | } | |
4193 | #endif | |
4194 | } | |
4195 | ||
c713216d MG |
4196 | /** |
4197 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4198 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4199 | * |
4200 | * This will call free_area_init_node() for each active node in the system. | |
4201 | * Using the page ranges provided by add_active_range(), the size of each | |
4202 | * zone in each node and their holes is calculated. If the maximum PFN | |
4203 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4204 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4205 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4206 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4207 | * at arch_max_dma_pfn. | |
4208 | */ | |
4209 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4210 | { | |
4211 | unsigned long nid; | |
db99100d | 4212 | int i; |
c713216d | 4213 | |
a6af2bc3 MG |
4214 | /* Sort early_node_map as initialisation assumes it is sorted */ |
4215 | sort_node_map(); | |
4216 | ||
c713216d MG |
4217 | /* Record where the zone boundaries are */ |
4218 | memset(arch_zone_lowest_possible_pfn, 0, | |
4219 | sizeof(arch_zone_lowest_possible_pfn)); | |
4220 | memset(arch_zone_highest_possible_pfn, 0, | |
4221 | sizeof(arch_zone_highest_possible_pfn)); | |
4222 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4223 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4224 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4225 | if (i == ZONE_MOVABLE) |
4226 | continue; | |
c713216d MG |
4227 | arch_zone_lowest_possible_pfn[i] = |
4228 | arch_zone_highest_possible_pfn[i-1]; | |
4229 | arch_zone_highest_possible_pfn[i] = | |
4230 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4231 | } | |
2a1e274a MG |
4232 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4233 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4234 | ||
4235 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4236 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4237 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4238 | |
c713216d MG |
4239 | /* Print out the zone ranges */ |
4240 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4241 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4242 | if (i == ZONE_MOVABLE) | |
4243 | continue; | |
5dab8ec1 | 4244 | printk(" %-8s %0#10lx -> %0#10lx\n", |
c713216d MG |
4245 | zone_names[i], |
4246 | arch_zone_lowest_possible_pfn[i], | |
4247 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4248 | } |
4249 | ||
4250 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4251 | printk("Movable zone start PFN for each node\n"); | |
4252 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4253 | if (zone_movable_pfn[i]) | |
4254 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4255 | } | |
c713216d MG |
4256 | |
4257 | /* Print out the early_node_map[] */ | |
4258 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4259 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 4260 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
4261 | early_node_map[i].start_pfn, |
4262 | early_node_map[i].end_pfn); | |
4263 | ||
4264 | /* Initialise every node */ | |
708614e6 | 4265 | mminit_verify_pageflags_layout(); |
8ef82866 | 4266 | setup_nr_node_ids(); |
c713216d MG |
4267 | for_each_online_node(nid) { |
4268 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4269 | free_area_init_node(nid, NULL, |
c713216d | 4270 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4271 | |
4272 | /* Any memory on that node */ | |
4273 | if (pgdat->node_present_pages) | |
4274 | node_set_state(nid, N_HIGH_MEMORY); | |
4275 | check_for_regular_memory(pgdat); | |
c713216d MG |
4276 | } |
4277 | } | |
2a1e274a | 4278 | |
7e63efef | 4279 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4280 | { |
4281 | unsigned long long coremem; | |
4282 | if (!p) | |
4283 | return -EINVAL; | |
4284 | ||
4285 | coremem = memparse(p, &p); | |
7e63efef | 4286 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4287 | |
7e63efef | 4288 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4289 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4290 | ||
4291 | return 0; | |
4292 | } | |
ed7ed365 | 4293 | |
7e63efef MG |
4294 | /* |
4295 | * kernelcore=size sets the amount of memory for use for allocations that | |
4296 | * cannot be reclaimed or migrated. | |
4297 | */ | |
4298 | static int __init cmdline_parse_kernelcore(char *p) | |
4299 | { | |
4300 | return cmdline_parse_core(p, &required_kernelcore); | |
4301 | } | |
4302 | ||
4303 | /* | |
4304 | * movablecore=size sets the amount of memory for use for allocations that | |
4305 | * can be reclaimed or migrated. | |
4306 | */ | |
4307 | static int __init cmdline_parse_movablecore(char *p) | |
4308 | { | |
4309 | return cmdline_parse_core(p, &required_movablecore); | |
4310 | } | |
4311 | ||
ed7ed365 | 4312 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4313 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4314 | |
c713216d MG |
4315 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4316 | ||
0e0b864e | 4317 | /** |
88ca3b94 RD |
4318 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4319 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4320 | * |
4321 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4322 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4323 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4324 | * function may optionally be used to account for unfreeable pages in the |
4325 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4326 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4327 | */ |
4328 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4329 | { | |
4330 | dma_reserve = new_dma_reserve; | |
4331 | } | |
4332 | ||
93b7504e | 4333 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
52765583 | 4334 | struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] }; |
1da177e4 | 4335 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4336 | #endif |
1da177e4 LT |
4337 | |
4338 | void __init free_area_init(unsigned long *zones_size) | |
4339 | { | |
9109fb7b | 4340 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4341 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4342 | } | |
1da177e4 | 4343 | |
1da177e4 LT |
4344 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4345 | unsigned long action, void *hcpu) | |
4346 | { | |
4347 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4348 | |
8bb78442 | 4349 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4350 | drain_pages(cpu); |
4351 | ||
4352 | /* | |
4353 | * Spill the event counters of the dead processor | |
4354 | * into the current processors event counters. | |
4355 | * This artificially elevates the count of the current | |
4356 | * processor. | |
4357 | */ | |
f8891e5e | 4358 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4359 | |
4360 | /* | |
4361 | * Zero the differential counters of the dead processor | |
4362 | * so that the vm statistics are consistent. | |
4363 | * | |
4364 | * This is only okay since the processor is dead and cannot | |
4365 | * race with what we are doing. | |
4366 | */ | |
2244b95a | 4367 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4368 | } |
4369 | return NOTIFY_OK; | |
4370 | } | |
1da177e4 LT |
4371 | |
4372 | void __init page_alloc_init(void) | |
4373 | { | |
4374 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4375 | } | |
4376 | ||
cb45b0e9 HA |
4377 | /* |
4378 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4379 | * or min_free_kbytes changes. | |
4380 | */ | |
4381 | static void calculate_totalreserve_pages(void) | |
4382 | { | |
4383 | struct pglist_data *pgdat; | |
4384 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4385 | enum zone_type i, j; |
cb45b0e9 HA |
4386 | |
4387 | for_each_online_pgdat(pgdat) { | |
4388 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4389 | struct zone *zone = pgdat->node_zones + i; | |
4390 | unsigned long max = 0; | |
4391 | ||
4392 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4393 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4394 | if (zone->lowmem_reserve[j] > max) | |
4395 | max = zone->lowmem_reserve[j]; | |
4396 | } | |
4397 | ||
41858966 MG |
4398 | /* we treat the high watermark as reserved pages. */ |
4399 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
4400 | |
4401 | if (max > zone->present_pages) | |
4402 | max = zone->present_pages; | |
4403 | reserve_pages += max; | |
4404 | } | |
4405 | } | |
4406 | totalreserve_pages = reserve_pages; | |
4407 | } | |
4408 | ||
1da177e4 LT |
4409 | /* |
4410 | * setup_per_zone_lowmem_reserve - called whenever | |
4411 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4412 | * has a correct pages reserved value, so an adequate number of | |
4413 | * pages are left in the zone after a successful __alloc_pages(). | |
4414 | */ | |
4415 | static void setup_per_zone_lowmem_reserve(void) | |
4416 | { | |
4417 | struct pglist_data *pgdat; | |
2f6726e5 | 4418 | enum zone_type j, idx; |
1da177e4 | 4419 | |
ec936fc5 | 4420 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4421 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4422 | struct zone *zone = pgdat->node_zones + j; | |
4423 | unsigned long present_pages = zone->present_pages; | |
4424 | ||
4425 | zone->lowmem_reserve[j] = 0; | |
4426 | ||
2f6726e5 CL |
4427 | idx = j; |
4428 | while (idx) { | |
1da177e4 LT |
4429 | struct zone *lower_zone; |
4430 | ||
2f6726e5 CL |
4431 | idx--; |
4432 | ||
1da177e4 LT |
4433 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4434 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4435 | ||
4436 | lower_zone = pgdat->node_zones + idx; | |
4437 | lower_zone->lowmem_reserve[j] = present_pages / | |
4438 | sysctl_lowmem_reserve_ratio[idx]; | |
4439 | present_pages += lower_zone->present_pages; | |
4440 | } | |
4441 | } | |
4442 | } | |
cb45b0e9 HA |
4443 | |
4444 | /* update totalreserve_pages */ | |
4445 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4446 | } |
4447 | ||
88ca3b94 | 4448 | /** |
bc75d33f | 4449 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
bce7394a | 4450 | * or when memory is hot-{added|removed} |
88ca3b94 | 4451 | * |
bc75d33f MK |
4452 | * Ensures that the watermark[min,low,high] values for each zone are set |
4453 | * correctly with respect to min_free_kbytes. | |
1da177e4 | 4454 | */ |
bc75d33f | 4455 | void setup_per_zone_wmarks(void) |
1da177e4 LT |
4456 | { |
4457 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4458 | unsigned long lowmem_pages = 0; | |
4459 | struct zone *zone; | |
4460 | unsigned long flags; | |
4461 | ||
4462 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4463 | for_each_zone(zone) { | |
4464 | if (!is_highmem(zone)) | |
4465 | lowmem_pages += zone->present_pages; | |
4466 | } | |
4467 | ||
4468 | for_each_zone(zone) { | |
ac924c60 AM |
4469 | u64 tmp; |
4470 | ||
1125b4e3 | 4471 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4472 | tmp = (u64)pages_min * zone->present_pages; |
4473 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4474 | if (is_highmem(zone)) { |
4475 | /* | |
669ed175 NP |
4476 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4477 | * need highmem pages, so cap pages_min to a small | |
4478 | * value here. | |
4479 | * | |
41858966 | 4480 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
4481 | * deltas controls asynch page reclaim, and so should |
4482 | * not be capped for highmem. | |
1da177e4 LT |
4483 | */ |
4484 | int min_pages; | |
4485 | ||
4486 | min_pages = zone->present_pages / 1024; | |
4487 | if (min_pages < SWAP_CLUSTER_MAX) | |
4488 | min_pages = SWAP_CLUSTER_MAX; | |
4489 | if (min_pages > 128) | |
4490 | min_pages = 128; | |
41858966 | 4491 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 4492 | } else { |
669ed175 NP |
4493 | /* |
4494 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4495 | * proportionate to the zone's size. |
4496 | */ | |
41858966 | 4497 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
4498 | } |
4499 | ||
41858966 MG |
4500 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
4501 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
56fd56b8 | 4502 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4503 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4504 | } |
cb45b0e9 HA |
4505 | |
4506 | /* update totalreserve_pages */ | |
4507 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4508 | } |
4509 | ||
556adecb | 4510 | /** |
556adecb RR |
4511 | * The inactive anon list should be small enough that the VM never has to |
4512 | * do too much work, but large enough that each inactive page has a chance | |
4513 | * to be referenced again before it is swapped out. | |
4514 | * | |
4515 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4516 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4517 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4518 | * the anonymous pages are kept on the inactive list. | |
4519 | * | |
4520 | * total target max | |
4521 | * memory ratio inactive anon | |
4522 | * ------------------------------------- | |
4523 | * 10MB 1 5MB | |
4524 | * 100MB 1 50MB | |
4525 | * 1GB 3 250MB | |
4526 | * 10GB 10 0.9GB | |
4527 | * 100GB 31 3GB | |
4528 | * 1TB 101 10GB | |
4529 | * 10TB 320 32GB | |
4530 | */ | |
96cb4df5 | 4531 | void calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 4532 | { |
96cb4df5 | 4533 | unsigned int gb, ratio; |
556adecb | 4534 | |
96cb4df5 MK |
4535 | /* Zone size in gigabytes */ |
4536 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
4537 | if (gb) | |
556adecb | 4538 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
4539 | else |
4540 | ratio = 1; | |
556adecb | 4541 | |
96cb4df5 MK |
4542 | zone->inactive_ratio = ratio; |
4543 | } | |
556adecb | 4544 | |
96cb4df5 MK |
4545 | static void __init setup_per_zone_inactive_ratio(void) |
4546 | { | |
4547 | struct zone *zone; | |
4548 | ||
4549 | for_each_zone(zone) | |
4550 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
4551 | } |
4552 | ||
1da177e4 LT |
4553 | /* |
4554 | * Initialise min_free_kbytes. | |
4555 | * | |
4556 | * For small machines we want it small (128k min). For large machines | |
4557 | * we want it large (64MB max). But it is not linear, because network | |
4558 | * bandwidth does not increase linearly with machine size. We use | |
4559 | * | |
4560 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4561 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4562 | * | |
4563 | * which yields | |
4564 | * | |
4565 | * 16MB: 512k | |
4566 | * 32MB: 724k | |
4567 | * 64MB: 1024k | |
4568 | * 128MB: 1448k | |
4569 | * 256MB: 2048k | |
4570 | * 512MB: 2896k | |
4571 | * 1024MB: 4096k | |
4572 | * 2048MB: 5792k | |
4573 | * 4096MB: 8192k | |
4574 | * 8192MB: 11584k | |
4575 | * 16384MB: 16384k | |
4576 | */ | |
bc75d33f | 4577 | static int __init init_per_zone_wmark_min(void) |
1da177e4 LT |
4578 | { |
4579 | unsigned long lowmem_kbytes; | |
4580 | ||
4581 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4582 | ||
4583 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4584 | if (min_free_kbytes < 128) | |
4585 | min_free_kbytes = 128; | |
4586 | if (min_free_kbytes > 65536) | |
4587 | min_free_kbytes = 65536; | |
bc75d33f | 4588 | setup_per_zone_wmarks(); |
1da177e4 | 4589 | setup_per_zone_lowmem_reserve(); |
556adecb | 4590 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
4591 | return 0; |
4592 | } | |
bc75d33f | 4593 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
4594 | |
4595 | /* | |
4596 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4597 | * that we can call two helper functions whenever min_free_kbytes | |
4598 | * changes. | |
4599 | */ | |
4600 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4601 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4602 | { | |
4603 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 | 4604 | if (write) |
bc75d33f | 4605 | setup_per_zone_wmarks(); |
1da177e4 LT |
4606 | return 0; |
4607 | } | |
4608 | ||
9614634f CL |
4609 | #ifdef CONFIG_NUMA |
4610 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4611 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4612 | { | |
4613 | struct zone *zone; | |
4614 | int rc; | |
4615 | ||
4616 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4617 | if (rc) | |
4618 | return rc; | |
4619 | ||
4620 | for_each_zone(zone) | |
8417bba4 | 4621 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4622 | sysctl_min_unmapped_ratio) / 100; |
4623 | return 0; | |
4624 | } | |
0ff38490 CL |
4625 | |
4626 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4627 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4628 | { | |
4629 | struct zone *zone; | |
4630 | int rc; | |
4631 | ||
4632 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4633 | if (rc) | |
4634 | return rc; | |
4635 | ||
4636 | for_each_zone(zone) | |
4637 | zone->min_slab_pages = (zone->present_pages * | |
4638 | sysctl_min_slab_ratio) / 100; | |
4639 | return 0; | |
4640 | } | |
9614634f CL |
4641 | #endif |
4642 | ||
1da177e4 LT |
4643 | /* |
4644 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4645 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4646 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4647 | * | |
4648 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 4649 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
4650 | * if in function of the boot time zone sizes. |
4651 | */ | |
4652 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4653 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4654 | { | |
4655 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4656 | setup_per_zone_lowmem_reserve(); | |
4657 | return 0; | |
4658 | } | |
4659 | ||
8ad4b1fb RS |
4660 | /* |
4661 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4662 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4663 | * can have before it gets flushed back to buddy allocator. | |
4664 | */ | |
4665 | ||
4666 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4667 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4668 | { | |
4669 | struct zone *zone; | |
4670 | unsigned int cpu; | |
4671 | int ret; | |
4672 | ||
4673 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4674 | if (!write || (ret == -EINVAL)) | |
4675 | return ret; | |
364df0eb | 4676 | for_each_populated_zone(zone) { |
8ad4b1fb RS |
4677 | for_each_online_cpu(cpu) { |
4678 | unsigned long high; | |
4679 | high = zone->present_pages / percpu_pagelist_fraction; | |
4680 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4681 | } | |
4682 | } | |
4683 | return 0; | |
4684 | } | |
4685 | ||
f034b5d4 | 4686 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4687 | |
4688 | #ifdef CONFIG_NUMA | |
4689 | static int __init set_hashdist(char *str) | |
4690 | { | |
4691 | if (!str) | |
4692 | return 0; | |
4693 | hashdist = simple_strtoul(str, &str, 0); | |
4694 | return 1; | |
4695 | } | |
4696 | __setup("hashdist=", set_hashdist); | |
4697 | #endif | |
4698 | ||
4699 | /* | |
4700 | * allocate a large system hash table from bootmem | |
4701 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4702 | * quantity of entries | |
4703 | * - limit is the number of hash buckets, not the total allocation size | |
4704 | */ | |
4705 | void *__init alloc_large_system_hash(const char *tablename, | |
4706 | unsigned long bucketsize, | |
4707 | unsigned long numentries, | |
4708 | int scale, | |
4709 | int flags, | |
4710 | unsigned int *_hash_shift, | |
4711 | unsigned int *_hash_mask, | |
4712 | unsigned long limit) | |
4713 | { | |
4714 | unsigned long long max = limit; | |
4715 | unsigned long log2qty, size; | |
4716 | void *table = NULL; | |
4717 | ||
4718 | /* allow the kernel cmdline to have a say */ | |
4719 | if (!numentries) { | |
4720 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4721 | numentries = nr_kernel_pages; |
1da177e4 LT |
4722 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4723 | numentries >>= 20 - PAGE_SHIFT; | |
4724 | numentries <<= 20 - PAGE_SHIFT; | |
4725 | ||
4726 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4727 | if (scale > PAGE_SHIFT) | |
4728 | numentries >>= (scale - PAGE_SHIFT); | |
4729 | else | |
4730 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4731 | |
4732 | /* Make sure we've got at least a 0-order allocation.. */ | |
4733 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4734 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4735 | } |
6e692ed3 | 4736 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4737 | |
4738 | /* limit allocation size to 1/16 total memory by default */ | |
4739 | if (max == 0) { | |
4740 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4741 | do_div(max, bucketsize); | |
4742 | } | |
4743 | ||
4744 | if (numentries > max) | |
4745 | numentries = max; | |
4746 | ||
f0d1b0b3 | 4747 | log2qty = ilog2(numentries); |
1da177e4 LT |
4748 | |
4749 | do { | |
4750 | size = bucketsize << log2qty; | |
4751 | if (flags & HASH_EARLY) | |
74768ed8 | 4752 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
4753 | else if (hashdist) |
4754 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4755 | else { | |
1037b83b ED |
4756 | /* |
4757 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
4758 | * some pages at the end of hash table which |
4759 | * alloc_pages_exact() automatically does | |
1037b83b | 4760 | */ |
264ef8a9 | 4761 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 4762 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
4763 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
4764 | } | |
1da177e4 LT |
4765 | } |
4766 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4767 | ||
4768 | if (!table) | |
4769 | panic("Failed to allocate %s hash table\n", tablename); | |
4770 | ||
b49ad484 | 4771 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4772 | tablename, |
4773 | (1U << log2qty), | |
f0d1b0b3 | 4774 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4775 | size); |
4776 | ||
4777 | if (_hash_shift) | |
4778 | *_hash_shift = log2qty; | |
4779 | if (_hash_mask) | |
4780 | *_hash_mask = (1 << log2qty) - 1; | |
4781 | ||
4782 | return table; | |
4783 | } | |
a117e66e | 4784 | |
835c134e MG |
4785 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4786 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4787 | unsigned long pfn) | |
4788 | { | |
4789 | #ifdef CONFIG_SPARSEMEM | |
4790 | return __pfn_to_section(pfn)->pageblock_flags; | |
4791 | #else | |
4792 | return zone->pageblock_flags; | |
4793 | #endif /* CONFIG_SPARSEMEM */ | |
4794 | } | |
4795 | ||
4796 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4797 | { | |
4798 | #ifdef CONFIG_SPARSEMEM | |
4799 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4800 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4801 | #else |
4802 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4803 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4804 | #endif /* CONFIG_SPARSEMEM */ |
4805 | } | |
4806 | ||
4807 | /** | |
d9c23400 | 4808 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4809 | * @page: The page within the block of interest |
4810 | * @start_bitidx: The first bit of interest to retrieve | |
4811 | * @end_bitidx: The last bit of interest | |
4812 | * returns pageblock_bits flags | |
4813 | */ | |
4814 | unsigned long get_pageblock_flags_group(struct page *page, | |
4815 | int start_bitidx, int end_bitidx) | |
4816 | { | |
4817 | struct zone *zone; | |
4818 | unsigned long *bitmap; | |
4819 | unsigned long pfn, bitidx; | |
4820 | unsigned long flags = 0; | |
4821 | unsigned long value = 1; | |
4822 | ||
4823 | zone = page_zone(page); | |
4824 | pfn = page_to_pfn(page); | |
4825 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4826 | bitidx = pfn_to_bitidx(zone, pfn); | |
4827 | ||
4828 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4829 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4830 | flags |= value; | |
6220ec78 | 4831 | |
835c134e MG |
4832 | return flags; |
4833 | } | |
4834 | ||
4835 | /** | |
d9c23400 | 4836 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4837 | * @page: The page within the block of interest |
4838 | * @start_bitidx: The first bit of interest | |
4839 | * @end_bitidx: The last bit of interest | |
4840 | * @flags: The flags to set | |
4841 | */ | |
4842 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4843 | int start_bitidx, int end_bitidx) | |
4844 | { | |
4845 | struct zone *zone; | |
4846 | unsigned long *bitmap; | |
4847 | unsigned long pfn, bitidx; | |
4848 | unsigned long value = 1; | |
4849 | ||
4850 | zone = page_zone(page); | |
4851 | pfn = page_to_pfn(page); | |
4852 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4853 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
4854 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
4855 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
4856 | |
4857 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4858 | if (flags & value) | |
4859 | __set_bit(bitidx + start_bitidx, bitmap); | |
4860 | else | |
4861 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4862 | } | |
a5d76b54 KH |
4863 | |
4864 | /* | |
4865 | * This is designed as sub function...plz see page_isolation.c also. | |
4866 | * set/clear page block's type to be ISOLATE. | |
4867 | * page allocater never alloc memory from ISOLATE block. | |
4868 | */ | |
4869 | ||
4870 | int set_migratetype_isolate(struct page *page) | |
4871 | { | |
4872 | struct zone *zone; | |
4873 | unsigned long flags; | |
4874 | int ret = -EBUSY; | |
4875 | ||
4876 | zone = page_zone(page); | |
4877 | spin_lock_irqsave(&zone->lock, flags); | |
4878 | /* | |
4879 | * In future, more migrate types will be able to be isolation target. | |
4880 | */ | |
4881 | if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE) | |
4882 | goto out; | |
4883 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
4884 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
4885 | ret = 0; | |
4886 | out: | |
4887 | spin_unlock_irqrestore(&zone->lock, flags); | |
4888 | if (!ret) | |
9f8f2172 | 4889 | drain_all_pages(); |
a5d76b54 KH |
4890 | return ret; |
4891 | } | |
4892 | ||
4893 | void unset_migratetype_isolate(struct page *page) | |
4894 | { | |
4895 | struct zone *zone; | |
4896 | unsigned long flags; | |
4897 | zone = page_zone(page); | |
4898 | spin_lock_irqsave(&zone->lock, flags); | |
4899 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
4900 | goto out; | |
4901 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4902 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4903 | out: | |
4904 | spin_unlock_irqrestore(&zone->lock, flags); | |
4905 | } | |
0c0e6195 KH |
4906 | |
4907 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
4908 | /* | |
4909 | * All pages in the range must be isolated before calling this. | |
4910 | */ | |
4911 | void | |
4912 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
4913 | { | |
4914 | struct page *page; | |
4915 | struct zone *zone; | |
4916 | int order, i; | |
4917 | unsigned long pfn; | |
4918 | unsigned long flags; | |
4919 | /* find the first valid pfn */ | |
4920 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
4921 | if (pfn_valid(pfn)) | |
4922 | break; | |
4923 | if (pfn == end_pfn) | |
4924 | return; | |
4925 | zone = page_zone(pfn_to_page(pfn)); | |
4926 | spin_lock_irqsave(&zone->lock, flags); | |
4927 | pfn = start_pfn; | |
4928 | while (pfn < end_pfn) { | |
4929 | if (!pfn_valid(pfn)) { | |
4930 | pfn++; | |
4931 | continue; | |
4932 | } | |
4933 | page = pfn_to_page(pfn); | |
4934 | BUG_ON(page_count(page)); | |
4935 | BUG_ON(!PageBuddy(page)); | |
4936 | order = page_order(page); | |
4937 | #ifdef CONFIG_DEBUG_VM | |
4938 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
4939 | pfn, 1 << order, end_pfn); | |
4940 | #endif | |
4941 | list_del(&page->lru); | |
4942 | rmv_page_order(page); | |
4943 | zone->free_area[order].nr_free--; | |
4944 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
4945 | - (1UL << order)); | |
4946 | for (i = 0; i < (1 << order); i++) | |
4947 | SetPageReserved((page+i)); | |
4948 | pfn += (1 << order); | |
4949 | } | |
4950 | spin_unlock_irqrestore(&zone->lock, flags); | |
4951 | } | |
4952 | #endif |