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> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/compiler.h> | |
9f158333 | 24 | #include <linux/kernel.h> |
1da177e4 LT |
25 | #include <linux/module.h> |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/topology.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
bdc8cb98 | 35 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
36 | #include <linux/nodemask.h> |
37 | #include <linux/vmalloc.h> | |
4be38e35 | 38 | #include <linux/mempolicy.h> |
6811378e | 39 | #include <linux/stop_machine.h> |
c713216d MG |
40 | #include <linux/sort.h> |
41 | #include <linux/pfn.h> | |
3fcfab16 | 42 | #include <linux/backing-dev.h> |
933e312e | 43 | #include <linux/fault-inject.h> |
1da177e4 LT |
44 | |
45 | #include <asm/tlbflush.h> | |
ac924c60 | 46 | #include <asm/div64.h> |
1da177e4 LT |
47 | #include "internal.h" |
48 | ||
49 | /* | |
13808910 | 50 | * Array of node states. |
1da177e4 | 51 | */ |
13808910 CL |
52 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
53 | [N_POSSIBLE] = NODE_MASK_ALL, | |
54 | [N_ONLINE] = { { [0] = 1UL } }, | |
55 | #ifndef CONFIG_NUMA | |
56 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
57 | #ifdef CONFIG_HIGHMEM | |
58 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
59 | #endif | |
60 | [N_CPU] = { { [0] = 1UL } }, | |
61 | #endif /* NUMA */ | |
62 | }; | |
63 | EXPORT_SYMBOL(node_states); | |
64 | ||
6c231b7b | 65 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 66 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 67 | long nr_swap_pages; |
8ad4b1fb | 68 | int percpu_pagelist_fraction; |
1da177e4 | 69 | |
d98c7a09 | 70 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 71 | |
1da177e4 LT |
72 | /* |
73 | * results with 256, 32 in the lowmem_reserve sysctl: | |
74 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
75 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
76 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
77 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
78 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
79 | * |
80 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
81 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 82 | */ |
2f1b6248 | 83 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 84 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 85 | 256, |
4b51d669 | 86 | #endif |
fb0e7942 | 87 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 88 | 256, |
fb0e7942 | 89 | #endif |
e53ef38d | 90 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 91 | 32, |
e53ef38d | 92 | #endif |
2a1e274a | 93 | 32, |
2f1b6248 | 94 | }; |
1da177e4 LT |
95 | |
96 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 97 | |
15ad7cdc | 98 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 99 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 100 | "DMA", |
4b51d669 | 101 | #endif |
fb0e7942 | 102 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 103 | "DMA32", |
fb0e7942 | 104 | #endif |
2f1b6248 | 105 | "Normal", |
e53ef38d | 106 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 107 | "HighMem", |
e53ef38d | 108 | #endif |
2a1e274a | 109 | "Movable", |
2f1b6248 CL |
110 | }; |
111 | ||
1da177e4 LT |
112 | int min_free_kbytes = 1024; |
113 | ||
86356ab1 YG |
114 | unsigned long __meminitdata nr_kernel_pages; |
115 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 116 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 117 | |
c713216d MG |
118 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
119 | /* | |
120 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
121 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
122 | * Ranges passed to add_active_range() will be merged if possible | |
123 | * so the number of times add_active_range() can be called is | |
124 | * related to the number of nodes and the number of holes | |
125 | */ | |
126 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
127 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
128 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
129 | #else | |
130 | #if MAX_NUMNODES >= 32 | |
131 | /* If there can be many nodes, allow up to 50 holes per node */ | |
132 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
133 | #else | |
134 | /* By default, allow up to 256 distinct regions */ | |
135 | #define MAX_ACTIVE_REGIONS 256 | |
136 | #endif | |
137 | #endif | |
138 | ||
98011f56 JB |
139 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
140 | static int __meminitdata nr_nodemap_entries; | |
141 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
142 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c | 143 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
98011f56 JB |
144 | static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES]; |
145 | static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES]; | |
fb01439c | 146 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ |
2a1e274a | 147 | unsigned long __initdata required_kernelcore; |
7e63efef | 148 | unsigned long __initdata required_movablecore; |
e228929b | 149 | unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
150 | |
151 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
152 | int movable_zone; | |
153 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
154 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
155 | ||
418508c1 MS |
156 | #if MAX_NUMNODES > 1 |
157 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
158 | EXPORT_SYMBOL(nr_node_ids); | |
159 | #endif | |
160 | ||
b2a0ac88 MG |
161 | static inline int get_pageblock_migratetype(struct page *page) |
162 | { | |
163 | return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end); | |
164 | } | |
165 | ||
166 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
167 | { | |
168 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
169 | PB_migrate, PB_migrate_end); | |
170 | } | |
171 | ||
172 | static inline int gfpflags_to_migratetype(gfp_t gfp_flags) | |
173 | { | |
174 | return ((gfp_flags & __GFP_MOVABLE) != 0); | |
175 | } | |
176 | ||
13e7444b | 177 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 178 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 179 | { |
bdc8cb98 DH |
180 | int ret = 0; |
181 | unsigned seq; | |
182 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 183 | |
bdc8cb98 DH |
184 | do { |
185 | seq = zone_span_seqbegin(zone); | |
186 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
187 | ret = 1; | |
188 | else if (pfn < zone->zone_start_pfn) | |
189 | ret = 1; | |
190 | } while (zone_span_seqretry(zone, seq)); | |
191 | ||
192 | return ret; | |
c6a57e19 DH |
193 | } |
194 | ||
195 | static int page_is_consistent(struct zone *zone, struct page *page) | |
196 | { | |
14e07298 | 197 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 198 | return 0; |
1da177e4 | 199 | if (zone != page_zone(page)) |
c6a57e19 DH |
200 | return 0; |
201 | ||
202 | return 1; | |
203 | } | |
204 | /* | |
205 | * Temporary debugging check for pages not lying within a given zone. | |
206 | */ | |
207 | static int bad_range(struct zone *zone, struct page *page) | |
208 | { | |
209 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 210 | return 1; |
c6a57e19 DH |
211 | if (!page_is_consistent(zone, page)) |
212 | return 1; | |
213 | ||
1da177e4 LT |
214 | return 0; |
215 | } | |
13e7444b NP |
216 | #else |
217 | static inline int bad_range(struct zone *zone, struct page *page) | |
218 | { | |
219 | return 0; | |
220 | } | |
221 | #endif | |
222 | ||
224abf92 | 223 | static void bad_page(struct page *page) |
1da177e4 | 224 | { |
224abf92 | 225 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
226 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
227 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
228 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
229 | current->comm, page, (int)(2*sizeof(unsigned long)), |
230 | (unsigned long)page->flags, page->mapping, | |
231 | page_mapcount(page), page_count(page)); | |
1da177e4 | 232 | dump_stack(); |
334795ec HD |
233 | page->flags &= ~(1 << PG_lru | |
234 | 1 << PG_private | | |
1da177e4 | 235 | 1 << PG_locked | |
1da177e4 LT |
236 | 1 << PG_active | |
237 | 1 << PG_dirty | | |
334795ec HD |
238 | 1 << PG_reclaim | |
239 | 1 << PG_slab | | |
1da177e4 | 240 | 1 << PG_swapcache | |
676165a8 NP |
241 | 1 << PG_writeback | |
242 | 1 << PG_buddy ); | |
1da177e4 LT |
243 | set_page_count(page, 0); |
244 | reset_page_mapcount(page); | |
245 | page->mapping = NULL; | |
9f158333 | 246 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
247 | } |
248 | ||
1da177e4 LT |
249 | /* |
250 | * Higher-order pages are called "compound pages". They are structured thusly: | |
251 | * | |
252 | * The first PAGE_SIZE page is called the "head page". | |
253 | * | |
254 | * The remaining PAGE_SIZE pages are called "tail pages". | |
255 | * | |
256 | * All pages have PG_compound set. All pages have their ->private pointing at | |
257 | * the head page (even the head page has this). | |
258 | * | |
41d78ba5 HD |
259 | * The first tail page's ->lru.next holds the address of the compound page's |
260 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
261 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 262 | */ |
d98c7a09 HD |
263 | |
264 | static void free_compound_page(struct page *page) | |
265 | { | |
d85f3385 | 266 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
267 | } |
268 | ||
1da177e4 LT |
269 | static void prep_compound_page(struct page *page, unsigned long order) |
270 | { | |
271 | int i; | |
272 | int nr_pages = 1 << order; | |
273 | ||
33f2ef89 | 274 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 275 | set_compound_order(page, order); |
6d777953 | 276 | __SetPageHead(page); |
d85f3385 | 277 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
278 | struct page *p = page + i; |
279 | ||
d85f3385 | 280 | __SetPageTail(p); |
d85f3385 | 281 | p->first_page = page; |
1da177e4 LT |
282 | } |
283 | } | |
284 | ||
285 | static void destroy_compound_page(struct page *page, unsigned long order) | |
286 | { | |
287 | int i; | |
288 | int nr_pages = 1 << order; | |
289 | ||
d85f3385 | 290 | if (unlikely(compound_order(page) != order)) |
224abf92 | 291 | bad_page(page); |
1da177e4 | 292 | |
6d777953 | 293 | if (unlikely(!PageHead(page))) |
d85f3385 | 294 | bad_page(page); |
6d777953 | 295 | __ClearPageHead(page); |
d85f3385 | 296 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
297 | struct page *p = page + i; |
298 | ||
6d777953 | 299 | if (unlikely(!PageTail(p) | |
d85f3385 | 300 | (p->first_page != page))) |
224abf92 | 301 | bad_page(page); |
d85f3385 | 302 | __ClearPageTail(p); |
1da177e4 LT |
303 | } |
304 | } | |
1da177e4 | 305 | |
17cf4406 NP |
306 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
307 | { | |
308 | int i; | |
309 | ||
725d704e | 310 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
311 | /* |
312 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
313 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
314 | */ | |
725d704e | 315 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
316 | for (i = 0; i < (1 << order); i++) |
317 | clear_highpage(page + i); | |
318 | } | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * function for dealing with page's order in buddy system. | |
322 | * zone->lock is already acquired when we use these. | |
323 | * So, we don't need atomic page->flags operations here. | |
324 | */ | |
6aa3001b AM |
325 | static inline unsigned long page_order(struct page *page) |
326 | { | |
4c21e2f2 | 327 | return page_private(page); |
1da177e4 LT |
328 | } |
329 | ||
6aa3001b AM |
330 | static inline void set_page_order(struct page *page, int order) |
331 | { | |
4c21e2f2 | 332 | set_page_private(page, order); |
676165a8 | 333 | __SetPageBuddy(page); |
1da177e4 LT |
334 | } |
335 | ||
336 | static inline void rmv_page_order(struct page *page) | |
337 | { | |
676165a8 | 338 | __ClearPageBuddy(page); |
4c21e2f2 | 339 | set_page_private(page, 0); |
1da177e4 LT |
340 | } |
341 | ||
342 | /* | |
343 | * Locate the struct page for both the matching buddy in our | |
344 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
345 | * | |
346 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
347 | * the following equation: | |
348 | * B2 = B1 ^ (1 << O) | |
349 | * For example, if the starting buddy (buddy2) is #8 its order | |
350 | * 1 buddy is #10: | |
351 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
352 | * | |
353 | * 2) Any buddy B will have an order O+1 parent P which | |
354 | * satisfies the following equation: | |
355 | * P = B & ~(1 << O) | |
356 | * | |
d6e05edc | 357 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
358 | */ |
359 | static inline struct page * | |
360 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
361 | { | |
362 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
363 | ||
364 | return page + (buddy_idx - page_idx); | |
365 | } | |
366 | ||
367 | static inline unsigned long | |
368 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
369 | { | |
370 | return (page_idx & ~(1 << order)); | |
371 | } | |
372 | ||
373 | /* | |
374 | * This function checks whether a page is free && is the buddy | |
375 | * we can do coalesce a page and its buddy if | |
13e7444b | 376 | * (a) the buddy is not in a hole && |
676165a8 | 377 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
378 | * (c) a page and its buddy have the same order && |
379 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
380 | * |
381 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
382 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 383 | * |
676165a8 | 384 | * For recording page's order, we use page_private(page). |
1da177e4 | 385 | */ |
cb2b95e1 AW |
386 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
387 | int order) | |
1da177e4 | 388 | { |
14e07298 | 389 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 390 | return 0; |
13e7444b | 391 | |
cb2b95e1 AW |
392 | if (page_zone_id(page) != page_zone_id(buddy)) |
393 | return 0; | |
394 | ||
395 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
396 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 397 | return 1; |
676165a8 | 398 | } |
6aa3001b | 399 | return 0; |
1da177e4 LT |
400 | } |
401 | ||
402 | /* | |
403 | * Freeing function for a buddy system allocator. | |
404 | * | |
405 | * The concept of a buddy system is to maintain direct-mapped table | |
406 | * (containing bit values) for memory blocks of various "orders". | |
407 | * The bottom level table contains the map for the smallest allocatable | |
408 | * units of memory (here, pages), and each level above it describes | |
409 | * pairs of units from the levels below, hence, "buddies". | |
410 | * At a high level, all that happens here is marking the table entry | |
411 | * at the bottom level available, and propagating the changes upward | |
412 | * as necessary, plus some accounting needed to play nicely with other | |
413 | * parts of the VM system. | |
414 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 415 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 416 | * order is recorded in page_private(page) field. |
1da177e4 LT |
417 | * So when we are allocating or freeing one, we can derive the state of the |
418 | * other. That is, if we allocate a small block, and both were | |
419 | * free, the remainder of the region must be split into blocks. | |
420 | * If a block is freed, and its buddy is also free, then this | |
421 | * triggers coalescing into a block of larger size. | |
422 | * | |
423 | * -- wli | |
424 | */ | |
425 | ||
48db57f8 | 426 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
427 | struct zone *zone, unsigned int order) |
428 | { | |
429 | unsigned long page_idx; | |
430 | int order_size = 1 << order; | |
b2a0ac88 | 431 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 432 | |
224abf92 | 433 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
434 | destroy_compound_page(page, order); |
435 | ||
436 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
437 | ||
725d704e NP |
438 | VM_BUG_ON(page_idx & (order_size - 1)); |
439 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 440 | |
d23ad423 | 441 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
442 | while (order < MAX_ORDER-1) { |
443 | unsigned long combined_idx; | |
1da177e4 LT |
444 | struct page *buddy; |
445 | ||
1da177e4 | 446 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 447 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 448 | break; /* Move the buddy up one level. */ |
13e7444b | 449 | |
1da177e4 | 450 | list_del(&buddy->lru); |
b2a0ac88 | 451 | zone->free_area[order].nr_free--; |
1da177e4 | 452 | rmv_page_order(buddy); |
13e7444b | 453 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
454 | page = page + (combined_idx - page_idx); |
455 | page_idx = combined_idx; | |
456 | order++; | |
457 | } | |
458 | set_page_order(page, order); | |
b2a0ac88 MG |
459 | list_add(&page->lru, |
460 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
461 | zone->free_area[order].nr_free++; |
462 | } | |
463 | ||
224abf92 | 464 | static inline int free_pages_check(struct page *page) |
1da177e4 | 465 | { |
92be2e33 NP |
466 | if (unlikely(page_mapcount(page) | |
467 | (page->mapping != NULL) | | |
468 | (page_count(page) != 0) | | |
1da177e4 LT |
469 | (page->flags & ( |
470 | 1 << PG_lru | | |
471 | 1 << PG_private | | |
472 | 1 << PG_locked | | |
473 | 1 << PG_active | | |
1da177e4 LT |
474 | 1 << PG_slab | |
475 | 1 << PG_swapcache | | |
b5810039 | 476 | 1 << PG_writeback | |
676165a8 NP |
477 | 1 << PG_reserved | |
478 | 1 << PG_buddy )))) | |
224abf92 | 479 | bad_page(page); |
1da177e4 | 480 | if (PageDirty(page)) |
242e5468 | 481 | __ClearPageDirty(page); |
689bcebf HD |
482 | /* |
483 | * For now, we report if PG_reserved was found set, but do not | |
484 | * clear it, and do not free the page. But we shall soon need | |
485 | * to do more, for when the ZERO_PAGE count wraps negative. | |
486 | */ | |
487 | return PageReserved(page); | |
1da177e4 LT |
488 | } |
489 | ||
490 | /* | |
491 | * Frees a list of pages. | |
492 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 493 | * count is the number of pages to free. |
1da177e4 LT |
494 | * |
495 | * If the zone was previously in an "all pages pinned" state then look to | |
496 | * see if this freeing clears that state. | |
497 | * | |
498 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
499 | * pinned" detection logic. | |
500 | */ | |
48db57f8 NP |
501 | static void free_pages_bulk(struct zone *zone, int count, |
502 | struct list_head *list, int order) | |
1da177e4 | 503 | { |
c54ad30c | 504 | spin_lock(&zone->lock); |
1da177e4 LT |
505 | zone->all_unreclaimable = 0; |
506 | zone->pages_scanned = 0; | |
48db57f8 NP |
507 | while (count--) { |
508 | struct page *page; | |
509 | ||
725d704e | 510 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 511 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 512 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 513 | list_del(&page->lru); |
48db57f8 | 514 | __free_one_page(page, zone, order); |
1da177e4 | 515 | } |
c54ad30c | 516 | spin_unlock(&zone->lock); |
1da177e4 LT |
517 | } |
518 | ||
48db57f8 | 519 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 520 | { |
006d22d9 CL |
521 | spin_lock(&zone->lock); |
522 | zone->all_unreclaimable = 0; | |
523 | zone->pages_scanned = 0; | |
0798e519 | 524 | __free_one_page(page, zone, order); |
006d22d9 | 525 | spin_unlock(&zone->lock); |
48db57f8 NP |
526 | } |
527 | ||
528 | static void __free_pages_ok(struct page *page, unsigned int order) | |
529 | { | |
530 | unsigned long flags; | |
1da177e4 | 531 | int i; |
689bcebf | 532 | int reserved = 0; |
1da177e4 | 533 | |
1da177e4 | 534 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 535 | reserved += free_pages_check(page + i); |
689bcebf HD |
536 | if (reserved) |
537 | return; | |
538 | ||
9858db50 NP |
539 | if (!PageHighMem(page)) |
540 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 541 | arch_free_page(page, order); |
48db57f8 | 542 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 543 | |
c54ad30c | 544 | local_irq_save(flags); |
f8891e5e | 545 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 546 | free_one_page(page_zone(page), page, order); |
c54ad30c | 547 | local_irq_restore(flags); |
1da177e4 LT |
548 | } |
549 | ||
a226f6c8 DH |
550 | /* |
551 | * permit the bootmem allocator to evade page validation on high-order frees | |
552 | */ | |
553 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
554 | { | |
555 | if (order == 0) { | |
556 | __ClearPageReserved(page); | |
557 | set_page_count(page, 0); | |
7835e98b | 558 | set_page_refcounted(page); |
545b1ea9 | 559 | __free_page(page); |
a226f6c8 | 560 | } else { |
a226f6c8 DH |
561 | int loop; |
562 | ||
545b1ea9 | 563 | prefetchw(page); |
a226f6c8 DH |
564 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
565 | struct page *p = &page[loop]; | |
566 | ||
545b1ea9 NP |
567 | if (loop + 1 < BITS_PER_LONG) |
568 | prefetchw(p + 1); | |
a226f6c8 DH |
569 | __ClearPageReserved(p); |
570 | set_page_count(p, 0); | |
571 | } | |
572 | ||
7835e98b | 573 | set_page_refcounted(page); |
545b1ea9 | 574 | __free_pages(page, order); |
a226f6c8 DH |
575 | } |
576 | } | |
577 | ||
1da177e4 LT |
578 | |
579 | /* | |
580 | * The order of subdivision here is critical for the IO subsystem. | |
581 | * Please do not alter this order without good reasons and regression | |
582 | * testing. Specifically, as large blocks of memory are subdivided, | |
583 | * the order in which smaller blocks are delivered depends on the order | |
584 | * they're subdivided in this function. This is the primary factor | |
585 | * influencing the order in which pages are delivered to the IO | |
586 | * subsystem according to empirical testing, and this is also justified | |
587 | * by considering the behavior of a buddy system containing a single | |
588 | * large block of memory acted on by a series of small allocations. | |
589 | * This behavior is a critical factor in sglist merging's success. | |
590 | * | |
591 | * -- wli | |
592 | */ | |
085cc7d5 | 593 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
594 | int low, int high, struct free_area *area, |
595 | int migratetype) | |
1da177e4 LT |
596 | { |
597 | unsigned long size = 1 << high; | |
598 | ||
599 | while (high > low) { | |
600 | area--; | |
601 | high--; | |
602 | size >>= 1; | |
725d704e | 603 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 604 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
605 | area->nr_free++; |
606 | set_page_order(&page[size], high); | |
607 | } | |
1da177e4 LT |
608 | } |
609 | ||
1da177e4 LT |
610 | /* |
611 | * This page is about to be returned from the page allocator | |
612 | */ | |
17cf4406 | 613 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 614 | { |
92be2e33 NP |
615 | if (unlikely(page_mapcount(page) | |
616 | (page->mapping != NULL) | | |
617 | (page_count(page) != 0) | | |
334795ec HD |
618 | (page->flags & ( |
619 | 1 << PG_lru | | |
1da177e4 LT |
620 | 1 << PG_private | |
621 | 1 << PG_locked | | |
1da177e4 LT |
622 | 1 << PG_active | |
623 | 1 << PG_dirty | | |
334795ec | 624 | 1 << PG_slab | |
1da177e4 | 625 | 1 << PG_swapcache | |
b5810039 | 626 | 1 << PG_writeback | |
676165a8 NP |
627 | 1 << PG_reserved | |
628 | 1 << PG_buddy )))) | |
224abf92 | 629 | bad_page(page); |
1da177e4 | 630 | |
689bcebf HD |
631 | /* |
632 | * For now, we report if PG_reserved was found set, but do not | |
633 | * clear it, and do not allocate the page: as a safety net. | |
634 | */ | |
635 | if (PageReserved(page)) | |
636 | return 1; | |
637 | ||
d77c2d7c | 638 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead | |
1da177e4 | 639 | 1 << PG_referenced | 1 << PG_arch_1 | |
5409bae0 | 640 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 641 | set_page_private(page, 0); |
7835e98b | 642 | set_page_refcounted(page); |
cc102509 NP |
643 | |
644 | arch_alloc_page(page, order); | |
1da177e4 | 645 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
646 | |
647 | if (gfp_flags & __GFP_ZERO) | |
648 | prep_zero_page(page, order, gfp_flags); | |
649 | ||
650 | if (order && (gfp_flags & __GFP_COMP)) | |
651 | prep_compound_page(page, order); | |
652 | ||
689bcebf | 653 | return 0; |
1da177e4 LT |
654 | } |
655 | ||
b2a0ac88 MG |
656 | /* |
657 | * This array describes the order lists are fallen back to when | |
658 | * the free lists for the desirable migrate type are depleted | |
659 | */ | |
660 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
661 | [MIGRATE_UNMOVABLE] = { MIGRATE_MOVABLE }, | |
662 | [MIGRATE_MOVABLE] = { MIGRATE_UNMOVABLE }, | |
663 | }; | |
664 | ||
665 | /* Remove an element from the buddy allocator from the fallback list */ | |
666 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
667 | int start_migratetype) | |
668 | { | |
669 | struct free_area * area; | |
670 | int current_order; | |
671 | struct page *page; | |
672 | int migratetype, i; | |
673 | ||
674 | /* Find the largest possible block of pages in the other list */ | |
675 | for (current_order = MAX_ORDER-1; current_order >= order; | |
676 | --current_order) { | |
677 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
678 | migratetype = fallbacks[start_migratetype][i]; | |
679 | ||
680 | area = &(zone->free_area[current_order]); | |
681 | if (list_empty(&area->free_list[migratetype])) | |
682 | continue; | |
683 | ||
684 | page = list_entry(area->free_list[migratetype].next, | |
685 | struct page, lru); | |
686 | area->nr_free--; | |
687 | ||
688 | /* | |
689 | * If breaking a large block of pages, place the buddies | |
690 | * on the preferred allocation list | |
691 | */ | |
692 | if (unlikely(current_order >= MAX_ORDER / 2)) | |
693 | migratetype = start_migratetype; | |
694 | ||
695 | /* Remove the page from the freelists */ | |
696 | list_del(&page->lru); | |
697 | rmv_page_order(page); | |
698 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
699 | -(1UL << order)); | |
700 | ||
701 | if (current_order == MAX_ORDER - 1) | |
702 | set_pageblock_migratetype(page, | |
703 | start_migratetype); | |
704 | ||
705 | expand(zone, page, order, current_order, area, migratetype); | |
706 | return page; | |
707 | } | |
708 | } | |
709 | ||
710 | return NULL; | |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * Do the hard work of removing an element from the buddy allocator. | |
715 | * Call me with the zone->lock already held. | |
716 | */ | |
b2a0ac88 MG |
717 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
718 | int migratetype) | |
1da177e4 LT |
719 | { |
720 | struct free_area * area; | |
721 | unsigned int current_order; | |
722 | struct page *page; | |
723 | ||
b2a0ac88 | 724 | /* Find a page of the appropriate size in the preferred list */ |
1da177e4 | 725 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
b2a0ac88 MG |
726 | area = &(zone->free_area[current_order]); |
727 | if (list_empty(&area->free_list[migratetype])) | |
1da177e4 LT |
728 | continue; |
729 | ||
b2a0ac88 MG |
730 | page = list_entry(area->free_list[migratetype].next, |
731 | struct page, lru); | |
1da177e4 LT |
732 | list_del(&page->lru); |
733 | rmv_page_order(page); | |
734 | area->nr_free--; | |
d23ad423 | 735 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); |
b2a0ac88 MG |
736 | expand(zone, page, order, current_order, area, migratetype); |
737 | goto got_page; | |
1da177e4 LT |
738 | } |
739 | ||
b2a0ac88 MG |
740 | page = __rmqueue_fallback(zone, order, migratetype); |
741 | ||
742 | got_page: | |
743 | ||
744 | return page; | |
1da177e4 LT |
745 | } |
746 | ||
747 | /* | |
748 | * Obtain a specified number of elements from the buddy allocator, all under | |
749 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
750 | * Returns the number of new pages which were placed at *list. | |
751 | */ | |
752 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
753 | unsigned long count, struct list_head *list, |
754 | int migratetype) | |
1da177e4 | 755 | { |
1da177e4 | 756 | int i; |
1da177e4 | 757 | |
c54ad30c | 758 | spin_lock(&zone->lock); |
1da177e4 | 759 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 760 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 761 | if (unlikely(page == NULL)) |
1da177e4 | 762 | break; |
1da177e4 LT |
763 | list_add_tail(&page->lru, list); |
764 | } | |
c54ad30c | 765 | spin_unlock(&zone->lock); |
085cc7d5 | 766 | return i; |
1da177e4 LT |
767 | } |
768 | ||
4ae7c039 | 769 | #ifdef CONFIG_NUMA |
8fce4d8e | 770 | /* |
4037d452 CL |
771 | * Called from the vmstat counter updater to drain pagesets of this |
772 | * currently executing processor on remote nodes after they have | |
773 | * expired. | |
774 | * | |
879336c3 CL |
775 | * Note that this function must be called with the thread pinned to |
776 | * a single processor. | |
8fce4d8e | 777 | */ |
4037d452 | 778 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 779 | { |
4ae7c039 | 780 | unsigned long flags; |
4037d452 | 781 | int to_drain; |
4ae7c039 | 782 | |
4037d452 CL |
783 | local_irq_save(flags); |
784 | if (pcp->count >= pcp->batch) | |
785 | to_drain = pcp->batch; | |
786 | else | |
787 | to_drain = pcp->count; | |
788 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
789 | pcp->count -= to_drain; | |
790 | local_irq_restore(flags); | |
4ae7c039 CL |
791 | } |
792 | #endif | |
793 | ||
1da177e4 LT |
794 | static void __drain_pages(unsigned int cpu) |
795 | { | |
c54ad30c | 796 | unsigned long flags; |
1da177e4 LT |
797 | struct zone *zone; |
798 | int i; | |
799 | ||
800 | for_each_zone(zone) { | |
801 | struct per_cpu_pageset *pset; | |
802 | ||
f2e12bb2 CL |
803 | if (!populated_zone(zone)) |
804 | continue; | |
805 | ||
e7c8d5c9 | 806 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
807 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
808 | struct per_cpu_pages *pcp; | |
809 | ||
810 | pcp = &pset->pcp[i]; | |
c54ad30c | 811 | local_irq_save(flags); |
48db57f8 NP |
812 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
813 | pcp->count = 0; | |
c54ad30c | 814 | local_irq_restore(flags); |
1da177e4 LT |
815 | } |
816 | } | |
817 | } | |
1da177e4 | 818 | |
296699de | 819 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
820 | |
821 | void mark_free_pages(struct zone *zone) | |
822 | { | |
f623f0db RW |
823 | unsigned long pfn, max_zone_pfn; |
824 | unsigned long flags; | |
b2a0ac88 | 825 | int order, t; |
1da177e4 LT |
826 | struct list_head *curr; |
827 | ||
828 | if (!zone->spanned_pages) | |
829 | return; | |
830 | ||
831 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
832 | |
833 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
834 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
835 | if (pfn_valid(pfn)) { | |
836 | struct page *page = pfn_to_page(pfn); | |
837 | ||
7be98234 RW |
838 | if (!swsusp_page_is_forbidden(page)) |
839 | swsusp_unset_page_free(page); | |
f623f0db | 840 | } |
1da177e4 | 841 | |
b2a0ac88 MG |
842 | for_each_migratetype_order(order, t) { |
843 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 844 | unsigned long i; |
1da177e4 | 845 | |
f623f0db RW |
846 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
847 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 848 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 849 | } |
b2a0ac88 | 850 | } |
1da177e4 LT |
851 | spin_unlock_irqrestore(&zone->lock, flags); |
852 | } | |
853 | ||
854 | /* | |
855 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
856 | */ | |
857 | void drain_local_pages(void) | |
858 | { | |
859 | unsigned long flags; | |
860 | ||
861 | local_irq_save(flags); | |
862 | __drain_pages(smp_processor_id()); | |
863 | local_irq_restore(flags); | |
864 | } | |
296699de | 865 | #endif /* CONFIG_HIBERNATION */ |
1da177e4 | 866 | |
1da177e4 LT |
867 | /* |
868 | * Free a 0-order page | |
869 | */ | |
1da177e4 LT |
870 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
871 | { | |
872 | struct zone *zone = page_zone(page); | |
873 | struct per_cpu_pages *pcp; | |
874 | unsigned long flags; | |
875 | ||
1da177e4 LT |
876 | if (PageAnon(page)) |
877 | page->mapping = NULL; | |
224abf92 | 878 | if (free_pages_check(page)) |
689bcebf HD |
879 | return; |
880 | ||
9858db50 NP |
881 | if (!PageHighMem(page)) |
882 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 883 | arch_free_page(page, 0); |
689bcebf HD |
884 | kernel_map_pages(page, 1, 0); |
885 | ||
e7c8d5c9 | 886 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 887 | local_irq_save(flags); |
f8891e5e | 888 | __count_vm_event(PGFREE); |
1da177e4 LT |
889 | list_add(&page->lru, &pcp->list); |
890 | pcp->count++; | |
48db57f8 NP |
891 | if (pcp->count >= pcp->high) { |
892 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
893 | pcp->count -= pcp->batch; | |
894 | } | |
1da177e4 LT |
895 | local_irq_restore(flags); |
896 | put_cpu(); | |
897 | } | |
898 | ||
899 | void fastcall free_hot_page(struct page *page) | |
900 | { | |
901 | free_hot_cold_page(page, 0); | |
902 | } | |
903 | ||
904 | void fastcall free_cold_page(struct page *page) | |
905 | { | |
906 | free_hot_cold_page(page, 1); | |
907 | } | |
908 | ||
8dfcc9ba NP |
909 | /* |
910 | * split_page takes a non-compound higher-order page, and splits it into | |
911 | * n (1<<order) sub-pages: page[0..n] | |
912 | * Each sub-page must be freed individually. | |
913 | * | |
914 | * Note: this is probably too low level an operation for use in drivers. | |
915 | * Please consult with lkml before using this in your driver. | |
916 | */ | |
917 | void split_page(struct page *page, unsigned int order) | |
918 | { | |
919 | int i; | |
920 | ||
725d704e NP |
921 | VM_BUG_ON(PageCompound(page)); |
922 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
923 | for (i = 1; i < (1 << order); i++) |
924 | set_page_refcounted(page + i); | |
8dfcc9ba | 925 | } |
8dfcc9ba | 926 | |
1da177e4 LT |
927 | /* |
928 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
929 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
930 | * or two. | |
931 | */ | |
a74609fa NP |
932 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
933 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
934 | { |
935 | unsigned long flags; | |
689bcebf | 936 | struct page *page; |
1da177e4 | 937 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 938 | int cpu; |
b2a0ac88 | 939 | int migratetype = gfpflags_to_migratetype(gfp_flags); |
1da177e4 | 940 | |
689bcebf | 941 | again: |
a74609fa | 942 | cpu = get_cpu(); |
48db57f8 | 943 | if (likely(order == 0)) { |
1da177e4 LT |
944 | struct per_cpu_pages *pcp; |
945 | ||
a74609fa | 946 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 947 | local_irq_save(flags); |
a74609fa | 948 | if (!pcp->count) { |
941c7105 | 949 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 950 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
951 | if (unlikely(!pcp->count)) |
952 | goto failed; | |
1da177e4 | 953 | } |
a74609fa NP |
954 | page = list_entry(pcp->list.next, struct page, lru); |
955 | list_del(&page->lru); | |
956 | pcp->count--; | |
7fb1d9fc | 957 | } else { |
1da177e4 | 958 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 959 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
960 | spin_unlock(&zone->lock); |
961 | if (!page) | |
962 | goto failed; | |
1da177e4 LT |
963 | } |
964 | ||
f8891e5e | 965 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 966 | zone_statistics(zonelist, zone); |
a74609fa NP |
967 | local_irq_restore(flags); |
968 | put_cpu(); | |
1da177e4 | 969 | |
725d704e | 970 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 971 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 972 | goto again; |
1da177e4 | 973 | return page; |
a74609fa NP |
974 | |
975 | failed: | |
976 | local_irq_restore(flags); | |
977 | put_cpu(); | |
978 | return NULL; | |
1da177e4 LT |
979 | } |
980 | ||
7fb1d9fc | 981 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
982 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
983 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
984 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
985 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
986 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
987 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 988 | |
933e312e AM |
989 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
990 | ||
991 | static struct fail_page_alloc_attr { | |
992 | struct fault_attr attr; | |
993 | ||
994 | u32 ignore_gfp_highmem; | |
995 | u32 ignore_gfp_wait; | |
54114994 | 996 | u32 min_order; |
933e312e AM |
997 | |
998 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
999 | ||
1000 | struct dentry *ignore_gfp_highmem_file; | |
1001 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1002 | struct dentry *min_order_file; |
933e312e AM |
1003 | |
1004 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1005 | ||
1006 | } fail_page_alloc = { | |
1007 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1008 | .ignore_gfp_wait = 1, |
1009 | .ignore_gfp_highmem = 1, | |
54114994 | 1010 | .min_order = 1, |
933e312e AM |
1011 | }; |
1012 | ||
1013 | static int __init setup_fail_page_alloc(char *str) | |
1014 | { | |
1015 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1016 | } | |
1017 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1018 | ||
1019 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1020 | { | |
54114994 AM |
1021 | if (order < fail_page_alloc.min_order) |
1022 | return 0; | |
933e312e AM |
1023 | if (gfp_mask & __GFP_NOFAIL) |
1024 | return 0; | |
1025 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1026 | return 0; | |
1027 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1028 | return 0; | |
1029 | ||
1030 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1031 | } | |
1032 | ||
1033 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1034 | ||
1035 | static int __init fail_page_alloc_debugfs(void) | |
1036 | { | |
1037 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1038 | struct dentry *dir; | |
1039 | int err; | |
1040 | ||
1041 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1042 | "fail_page_alloc"); | |
1043 | if (err) | |
1044 | return err; | |
1045 | dir = fail_page_alloc.attr.dentries.dir; | |
1046 | ||
1047 | fail_page_alloc.ignore_gfp_wait_file = | |
1048 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1049 | &fail_page_alloc.ignore_gfp_wait); | |
1050 | ||
1051 | fail_page_alloc.ignore_gfp_highmem_file = | |
1052 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1053 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1054 | fail_page_alloc.min_order_file = |
1055 | debugfs_create_u32("min-order", mode, dir, | |
1056 | &fail_page_alloc.min_order); | |
933e312e AM |
1057 | |
1058 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1059 | !fail_page_alloc.ignore_gfp_highmem_file || |
1060 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1061 | err = -ENOMEM; |
1062 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1063 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1064 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1065 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1066 | } | |
1067 | ||
1068 | return err; | |
1069 | } | |
1070 | ||
1071 | late_initcall(fail_page_alloc_debugfs); | |
1072 | ||
1073 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1074 | ||
1075 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1076 | ||
1077 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1078 | { | |
1079 | return 0; | |
1080 | } | |
1081 | ||
1082 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1083 | ||
1da177e4 LT |
1084 | /* |
1085 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1086 | * of the allocation. | |
1087 | */ | |
1088 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1089 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1090 | { |
1091 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1092 | long min = mark; |
1093 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1094 | int o; |
1095 | ||
7fb1d9fc | 1096 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1097 | min -= min / 2; |
7fb1d9fc | 1098 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1099 | min -= min / 4; |
1100 | ||
1101 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1102 | return 0; | |
1103 | for (o = 0; o < order; o++) { | |
1104 | /* At the next order, this order's pages become unavailable */ | |
1105 | free_pages -= z->free_area[o].nr_free << o; | |
1106 | ||
1107 | /* Require fewer higher order pages to be free */ | |
1108 | min >>= 1; | |
1109 | ||
1110 | if (free_pages <= min) | |
1111 | return 0; | |
1112 | } | |
1113 | return 1; | |
1114 | } | |
1115 | ||
9276b1bc PJ |
1116 | #ifdef CONFIG_NUMA |
1117 | /* | |
1118 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1119 | * skip over zones that are not allowed by the cpuset, or that have | |
1120 | * been recently (in last second) found to be nearly full. See further | |
1121 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1122 | * that have to skip over alot of full or unallowed zones. | |
1123 | * | |
1124 | * If the zonelist cache is present in the passed in zonelist, then | |
1125 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1126 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1127 | * |
1128 | * If the zonelist cache is not available for this zonelist, does | |
1129 | * nothing and returns NULL. | |
1130 | * | |
1131 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1132 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1133 | * | |
1134 | * We hold off even calling zlc_setup, until after we've checked the | |
1135 | * first zone in the zonelist, on the theory that most allocations will | |
1136 | * be satisfied from that first zone, so best to examine that zone as | |
1137 | * quickly as we can. | |
1138 | */ | |
1139 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1140 | { | |
1141 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1142 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1143 | ||
1144 | zlc = zonelist->zlcache_ptr; | |
1145 | if (!zlc) | |
1146 | return NULL; | |
1147 | ||
1148 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1149 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1150 | zlc->last_full_zap = jiffies; | |
1151 | } | |
1152 | ||
1153 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1154 | &cpuset_current_mems_allowed : | |
37b07e41 | 1155 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1156 | return allowednodes; |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1161 | * if it is worth looking at further for free memory: | |
1162 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1163 | * bit set in the zonelist_cache fullzones BITMAP). | |
1164 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1165 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1166 | * Return true (non-zero) if zone is worth looking at further, or | |
1167 | * else return false (zero) if it is not. | |
1168 | * | |
1169 | * This check -ignores- the distinction between various watermarks, | |
1170 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1171 | * found to be full for any variation of these watermarks, it will | |
1172 | * be considered full for up to one second by all requests, unless | |
1173 | * we are so low on memory on all allowed nodes that we are forced | |
1174 | * into the second scan of the zonelist. | |
1175 | * | |
1176 | * In the second scan we ignore this zonelist cache and exactly | |
1177 | * apply the watermarks to all zones, even it is slower to do so. | |
1178 | * We are low on memory in the second scan, and should leave no stone | |
1179 | * unturned looking for a free page. | |
1180 | */ | |
1181 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1182 | nodemask_t *allowednodes) | |
1183 | { | |
1184 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1185 | int i; /* index of *z in zonelist zones */ | |
1186 | int n; /* node that zone *z is on */ | |
1187 | ||
1188 | zlc = zonelist->zlcache_ptr; | |
1189 | if (!zlc) | |
1190 | return 1; | |
1191 | ||
1192 | i = z - zonelist->zones; | |
1193 | n = zlc->z_to_n[i]; | |
1194 | ||
1195 | /* This zone is worth trying if it is allowed but not full */ | |
1196 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1201 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1202 | * from that zone don't waste time re-examining it. | |
1203 | */ | |
1204 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1205 | { | |
1206 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1207 | int i; /* index of *z in zonelist zones */ | |
1208 | ||
1209 | zlc = zonelist->zlcache_ptr; | |
1210 | if (!zlc) | |
1211 | return; | |
1212 | ||
1213 | i = z - zonelist->zones; | |
1214 | ||
1215 | set_bit(i, zlc->fullzones); | |
1216 | } | |
1217 | ||
1218 | #else /* CONFIG_NUMA */ | |
1219 | ||
1220 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1221 | { | |
1222 | return NULL; | |
1223 | } | |
1224 | ||
1225 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1226 | nodemask_t *allowednodes) | |
1227 | { | |
1228 | return 1; | |
1229 | } | |
1230 | ||
1231 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1232 | { | |
1233 | } | |
1234 | #endif /* CONFIG_NUMA */ | |
1235 | ||
7fb1d9fc | 1236 | /* |
0798e519 | 1237 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1238 | * a page. |
1239 | */ | |
1240 | static struct page * | |
1241 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1242 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1243 | { |
9276b1bc | 1244 | struct zone **z; |
7fb1d9fc | 1245 | struct page *page = NULL; |
9276b1bc | 1246 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1247 | struct zone *zone; |
9276b1bc PJ |
1248 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1249 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1250 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
b377fd39 | 1251 | enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */ |
7fb1d9fc | 1252 | |
9276b1bc | 1253 | zonelist_scan: |
7fb1d9fc | 1254 | /* |
9276b1bc | 1255 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1256 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1257 | */ | |
9276b1bc PJ |
1258 | z = zonelist->zones; |
1259 | ||
7fb1d9fc | 1260 | do { |
b377fd39 MG |
1261 | /* |
1262 | * In NUMA, this could be a policy zonelist which contains | |
1263 | * zones that may not be allowed by the current gfp_mask. | |
1264 | * Check the zone is allowed by the current flags | |
1265 | */ | |
1266 | if (unlikely(alloc_should_filter_zonelist(zonelist))) { | |
1267 | if (highest_zoneidx == -1) | |
1268 | highest_zoneidx = gfp_zone(gfp_mask); | |
1269 | if (zone_idx(*z) > highest_zoneidx) | |
1270 | continue; | |
1271 | } | |
1272 | ||
9276b1bc PJ |
1273 | if (NUMA_BUILD && zlc_active && |
1274 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1275 | continue; | |
1192d526 | 1276 | zone = *z; |
7fb1d9fc | 1277 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1278 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1279 | goto try_next_zone; |
7fb1d9fc RS |
1280 | |
1281 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1282 | unsigned long mark; |
1283 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1284 | mark = zone->pages_min; |
3148890b | 1285 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1286 | mark = zone->pages_low; |
3148890b | 1287 | else |
1192d526 | 1288 | mark = zone->pages_high; |
0798e519 PJ |
1289 | if (!zone_watermark_ok(zone, order, mark, |
1290 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1291 | if (!zone_reclaim_mode || |
1192d526 | 1292 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1293 | goto this_zone_full; |
0798e519 | 1294 | } |
7fb1d9fc RS |
1295 | } |
1296 | ||
1192d526 | 1297 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1298 | if (page) |
7fb1d9fc | 1299 | break; |
9276b1bc PJ |
1300 | this_zone_full: |
1301 | if (NUMA_BUILD) | |
1302 | zlc_mark_zone_full(zonelist, z); | |
1303 | try_next_zone: | |
1304 | if (NUMA_BUILD && !did_zlc_setup) { | |
1305 | /* we do zlc_setup after the first zone is tried */ | |
1306 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1307 | zlc_active = 1; | |
1308 | did_zlc_setup = 1; | |
1309 | } | |
7fb1d9fc | 1310 | } while (*(++z) != NULL); |
9276b1bc PJ |
1311 | |
1312 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1313 | /* Disable zlc cache for second zonelist scan */ | |
1314 | zlc_active = 0; | |
1315 | goto zonelist_scan; | |
1316 | } | |
7fb1d9fc | 1317 | return page; |
753ee728 MH |
1318 | } |
1319 | ||
1da177e4 LT |
1320 | /* |
1321 | * This is the 'heart' of the zoned buddy allocator. | |
1322 | */ | |
1323 | struct page * fastcall | |
dd0fc66f | 1324 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1325 | struct zonelist *zonelist) |
1326 | { | |
260b2367 | 1327 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1328 | struct zone **z; |
1da177e4 LT |
1329 | struct page *page; |
1330 | struct reclaim_state reclaim_state; | |
1331 | struct task_struct *p = current; | |
1da177e4 | 1332 | int do_retry; |
7fb1d9fc | 1333 | int alloc_flags; |
1da177e4 LT |
1334 | int did_some_progress; |
1335 | ||
1336 | might_sleep_if(wait); | |
1337 | ||
933e312e AM |
1338 | if (should_fail_alloc_page(gfp_mask, order)) |
1339 | return NULL; | |
1340 | ||
6b1de916 | 1341 | restart: |
7fb1d9fc | 1342 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1343 | |
7fb1d9fc | 1344 | if (unlikely(*z == NULL)) { |
523b9458 CL |
1345 | /* |
1346 | * Happens if we have an empty zonelist as a result of | |
1347 | * GFP_THISNODE being used on a memoryless node | |
1348 | */ | |
1da177e4 LT |
1349 | return NULL; |
1350 | } | |
6b1de916 | 1351 | |
7fb1d9fc | 1352 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1353 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1354 | if (page) |
1355 | goto got_pg; | |
1da177e4 | 1356 | |
952f3b51 CL |
1357 | /* |
1358 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1359 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1360 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1361 | * using a larger set of nodes after it has established that the | |
1362 | * allowed per node queues are empty and that nodes are | |
1363 | * over allocated. | |
1364 | */ | |
1365 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1366 | goto nopage; | |
1367 | ||
0798e519 | 1368 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1369 | wakeup_kswapd(*z, order); |
1da177e4 | 1370 | |
9bf2229f | 1371 | /* |
7fb1d9fc RS |
1372 | * OK, we're below the kswapd watermark and have kicked background |
1373 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1374 | * to how we want to proceed. | |
1375 | * | |
1376 | * The caller may dip into page reserves a bit more if the caller | |
1377 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1378 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1379 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1380 | */ |
3148890b | 1381 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1382 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1383 | alloc_flags |= ALLOC_HARDER; | |
1384 | if (gfp_mask & __GFP_HIGH) | |
1385 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1386 | if (wait) |
1387 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1388 | |
1389 | /* | |
1390 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1391 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1392 | * |
1393 | * This is the last chance, in general, before the goto nopage. | |
1394 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1395 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1396 | */ |
7fb1d9fc RS |
1397 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1398 | if (page) | |
1399 | goto got_pg; | |
1da177e4 LT |
1400 | |
1401 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1402 | |
b43a57bb | 1403 | rebalance: |
b84a35be NP |
1404 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1405 | && !in_interrupt()) { | |
1406 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1407 | nofail_alloc: |
b84a35be | 1408 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1409 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1410 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1411 | if (page) |
1412 | goto got_pg; | |
885036d3 | 1413 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1414 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1415 | goto nofail_alloc; |
1416 | } | |
1da177e4 LT |
1417 | } |
1418 | goto nopage; | |
1419 | } | |
1420 | ||
1421 | /* Atomic allocations - we can't balance anything */ | |
1422 | if (!wait) | |
1423 | goto nopage; | |
1424 | ||
1da177e4 LT |
1425 | cond_resched(); |
1426 | ||
1427 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1428 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1429 | p->flags |= PF_MEMALLOC; |
1430 | reclaim_state.reclaimed_slab = 0; | |
1431 | p->reclaim_state = &reclaim_state; | |
1432 | ||
5ad333eb | 1433 | did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask); |
1da177e4 LT |
1434 | |
1435 | p->reclaim_state = NULL; | |
1436 | p->flags &= ~PF_MEMALLOC; | |
1437 | ||
1438 | cond_resched(); | |
1439 | ||
1440 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1441 | page = get_page_from_freelist(gfp_mask, order, |
1442 | zonelist, alloc_flags); | |
1443 | if (page) | |
1444 | goto got_pg; | |
1da177e4 LT |
1445 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1446 | /* | |
1447 | * Go through the zonelist yet one more time, keep | |
1448 | * very high watermark here, this is only to catch | |
1449 | * a parallel oom killing, we must fail if we're still | |
1450 | * under heavy pressure. | |
1451 | */ | |
7fb1d9fc | 1452 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1453 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1454 | if (page) |
1455 | goto got_pg; | |
1da177e4 | 1456 | |
a8bbf72a MG |
1457 | /* The OOM killer will not help higher order allocs so fail */ |
1458 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1459 | goto nopage; | |
1460 | ||
9b0f8b04 | 1461 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1462 | goto restart; |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Don't let big-order allocations loop unless the caller explicitly | |
1467 | * requests that. Wait for some write requests to complete then retry. | |
1468 | * | |
1469 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1470 | * <= 3, but that may not be true in other implementations. | |
1471 | */ | |
1472 | do_retry = 0; | |
1473 | if (!(gfp_mask & __GFP_NORETRY)) { | |
5ad333eb AW |
1474 | if ((order <= PAGE_ALLOC_COSTLY_ORDER) || |
1475 | (gfp_mask & __GFP_REPEAT)) | |
1da177e4 LT |
1476 | do_retry = 1; |
1477 | if (gfp_mask & __GFP_NOFAIL) | |
1478 | do_retry = 1; | |
1479 | } | |
1480 | if (do_retry) { | |
3fcfab16 | 1481 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1482 | goto rebalance; |
1483 | } | |
1484 | ||
1485 | nopage: | |
1486 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1487 | printk(KERN_WARNING "%s: page allocation failure." | |
1488 | " order:%d, mode:0x%x\n", | |
1489 | p->comm, order, gfp_mask); | |
1490 | dump_stack(); | |
578c2fd6 | 1491 | show_mem(); |
1da177e4 | 1492 | } |
1da177e4 | 1493 | got_pg: |
1da177e4 LT |
1494 | return page; |
1495 | } | |
1496 | ||
1497 | EXPORT_SYMBOL(__alloc_pages); | |
1498 | ||
1499 | /* | |
1500 | * Common helper functions. | |
1501 | */ | |
dd0fc66f | 1502 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1503 | { |
1504 | struct page * page; | |
1505 | page = alloc_pages(gfp_mask, order); | |
1506 | if (!page) | |
1507 | return 0; | |
1508 | return (unsigned long) page_address(page); | |
1509 | } | |
1510 | ||
1511 | EXPORT_SYMBOL(__get_free_pages); | |
1512 | ||
dd0fc66f | 1513 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1514 | { |
1515 | struct page * page; | |
1516 | ||
1517 | /* | |
1518 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1519 | * a highmem page | |
1520 | */ | |
725d704e | 1521 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1522 | |
1523 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1524 | if (page) | |
1525 | return (unsigned long) page_address(page); | |
1526 | return 0; | |
1527 | } | |
1528 | ||
1529 | EXPORT_SYMBOL(get_zeroed_page); | |
1530 | ||
1531 | void __pagevec_free(struct pagevec *pvec) | |
1532 | { | |
1533 | int i = pagevec_count(pvec); | |
1534 | ||
1535 | while (--i >= 0) | |
1536 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1537 | } | |
1538 | ||
1539 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1540 | { | |
b5810039 | 1541 | if (put_page_testzero(page)) { |
1da177e4 LT |
1542 | if (order == 0) |
1543 | free_hot_page(page); | |
1544 | else | |
1545 | __free_pages_ok(page, order); | |
1546 | } | |
1547 | } | |
1548 | ||
1549 | EXPORT_SYMBOL(__free_pages); | |
1550 | ||
1551 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1552 | { | |
1553 | if (addr != 0) { | |
725d704e | 1554 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1555 | __free_pages(virt_to_page((void *)addr), order); |
1556 | } | |
1557 | } | |
1558 | ||
1559 | EXPORT_SYMBOL(free_pages); | |
1560 | ||
1da177e4 LT |
1561 | static unsigned int nr_free_zone_pages(int offset) |
1562 | { | |
e310fd43 MB |
1563 | /* Just pick one node, since fallback list is circular */ |
1564 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1565 | unsigned int sum = 0; |
1566 | ||
e310fd43 MB |
1567 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1568 | struct zone **zonep = zonelist->zones; | |
1569 | struct zone *zone; | |
1da177e4 | 1570 | |
e310fd43 MB |
1571 | for (zone = *zonep++; zone; zone = *zonep++) { |
1572 | unsigned long size = zone->present_pages; | |
1573 | unsigned long high = zone->pages_high; | |
1574 | if (size > high) | |
1575 | sum += size - high; | |
1da177e4 LT |
1576 | } |
1577 | ||
1578 | return sum; | |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1583 | */ | |
1584 | unsigned int nr_free_buffer_pages(void) | |
1585 | { | |
af4ca457 | 1586 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1587 | } |
c2f1a551 | 1588 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1589 | |
1590 | /* | |
1591 | * Amount of free RAM allocatable within all zones | |
1592 | */ | |
1593 | unsigned int nr_free_pagecache_pages(void) | |
1594 | { | |
2a1e274a | 1595 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1596 | } |
08e0f6a9 CL |
1597 | |
1598 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1599 | { |
08e0f6a9 | 1600 | if (NUMA_BUILD) |
25ba77c1 | 1601 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1602 | } |
1da177e4 | 1603 | |
1da177e4 LT |
1604 | void si_meminfo(struct sysinfo *val) |
1605 | { | |
1606 | val->totalram = totalram_pages; | |
1607 | val->sharedram = 0; | |
d23ad423 | 1608 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1609 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1610 | val->totalhigh = totalhigh_pages; |
1611 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1612 | val->mem_unit = PAGE_SIZE; |
1613 | } | |
1614 | ||
1615 | EXPORT_SYMBOL(si_meminfo); | |
1616 | ||
1617 | #ifdef CONFIG_NUMA | |
1618 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1619 | { | |
1620 | pg_data_t *pgdat = NODE_DATA(nid); | |
1621 | ||
1622 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1623 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1624 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1625 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1626 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1627 | NR_FREE_PAGES); | |
98d2b0eb CL |
1628 | #else |
1629 | val->totalhigh = 0; | |
1630 | val->freehigh = 0; | |
1631 | #endif | |
1da177e4 LT |
1632 | val->mem_unit = PAGE_SIZE; |
1633 | } | |
1634 | #endif | |
1635 | ||
1636 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1637 | ||
1638 | /* | |
1639 | * Show free area list (used inside shift_scroll-lock stuff) | |
1640 | * We also calculate the percentage fragmentation. We do this by counting the | |
1641 | * memory on each free list with the exception of the first item on the list. | |
1642 | */ | |
1643 | void show_free_areas(void) | |
1644 | { | |
c7241913 | 1645 | int cpu; |
1da177e4 LT |
1646 | struct zone *zone; |
1647 | ||
1648 | for_each_zone(zone) { | |
c7241913 | 1649 | if (!populated_zone(zone)) |
1da177e4 | 1650 | continue; |
c7241913 JS |
1651 | |
1652 | show_node(zone); | |
1653 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1654 | |
6b482c67 | 1655 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1656 | struct per_cpu_pageset *pageset; |
1657 | ||
e7c8d5c9 | 1658 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1659 | |
c7241913 JS |
1660 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1661 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1662 | cpu, pageset->pcp[0].high, | |
1663 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1664 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1665 | pageset->pcp[1].count); | |
1da177e4 LT |
1666 | } |
1667 | } | |
1668 | ||
a25700a5 | 1669 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1670 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1671 | global_page_state(NR_ACTIVE), |
1672 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1673 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1674 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1675 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1676 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1677 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1678 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1679 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1680 | global_page_state(NR_PAGETABLE), |
1681 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1682 | |
1683 | for_each_zone(zone) { | |
1684 | int i; | |
1685 | ||
c7241913 JS |
1686 | if (!populated_zone(zone)) |
1687 | continue; | |
1688 | ||
1da177e4 LT |
1689 | show_node(zone); |
1690 | printk("%s" | |
1691 | " free:%lukB" | |
1692 | " min:%lukB" | |
1693 | " low:%lukB" | |
1694 | " high:%lukB" | |
1695 | " active:%lukB" | |
1696 | " inactive:%lukB" | |
1697 | " present:%lukB" | |
1698 | " pages_scanned:%lu" | |
1699 | " all_unreclaimable? %s" | |
1700 | "\n", | |
1701 | zone->name, | |
d23ad423 | 1702 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1703 | K(zone->pages_min), |
1704 | K(zone->pages_low), | |
1705 | K(zone->pages_high), | |
c8785385 CL |
1706 | K(zone_page_state(zone, NR_ACTIVE)), |
1707 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1708 | K(zone->present_pages), |
1709 | zone->pages_scanned, | |
1710 | (zone->all_unreclaimable ? "yes" : "no") | |
1711 | ); | |
1712 | printk("lowmem_reserve[]:"); | |
1713 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1714 | printk(" %lu", zone->lowmem_reserve[i]); | |
1715 | printk("\n"); | |
1716 | } | |
1717 | ||
1718 | for_each_zone(zone) { | |
8f9de51a | 1719 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1720 | |
c7241913 JS |
1721 | if (!populated_zone(zone)) |
1722 | continue; | |
1723 | ||
1da177e4 LT |
1724 | show_node(zone); |
1725 | printk("%s: ", zone->name); | |
1da177e4 LT |
1726 | |
1727 | spin_lock_irqsave(&zone->lock, flags); | |
1728 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1729 | nr[order] = zone->free_area[order].nr_free; |
1730 | total += nr[order] << order; | |
1da177e4 LT |
1731 | } |
1732 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1733 | for (order = 0; order < MAX_ORDER; order++) |
1734 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1735 | printk("= %lukB\n", K(total)); |
1736 | } | |
1737 | ||
1738 | show_swap_cache_info(); | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1743 | * |
1744 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1745 | */ |
f0c0b2b8 KH |
1746 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1747 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1748 | { |
1a93205b CL |
1749 | struct zone *zone; |
1750 | ||
98d2b0eb | 1751 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1752 | zone_type++; |
02a68a5e CL |
1753 | |
1754 | do { | |
2f6726e5 | 1755 | zone_type--; |
070f8032 | 1756 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1757 | if (populated_zone(zone)) { |
070f8032 CL |
1758 | zonelist->zones[nr_zones++] = zone; |
1759 | check_highest_zone(zone_type); | |
1da177e4 | 1760 | } |
02a68a5e | 1761 | |
2f6726e5 | 1762 | } while (zone_type); |
070f8032 | 1763 | return nr_zones; |
1da177e4 LT |
1764 | } |
1765 | ||
f0c0b2b8 KH |
1766 | |
1767 | /* | |
1768 | * zonelist_order: | |
1769 | * 0 = automatic detection of better ordering. | |
1770 | * 1 = order by ([node] distance, -zonetype) | |
1771 | * 2 = order by (-zonetype, [node] distance) | |
1772 | * | |
1773 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1774 | * the same zonelist. So only NUMA can configure this param. | |
1775 | */ | |
1776 | #define ZONELIST_ORDER_DEFAULT 0 | |
1777 | #define ZONELIST_ORDER_NODE 1 | |
1778 | #define ZONELIST_ORDER_ZONE 2 | |
1779 | ||
1780 | /* zonelist order in the kernel. | |
1781 | * set_zonelist_order() will set this to NODE or ZONE. | |
1782 | */ | |
1783 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1784 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1785 | ||
1786 | ||
1da177e4 | 1787 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1788 | /* The value user specified ....changed by config */ |
1789 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1790 | /* string for sysctl */ | |
1791 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1792 | char numa_zonelist_order[16] = "default"; | |
1793 | ||
1794 | /* | |
1795 | * interface for configure zonelist ordering. | |
1796 | * command line option "numa_zonelist_order" | |
1797 | * = "[dD]efault - default, automatic configuration. | |
1798 | * = "[nN]ode - order by node locality, then by zone within node | |
1799 | * = "[zZ]one - order by zone, then by locality within zone | |
1800 | */ | |
1801 | ||
1802 | static int __parse_numa_zonelist_order(char *s) | |
1803 | { | |
1804 | if (*s == 'd' || *s == 'D') { | |
1805 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1806 | } else if (*s == 'n' || *s == 'N') { | |
1807 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
1808 | } else if (*s == 'z' || *s == 'Z') { | |
1809 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
1810 | } else { | |
1811 | printk(KERN_WARNING | |
1812 | "Ignoring invalid numa_zonelist_order value: " | |
1813 | "%s\n", s); | |
1814 | return -EINVAL; | |
1815 | } | |
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | static __init int setup_numa_zonelist_order(char *s) | |
1820 | { | |
1821 | if (s) | |
1822 | return __parse_numa_zonelist_order(s); | |
1823 | return 0; | |
1824 | } | |
1825 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
1826 | ||
1827 | /* | |
1828 | * sysctl handler for numa_zonelist_order | |
1829 | */ | |
1830 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
1831 | struct file *file, void __user *buffer, size_t *length, | |
1832 | loff_t *ppos) | |
1833 | { | |
1834 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
1835 | int ret; | |
1836 | ||
1837 | if (write) | |
1838 | strncpy(saved_string, (char*)table->data, | |
1839 | NUMA_ZONELIST_ORDER_LEN); | |
1840 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
1841 | if (ret) | |
1842 | return ret; | |
1843 | if (write) { | |
1844 | int oldval = user_zonelist_order; | |
1845 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
1846 | /* | |
1847 | * bogus value. restore saved string | |
1848 | */ | |
1849 | strncpy((char*)table->data, saved_string, | |
1850 | NUMA_ZONELIST_ORDER_LEN); | |
1851 | user_zonelist_order = oldval; | |
1852 | } else if (oldval != user_zonelist_order) | |
1853 | build_all_zonelists(); | |
1854 | } | |
1855 | return 0; | |
1856 | } | |
1857 | ||
1858 | ||
1da177e4 | 1859 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
1860 | static int node_load[MAX_NUMNODES]; |
1861 | ||
1da177e4 | 1862 | /** |
4dc3b16b | 1863 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1864 | * @node: node whose fallback list we're appending |
1865 | * @used_node_mask: nodemask_t of already used nodes | |
1866 | * | |
1867 | * We use a number of factors to determine which is the next node that should | |
1868 | * appear on a given node's fallback list. The node should not have appeared | |
1869 | * already in @node's fallback list, and it should be the next closest node | |
1870 | * according to the distance array (which contains arbitrary distance values | |
1871 | * from each node to each node in the system), and should also prefer nodes | |
1872 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1873 | * on them otherwise. | |
1874 | * It returns -1 if no node is found. | |
1875 | */ | |
f0c0b2b8 | 1876 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1877 | { |
4cf808eb | 1878 | int n, val; |
1da177e4 LT |
1879 | int min_val = INT_MAX; |
1880 | int best_node = -1; | |
1881 | ||
4cf808eb LT |
1882 | /* Use the local node if we haven't already */ |
1883 | if (!node_isset(node, *used_node_mask)) { | |
1884 | node_set(node, *used_node_mask); | |
1885 | return node; | |
1886 | } | |
1da177e4 | 1887 | |
37b07e41 | 1888 | for_each_node_state(n, N_HIGH_MEMORY) { |
4cf808eb | 1889 | cpumask_t tmp; |
1da177e4 LT |
1890 | |
1891 | /* Don't want a node to appear more than once */ | |
1892 | if (node_isset(n, *used_node_mask)) | |
1893 | continue; | |
1894 | ||
1da177e4 LT |
1895 | /* Use the distance array to find the distance */ |
1896 | val = node_distance(node, n); | |
1897 | ||
4cf808eb LT |
1898 | /* Penalize nodes under us ("prefer the next node") */ |
1899 | val += (n < node); | |
1900 | ||
1da177e4 LT |
1901 | /* Give preference to headless and unused nodes */ |
1902 | tmp = node_to_cpumask(n); | |
1903 | if (!cpus_empty(tmp)) | |
1904 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1905 | ||
1906 | /* Slight preference for less loaded node */ | |
1907 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1908 | val += node_load[n]; | |
1909 | ||
1910 | if (val < min_val) { | |
1911 | min_val = val; | |
1912 | best_node = n; | |
1913 | } | |
1914 | } | |
1915 | ||
1916 | if (best_node >= 0) | |
1917 | node_set(best_node, *used_node_mask); | |
1918 | ||
1919 | return best_node; | |
1920 | } | |
1921 | ||
f0c0b2b8 KH |
1922 | |
1923 | /* | |
1924 | * Build zonelists ordered by node and zones within node. | |
1925 | * This results in maximum locality--normal zone overflows into local | |
1926 | * DMA zone, if any--but risks exhausting DMA zone. | |
1927 | */ | |
1928 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 1929 | { |
19655d34 | 1930 | enum zone_type i; |
f0c0b2b8 | 1931 | int j; |
1da177e4 | 1932 | struct zonelist *zonelist; |
f0c0b2b8 KH |
1933 | |
1934 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1935 | zonelist = pgdat->node_zonelists + i; | |
1936 | for (j = 0; zonelist->zones[j] != NULL; j++) | |
1937 | ; | |
1938 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
1939 | zonelist->zones[j] = NULL; | |
1940 | } | |
1941 | } | |
1942 | ||
523b9458 CL |
1943 | /* |
1944 | * Build gfp_thisnode zonelists | |
1945 | */ | |
1946 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
1947 | { | |
1948 | enum zone_type i; | |
1949 | int j; | |
1950 | struct zonelist *zonelist; | |
1951 | ||
1952 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1953 | zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i; | |
1954 | j = build_zonelists_node(pgdat, zonelist, 0, i); | |
1955 | zonelist->zones[j] = NULL; | |
1956 | } | |
1957 | } | |
1958 | ||
f0c0b2b8 KH |
1959 | /* |
1960 | * Build zonelists ordered by zone and nodes within zones. | |
1961 | * This results in conserving DMA zone[s] until all Normal memory is | |
1962 | * exhausted, but results in overflowing to remote node while memory | |
1963 | * may still exist in local DMA zone. | |
1964 | */ | |
1965 | static int node_order[MAX_NUMNODES]; | |
1966 | ||
1967 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
1968 | { | |
1969 | enum zone_type i; | |
1970 | int pos, j, node; | |
1971 | int zone_type; /* needs to be signed */ | |
1972 | struct zone *z; | |
1973 | struct zonelist *zonelist; | |
1974 | ||
1975 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1976 | zonelist = pgdat->node_zonelists + i; | |
1977 | pos = 0; | |
1978 | for (zone_type = i; zone_type >= 0; zone_type--) { | |
1979 | for (j = 0; j < nr_nodes; j++) { | |
1980 | node = node_order[j]; | |
1981 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
1982 | if (populated_zone(z)) { | |
1983 | zonelist->zones[pos++] = z; | |
1984 | check_highest_zone(zone_type); | |
1985 | } | |
1986 | } | |
1987 | } | |
1988 | zonelist->zones[pos] = NULL; | |
1989 | } | |
1990 | } | |
1991 | ||
1992 | static int default_zonelist_order(void) | |
1993 | { | |
1994 | int nid, zone_type; | |
1995 | unsigned long low_kmem_size,total_size; | |
1996 | struct zone *z; | |
1997 | int average_size; | |
1998 | /* | |
1999 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2000 | * If they are really small and used heavily, the system can fall | |
2001 | * into OOM very easily. | |
2002 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2003 | */ | |
2004 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2005 | low_kmem_size = 0; | |
2006 | total_size = 0; | |
2007 | for_each_online_node(nid) { | |
2008 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2009 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2010 | if (populated_zone(z)) { | |
2011 | if (zone_type < ZONE_NORMAL) | |
2012 | low_kmem_size += z->present_pages; | |
2013 | total_size += z->present_pages; | |
2014 | } | |
2015 | } | |
2016 | } | |
2017 | if (!low_kmem_size || /* there are no DMA area. */ | |
2018 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2019 | return ZONELIST_ORDER_NODE; | |
2020 | /* | |
2021 | * look into each node's config. | |
2022 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2023 | * local memory, NODE_ORDER may be suitable. | |
2024 | */ | |
37b07e41 LS |
2025 | average_size = total_size / |
2026 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2027 | for_each_online_node(nid) { |
2028 | low_kmem_size = 0; | |
2029 | total_size = 0; | |
2030 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2031 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2032 | if (populated_zone(z)) { | |
2033 | if (zone_type < ZONE_NORMAL) | |
2034 | low_kmem_size += z->present_pages; | |
2035 | total_size += z->present_pages; | |
2036 | } | |
2037 | } | |
2038 | if (low_kmem_size && | |
2039 | total_size > average_size && /* ignore small node */ | |
2040 | low_kmem_size > total_size * 70/100) | |
2041 | return ZONELIST_ORDER_NODE; | |
2042 | } | |
2043 | return ZONELIST_ORDER_ZONE; | |
2044 | } | |
2045 | ||
2046 | static void set_zonelist_order(void) | |
2047 | { | |
2048 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2049 | current_zonelist_order = default_zonelist_order(); | |
2050 | else | |
2051 | current_zonelist_order = user_zonelist_order; | |
2052 | } | |
2053 | ||
2054 | static void build_zonelists(pg_data_t *pgdat) | |
2055 | { | |
2056 | int j, node, load; | |
2057 | enum zone_type i; | |
1da177e4 | 2058 | nodemask_t used_mask; |
f0c0b2b8 KH |
2059 | int local_node, prev_node; |
2060 | struct zonelist *zonelist; | |
2061 | int order = current_zonelist_order; | |
1da177e4 LT |
2062 | |
2063 | /* initialize zonelists */ | |
523b9458 | 2064 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 LT |
2065 | zonelist = pgdat->node_zonelists + i; |
2066 | zonelist->zones[0] = NULL; | |
2067 | } | |
2068 | ||
2069 | /* NUMA-aware ordering of nodes */ | |
2070 | local_node = pgdat->node_id; | |
2071 | load = num_online_nodes(); | |
2072 | prev_node = local_node; | |
2073 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2074 | |
2075 | memset(node_load, 0, sizeof(node_load)); | |
2076 | memset(node_order, 0, sizeof(node_order)); | |
2077 | j = 0; | |
2078 | ||
1da177e4 | 2079 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2080 | int distance = node_distance(local_node, node); |
2081 | ||
2082 | /* | |
2083 | * If another node is sufficiently far away then it is better | |
2084 | * to reclaim pages in a zone before going off node. | |
2085 | */ | |
2086 | if (distance > RECLAIM_DISTANCE) | |
2087 | zone_reclaim_mode = 1; | |
2088 | ||
1da177e4 LT |
2089 | /* |
2090 | * We don't want to pressure a particular node. | |
2091 | * So adding penalty to the first node in same | |
2092 | * distance group to make it round-robin. | |
2093 | */ | |
9eeff239 | 2094 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2095 | node_load[node] = load; |
2096 | ||
1da177e4 LT |
2097 | prev_node = node; |
2098 | load--; | |
f0c0b2b8 KH |
2099 | if (order == ZONELIST_ORDER_NODE) |
2100 | build_zonelists_in_node_order(pgdat, node); | |
2101 | else | |
2102 | node_order[j++] = node; /* remember order */ | |
2103 | } | |
1da177e4 | 2104 | |
f0c0b2b8 KH |
2105 | if (order == ZONELIST_ORDER_ZONE) { |
2106 | /* calculate node order -- i.e., DMA last! */ | |
2107 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2108 | } |
523b9458 CL |
2109 | |
2110 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2111 | } |
2112 | ||
9276b1bc | 2113 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2114 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2115 | { |
2116 | int i; | |
2117 | ||
2118 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2119 | struct zonelist *zonelist; | |
2120 | struct zonelist_cache *zlc; | |
2121 | struct zone **z; | |
2122 | ||
2123 | zonelist = pgdat->node_zonelists + i; | |
2124 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2125 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2126 | for (z = zonelist->zones; *z; z++) | |
2127 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
2128 | } | |
2129 | } | |
2130 | ||
f0c0b2b8 | 2131 | |
1da177e4 LT |
2132 | #else /* CONFIG_NUMA */ |
2133 | ||
f0c0b2b8 KH |
2134 | static void set_zonelist_order(void) |
2135 | { | |
2136 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2137 | } | |
2138 | ||
2139 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2140 | { |
19655d34 CL |
2141 | int node, local_node; |
2142 | enum zone_type i,j; | |
1da177e4 LT |
2143 | |
2144 | local_node = pgdat->node_id; | |
19655d34 | 2145 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
2146 | struct zonelist *zonelist; |
2147 | ||
2148 | zonelist = pgdat->node_zonelists + i; | |
2149 | ||
19655d34 | 2150 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
2151 | /* |
2152 | * Now we build the zonelist so that it contains the zones | |
2153 | * of all the other nodes. | |
2154 | * We don't want to pressure a particular node, so when | |
2155 | * building the zones for node N, we make sure that the | |
2156 | * zones coming right after the local ones are those from | |
2157 | * node N+1 (modulo N) | |
2158 | */ | |
2159 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2160 | if (!node_online(node)) | |
2161 | continue; | |
19655d34 | 2162 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2163 | } |
2164 | for (node = 0; node < local_node; node++) { | |
2165 | if (!node_online(node)) | |
2166 | continue; | |
19655d34 | 2167 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | zonelist->zones[j] = NULL; | |
2171 | } | |
2172 | } | |
2173 | ||
9276b1bc | 2174 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2175 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2176 | { |
2177 | int i; | |
2178 | ||
2179 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2180 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
2181 | } | |
2182 | ||
1da177e4 LT |
2183 | #endif /* CONFIG_NUMA */ |
2184 | ||
6811378e | 2185 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2186 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2187 | { |
6811378e | 2188 | int nid; |
9276b1bc PJ |
2189 | |
2190 | for_each_online_node(nid) { | |
7ea1530a CL |
2191 | pg_data_t *pgdat = NODE_DATA(nid); |
2192 | ||
2193 | build_zonelists(pgdat); | |
2194 | build_zonelist_cache(pgdat); | |
9276b1bc | 2195 | } |
6811378e YG |
2196 | return 0; |
2197 | } | |
2198 | ||
f0c0b2b8 | 2199 | void build_all_zonelists(void) |
6811378e | 2200 | { |
f0c0b2b8 KH |
2201 | set_zonelist_order(); |
2202 | ||
6811378e | 2203 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2204 | __build_all_zonelists(NULL); |
6811378e YG |
2205 | cpuset_init_current_mems_allowed(); |
2206 | } else { | |
2207 | /* we have to stop all cpus to guaranntee there is no user | |
2208 | of zonelist */ | |
2209 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2210 | /* cpuset refresh routine should be here */ | |
2211 | } | |
bd1e22b8 | 2212 | vm_total_pages = nr_free_pagecache_pages(); |
f0c0b2b8 KH |
2213 | printk("Built %i zonelists in %s order. Total pages: %ld\n", |
2214 | num_online_nodes(), | |
2215 | zonelist_order_name[current_zonelist_order], | |
2216 | vm_total_pages); | |
2217 | #ifdef CONFIG_NUMA | |
2218 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2219 | #endif | |
1da177e4 LT |
2220 | } |
2221 | ||
2222 | /* | |
2223 | * Helper functions to size the waitqueue hash table. | |
2224 | * Essentially these want to choose hash table sizes sufficiently | |
2225 | * large so that collisions trying to wait on pages are rare. | |
2226 | * But in fact, the number of active page waitqueues on typical | |
2227 | * systems is ridiculously low, less than 200. So this is even | |
2228 | * conservative, even though it seems large. | |
2229 | * | |
2230 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2231 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2232 | */ | |
2233 | #define PAGES_PER_WAITQUEUE 256 | |
2234 | ||
cca448fe | 2235 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2236 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2237 | { |
2238 | unsigned long size = 1; | |
2239 | ||
2240 | pages /= PAGES_PER_WAITQUEUE; | |
2241 | ||
2242 | while (size < pages) | |
2243 | size <<= 1; | |
2244 | ||
2245 | /* | |
2246 | * Once we have dozens or even hundreds of threads sleeping | |
2247 | * on IO we've got bigger problems than wait queue collision. | |
2248 | * Limit the size of the wait table to a reasonable size. | |
2249 | */ | |
2250 | size = min(size, 4096UL); | |
2251 | ||
2252 | return max(size, 4UL); | |
2253 | } | |
cca448fe YG |
2254 | #else |
2255 | /* | |
2256 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2257 | * a suitable size for its wait_table. So we use the maximum size now. | |
2258 | * | |
2259 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2260 | * | |
2261 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2262 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2263 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2264 | * | |
2265 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2266 | * or more by the traditional way. (See above). It equals: | |
2267 | * | |
2268 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2269 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2270 | * powerpc (64K page size) : = (32G +16M)byte. | |
2271 | */ | |
2272 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2273 | { | |
2274 | return 4096UL; | |
2275 | } | |
2276 | #endif | |
1da177e4 LT |
2277 | |
2278 | /* | |
2279 | * This is an integer logarithm so that shifts can be used later | |
2280 | * to extract the more random high bits from the multiplicative | |
2281 | * hash function before the remainder is taken. | |
2282 | */ | |
2283 | static inline unsigned long wait_table_bits(unsigned long size) | |
2284 | { | |
2285 | return ffz(~size); | |
2286 | } | |
2287 | ||
2288 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2289 | ||
1da177e4 LT |
2290 | /* |
2291 | * Initially all pages are reserved - free ones are freed | |
2292 | * up by free_all_bootmem() once the early boot process is | |
2293 | * done. Non-atomic initialization, single-pass. | |
2294 | */ | |
c09b4240 | 2295 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2296 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2297 | { |
1da177e4 | 2298 | struct page *page; |
29751f69 AW |
2299 | unsigned long end_pfn = start_pfn + size; |
2300 | unsigned long pfn; | |
1da177e4 | 2301 | |
cbe8dd4a | 2302 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2303 | /* |
2304 | * There can be holes in boot-time mem_map[]s | |
2305 | * handed to this function. They do not | |
2306 | * exist on hotplugged memory. | |
2307 | */ | |
2308 | if (context == MEMMAP_EARLY) { | |
2309 | if (!early_pfn_valid(pfn)) | |
2310 | continue; | |
2311 | if (!early_pfn_in_nid(pfn, nid)) | |
2312 | continue; | |
2313 | } | |
d41dee36 AW |
2314 | page = pfn_to_page(pfn); |
2315 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 2316 | init_page_count(page); |
1da177e4 LT |
2317 | reset_page_mapcount(page); |
2318 | SetPageReserved(page); | |
b2a0ac88 MG |
2319 | |
2320 | /* | |
2321 | * Mark the block movable so that blocks are reserved for | |
2322 | * movable at startup. This will force kernel allocations | |
2323 | * to reserve their blocks rather than leaking throughout | |
2324 | * the address space during boot when many long-lived | |
2325 | * kernel allocations are made | |
2326 | */ | |
2327 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2328 | ||
1da177e4 LT |
2329 | INIT_LIST_HEAD(&page->lru); |
2330 | #ifdef WANT_PAGE_VIRTUAL | |
2331 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2332 | if (!is_highmem_idx(zone)) | |
3212c6be | 2333 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2334 | #endif |
1da177e4 LT |
2335 | } |
2336 | } | |
2337 | ||
6ea6e688 PM |
2338 | static void __meminit zone_init_free_lists(struct pglist_data *pgdat, |
2339 | struct zone *zone, unsigned long size) | |
1da177e4 | 2340 | { |
b2a0ac88 MG |
2341 | int order, t; |
2342 | for_each_migratetype_order(order, t) { | |
2343 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2344 | zone->free_area[order].nr_free = 0; |
2345 | } | |
2346 | } | |
2347 | ||
2348 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2349 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2350 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2351 | #endif |
2352 | ||
d09c6b80 | 2353 | static int __devinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2354 | { |
2355 | int batch; | |
2356 | ||
2357 | /* | |
2358 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2359 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2360 | * |
2361 | * OK, so we don't know how big the cache is. So guess. | |
2362 | */ | |
2363 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2364 | if (batch * PAGE_SIZE > 512 * 1024) |
2365 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2366 | batch /= 4; /* We effectively *= 4 below */ |
2367 | if (batch < 1) | |
2368 | batch = 1; | |
2369 | ||
2370 | /* | |
0ceaacc9 NP |
2371 | * Clamp the batch to a 2^n - 1 value. Having a power |
2372 | * of 2 value was found to be more likely to have | |
2373 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2374 | * |
0ceaacc9 NP |
2375 | * For example if 2 tasks are alternately allocating |
2376 | * batches of pages, one task can end up with a lot | |
2377 | * of pages of one half of the possible page colors | |
2378 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2379 | */ |
0ceaacc9 | 2380 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2381 | |
e7c8d5c9 CL |
2382 | return batch; |
2383 | } | |
2384 | ||
2caaad41 CL |
2385 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2386 | { | |
2387 | struct per_cpu_pages *pcp; | |
2388 | ||
1c6fe946 MD |
2389 | memset(p, 0, sizeof(*p)); |
2390 | ||
2caaad41 CL |
2391 | pcp = &p->pcp[0]; /* hot */ |
2392 | pcp->count = 0; | |
2caaad41 CL |
2393 | pcp->high = 6 * batch; |
2394 | pcp->batch = max(1UL, 1 * batch); | |
2395 | INIT_LIST_HEAD(&pcp->list); | |
2396 | ||
2397 | pcp = &p->pcp[1]; /* cold*/ | |
2398 | pcp->count = 0; | |
2caaad41 | 2399 | pcp->high = 2 * batch; |
e46a5e28 | 2400 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
2401 | INIT_LIST_HEAD(&pcp->list); |
2402 | } | |
2403 | ||
8ad4b1fb RS |
2404 | /* |
2405 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2406 | * to the value high for the pageset p. | |
2407 | */ | |
2408 | ||
2409 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2410 | unsigned long high) | |
2411 | { | |
2412 | struct per_cpu_pages *pcp; | |
2413 | ||
2414 | pcp = &p->pcp[0]; /* hot list */ | |
2415 | pcp->high = high; | |
2416 | pcp->batch = max(1UL, high/4); | |
2417 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2418 | pcp->batch = PAGE_SHIFT * 8; | |
2419 | } | |
2420 | ||
2421 | ||
e7c8d5c9 CL |
2422 | #ifdef CONFIG_NUMA |
2423 | /* | |
2caaad41 CL |
2424 | * Boot pageset table. One per cpu which is going to be used for all |
2425 | * zones and all nodes. The parameters will be set in such a way | |
2426 | * that an item put on a list will immediately be handed over to | |
2427 | * the buddy list. This is safe since pageset manipulation is done | |
2428 | * with interrupts disabled. | |
2429 | * | |
2430 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2431 | * |
2432 | * The boot_pagesets must be kept even after bootup is complete for | |
2433 | * unused processors and/or zones. They do play a role for bootstrapping | |
2434 | * hotplugged processors. | |
2435 | * | |
2436 | * zoneinfo_show() and maybe other functions do | |
2437 | * not check if the processor is online before following the pageset pointer. | |
2438 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2439 | */ |
88a2a4ac | 2440 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2441 | |
2442 | /* | |
2443 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2444 | * per cpu pageset array in struct zone. |
2445 | */ | |
6292d9aa | 2446 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2447 | { |
2448 | struct zone *zone, *dzone; | |
37c0708d CL |
2449 | int node = cpu_to_node(cpu); |
2450 | ||
2451 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 CL |
2452 | |
2453 | for_each_zone(zone) { | |
e7c8d5c9 | 2454 | |
66a55030 CL |
2455 | if (!populated_zone(zone)) |
2456 | continue; | |
2457 | ||
23316bc8 | 2458 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2459 | GFP_KERNEL, node); |
23316bc8 | 2460 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2461 | goto bad; |
e7c8d5c9 | 2462 | |
23316bc8 | 2463 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2464 | |
2465 | if (percpu_pagelist_fraction) | |
2466 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2467 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2468 | } |
2469 | ||
2470 | return 0; | |
2471 | bad: | |
2472 | for_each_zone(dzone) { | |
64191688 AM |
2473 | if (!populated_zone(dzone)) |
2474 | continue; | |
e7c8d5c9 CL |
2475 | if (dzone == zone) |
2476 | break; | |
23316bc8 NP |
2477 | kfree(zone_pcp(dzone, cpu)); |
2478 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2479 | } |
2480 | return -ENOMEM; | |
2481 | } | |
2482 | ||
2483 | static inline void free_zone_pagesets(int cpu) | |
2484 | { | |
e7c8d5c9 CL |
2485 | struct zone *zone; |
2486 | ||
2487 | for_each_zone(zone) { | |
2488 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2489 | ||
f3ef9ead DR |
2490 | /* Free per_cpu_pageset if it is slab allocated */ |
2491 | if (pset != &boot_pageset[cpu]) | |
2492 | kfree(pset); | |
e7c8d5c9 | 2493 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2494 | } |
e7c8d5c9 CL |
2495 | } |
2496 | ||
9c7b216d | 2497 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2498 | unsigned long action, |
2499 | void *hcpu) | |
2500 | { | |
2501 | int cpu = (long)hcpu; | |
2502 | int ret = NOTIFY_OK; | |
2503 | ||
2504 | switch (action) { | |
ce421c79 | 2505 | case CPU_UP_PREPARE: |
8bb78442 | 2506 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2507 | if (process_zones(cpu)) |
2508 | ret = NOTIFY_BAD; | |
2509 | break; | |
2510 | case CPU_UP_CANCELED: | |
8bb78442 | 2511 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2512 | case CPU_DEAD: |
8bb78442 | 2513 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2514 | free_zone_pagesets(cpu); |
2515 | break; | |
2516 | default: | |
2517 | break; | |
e7c8d5c9 CL |
2518 | } |
2519 | return ret; | |
2520 | } | |
2521 | ||
74b85f37 | 2522 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2523 | { &pageset_cpuup_callback, NULL, 0 }; |
2524 | ||
78d9955b | 2525 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2526 | { |
2527 | int err; | |
2528 | ||
2529 | /* Initialize per_cpu_pageset for cpu 0. | |
2530 | * A cpuup callback will do this for every cpu | |
2531 | * as it comes online | |
2532 | */ | |
2533 | err = process_zones(smp_processor_id()); | |
2534 | BUG_ON(err); | |
2535 | register_cpu_notifier(&pageset_notifier); | |
2536 | } | |
2537 | ||
2538 | #endif | |
2539 | ||
577a32f6 | 2540 | static noinline __init_refok |
cca448fe | 2541 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2542 | { |
2543 | int i; | |
2544 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2545 | size_t alloc_size; |
ed8ece2e DH |
2546 | |
2547 | /* | |
2548 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2549 | * per zone. | |
2550 | */ | |
02b694de YG |
2551 | zone->wait_table_hash_nr_entries = |
2552 | wait_table_hash_nr_entries(zone_size_pages); | |
2553 | zone->wait_table_bits = | |
2554 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2555 | alloc_size = zone->wait_table_hash_nr_entries |
2556 | * sizeof(wait_queue_head_t); | |
2557 | ||
2558 | if (system_state == SYSTEM_BOOTING) { | |
2559 | zone->wait_table = (wait_queue_head_t *) | |
2560 | alloc_bootmem_node(pgdat, alloc_size); | |
2561 | } else { | |
2562 | /* | |
2563 | * This case means that a zone whose size was 0 gets new memory | |
2564 | * via memory hot-add. | |
2565 | * But it may be the case that a new node was hot-added. In | |
2566 | * this case vmalloc() will not be able to use this new node's | |
2567 | * memory - this wait_table must be initialized to use this new | |
2568 | * node itself as well. | |
2569 | * To use this new node's memory, further consideration will be | |
2570 | * necessary. | |
2571 | */ | |
8691f3a7 | 2572 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
2573 | } |
2574 | if (!zone->wait_table) | |
2575 | return -ENOMEM; | |
ed8ece2e | 2576 | |
02b694de | 2577 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2578 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2579 | |
2580 | return 0; | |
ed8ece2e DH |
2581 | } |
2582 | ||
c09b4240 | 2583 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2584 | { |
2585 | int cpu; | |
2586 | unsigned long batch = zone_batchsize(zone); | |
2587 | ||
2588 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2589 | #ifdef CONFIG_NUMA | |
2590 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2591 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2592 | setup_pageset(&boot_pageset[cpu],0); |
2593 | #else | |
2594 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2595 | #endif | |
2596 | } | |
f5335c0f AB |
2597 | if (zone->present_pages) |
2598 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2599 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2600 | } |
2601 | ||
718127cc YG |
2602 | __meminit int init_currently_empty_zone(struct zone *zone, |
2603 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2604 | unsigned long size, |
2605 | enum memmap_context context) | |
ed8ece2e DH |
2606 | { |
2607 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2608 | int ret; |
2609 | ret = zone_wait_table_init(zone, size); | |
2610 | if (ret) | |
2611 | return ret; | |
ed8ece2e DH |
2612 | pgdat->nr_zones = zone_idx(zone) + 1; |
2613 | ||
ed8ece2e DH |
2614 | zone->zone_start_pfn = zone_start_pfn; |
2615 | ||
2616 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2617 | ||
2618 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2619 | |
2620 | return 0; | |
ed8ece2e DH |
2621 | } |
2622 | ||
c713216d MG |
2623 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2624 | /* | |
2625 | * Basic iterator support. Return the first range of PFNs for a node | |
2626 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2627 | */ | |
a3142c8e | 2628 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2629 | { |
2630 | int i; | |
2631 | ||
2632 | for (i = 0; i < nr_nodemap_entries; i++) | |
2633 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2634 | return i; | |
2635 | ||
2636 | return -1; | |
2637 | } | |
2638 | ||
2639 | /* | |
2640 | * Basic iterator support. Return the next active range of PFNs for a node | |
2641 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2642 | */ | |
a3142c8e | 2643 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2644 | { |
2645 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2646 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2647 | return index; | |
2648 | ||
2649 | return -1; | |
2650 | } | |
2651 | ||
2652 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2653 | /* | |
2654 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2655 | * Architectures may implement their own version but if add_active_range() | |
2656 | * was used and there are no special requirements, this is a convenient | |
2657 | * alternative | |
2658 | */ | |
6f076f5d | 2659 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2660 | { |
2661 | int i; | |
2662 | ||
2663 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2664 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2665 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2666 | ||
2667 | if (start_pfn <= pfn && pfn < end_pfn) | |
2668 | return early_node_map[i].nid; | |
2669 | } | |
2670 | ||
2671 | return 0; | |
2672 | } | |
2673 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2674 | ||
2675 | /* Basic iterator support to walk early_node_map[] */ | |
2676 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2677 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2678 | i = next_active_region_index_in_nid(i, nid)) | |
2679 | ||
2680 | /** | |
2681 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2682 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2683 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2684 | * |
2685 | * If an architecture guarantees that all ranges registered with | |
2686 | * add_active_ranges() contain no holes and may be freed, this | |
2687 | * this function may be used instead of calling free_bootmem() manually. | |
2688 | */ | |
2689 | void __init free_bootmem_with_active_regions(int nid, | |
2690 | unsigned long max_low_pfn) | |
2691 | { | |
2692 | int i; | |
2693 | ||
2694 | for_each_active_range_index_in_nid(i, nid) { | |
2695 | unsigned long size_pages = 0; | |
2696 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2697 | ||
2698 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2699 | continue; | |
2700 | ||
2701 | if (end_pfn > max_low_pfn) | |
2702 | end_pfn = max_low_pfn; | |
2703 | ||
2704 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2705 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2706 | PFN_PHYS(early_node_map[i].start_pfn), | |
2707 | size_pages << PAGE_SHIFT); | |
2708 | } | |
2709 | } | |
2710 | ||
2711 | /** | |
2712 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2713 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2714 | * |
2715 | * If an architecture guarantees that all ranges registered with | |
2716 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2717 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2718 | */ |
2719 | void __init sparse_memory_present_with_active_regions(int nid) | |
2720 | { | |
2721 | int i; | |
2722 | ||
2723 | for_each_active_range_index_in_nid(i, nid) | |
2724 | memory_present(early_node_map[i].nid, | |
2725 | early_node_map[i].start_pfn, | |
2726 | early_node_map[i].end_pfn); | |
2727 | } | |
2728 | ||
fb01439c MG |
2729 | /** |
2730 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2731 | * @nid: The nid of the node to push the boundary for | |
2732 | * @start_pfn: The start pfn of the node | |
2733 | * @end_pfn: The end pfn of the node | |
2734 | * | |
2735 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2736 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2737 | * be hotplugged even though no physical memory exists. This function allows | |
2738 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2739 | * be used later. | |
2740 | */ | |
2741 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2742 | void __init push_node_boundaries(unsigned int nid, | |
2743 | unsigned long start_pfn, unsigned long end_pfn) | |
2744 | { | |
2745 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2746 | nid, start_pfn, end_pfn); | |
2747 | ||
2748 | /* Initialise the boundary for this node if necessary */ | |
2749 | if (node_boundary_end_pfn[nid] == 0) | |
2750 | node_boundary_start_pfn[nid] = -1UL; | |
2751 | ||
2752 | /* Update the boundaries */ | |
2753 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2754 | node_boundary_start_pfn[nid] = start_pfn; | |
2755 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2756 | node_boundary_end_pfn[nid] = end_pfn; | |
2757 | } | |
2758 | ||
2759 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2760 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2761 | unsigned long *start_pfn, unsigned long *end_pfn) |
2762 | { | |
2763 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2764 | nid, *start_pfn, *end_pfn); | |
2765 | ||
2766 | /* Return if boundary information has not been provided */ | |
2767 | if (node_boundary_end_pfn[nid] == 0) | |
2768 | return; | |
2769 | ||
2770 | /* Check the boundaries and update if necessary */ | |
2771 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2772 | *start_pfn = node_boundary_start_pfn[nid]; | |
2773 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2774 | *end_pfn = node_boundary_end_pfn[nid]; | |
2775 | } | |
2776 | #else | |
2777 | void __init push_node_boundaries(unsigned int nid, | |
2778 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2779 | ||
98011f56 | 2780 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2781 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
2782 | #endif | |
2783 | ||
2784 | ||
c713216d MG |
2785 | /** |
2786 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2787 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2788 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2789 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2790 | * |
2791 | * It returns the start and end page frame of a node based on information | |
2792 | * provided by an arch calling add_active_range(). If called for a node | |
2793 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2794 | * PFNs will be 0. |
c713216d | 2795 | */ |
a3142c8e | 2796 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
2797 | unsigned long *start_pfn, unsigned long *end_pfn) |
2798 | { | |
2799 | int i; | |
2800 | *start_pfn = -1UL; | |
2801 | *end_pfn = 0; | |
2802 | ||
2803 | for_each_active_range_index_in_nid(i, nid) { | |
2804 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2805 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2806 | } | |
2807 | ||
633c0666 | 2808 | if (*start_pfn == -1UL) |
c713216d | 2809 | *start_pfn = 0; |
fb01439c MG |
2810 | |
2811 | /* Push the node boundaries out if requested */ | |
2812 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
2813 | } |
2814 | ||
2a1e274a MG |
2815 | /* |
2816 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
2817 | * assumption is made that zones within a node are ordered in monotonic | |
2818 | * increasing memory addresses so that the "highest" populated zone is used | |
2819 | */ | |
2820 | void __init find_usable_zone_for_movable(void) | |
2821 | { | |
2822 | int zone_index; | |
2823 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
2824 | if (zone_index == ZONE_MOVABLE) | |
2825 | continue; | |
2826 | ||
2827 | if (arch_zone_highest_possible_pfn[zone_index] > | |
2828 | arch_zone_lowest_possible_pfn[zone_index]) | |
2829 | break; | |
2830 | } | |
2831 | ||
2832 | VM_BUG_ON(zone_index == -1); | |
2833 | movable_zone = zone_index; | |
2834 | } | |
2835 | ||
2836 | /* | |
2837 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
2838 | * because it is sized independant of architecture. Unlike the other zones, | |
2839 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
2840 | * in each node depending on the size of each node and how evenly kernelcore | |
2841 | * is distributed. This helper function adjusts the zone ranges | |
2842 | * provided by the architecture for a given node by using the end of the | |
2843 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
2844 | * zones within a node are in order of monotonic increases memory addresses | |
2845 | */ | |
2846 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
2847 | unsigned long zone_type, | |
2848 | unsigned long node_start_pfn, | |
2849 | unsigned long node_end_pfn, | |
2850 | unsigned long *zone_start_pfn, | |
2851 | unsigned long *zone_end_pfn) | |
2852 | { | |
2853 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
2854 | if (zone_movable_pfn[nid]) { | |
2855 | /* Size ZONE_MOVABLE */ | |
2856 | if (zone_type == ZONE_MOVABLE) { | |
2857 | *zone_start_pfn = zone_movable_pfn[nid]; | |
2858 | *zone_end_pfn = min(node_end_pfn, | |
2859 | arch_zone_highest_possible_pfn[movable_zone]); | |
2860 | ||
2861 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
2862 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
2863 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
2864 | *zone_end_pfn = zone_movable_pfn[nid]; | |
2865 | ||
2866 | /* Check if this whole range is within ZONE_MOVABLE */ | |
2867 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
2868 | *zone_start_pfn = *zone_end_pfn; | |
2869 | } | |
2870 | } | |
2871 | ||
c713216d MG |
2872 | /* |
2873 | * Return the number of pages a zone spans in a node, including holes | |
2874 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2875 | */ | |
6ea6e688 | 2876 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
2877 | unsigned long zone_type, |
2878 | unsigned long *ignored) | |
2879 | { | |
2880 | unsigned long node_start_pfn, node_end_pfn; | |
2881 | unsigned long zone_start_pfn, zone_end_pfn; | |
2882 | ||
2883 | /* Get the start and end of the node and zone */ | |
2884 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2885 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2886 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
2887 | adjust_zone_range_for_zone_movable(nid, zone_type, |
2888 | node_start_pfn, node_end_pfn, | |
2889 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
2890 | |
2891 | /* Check that this node has pages within the zone's required range */ | |
2892 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2893 | return 0; | |
2894 | ||
2895 | /* Move the zone boundaries inside the node if necessary */ | |
2896 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2897 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2898 | ||
2899 | /* Return the spanned pages */ | |
2900 | return zone_end_pfn - zone_start_pfn; | |
2901 | } | |
2902 | ||
2903 | /* | |
2904 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 2905 | * then all holes in the requested range will be accounted for. |
c713216d | 2906 | */ |
a3142c8e | 2907 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
2908 | unsigned long range_start_pfn, |
2909 | unsigned long range_end_pfn) | |
2910 | { | |
2911 | int i = 0; | |
2912 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2913 | unsigned long start_pfn; | |
2914 | ||
2915 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2916 | i = first_active_region_index_in_nid(nid); | |
2917 | if (i == -1) | |
2918 | return 0; | |
2919 | ||
b5445f95 MG |
2920 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
2921 | ||
9c7cd687 MG |
2922 | /* Account for ranges before physical memory on this node */ |
2923 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 2924 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
2925 | |
2926 | /* Find all holes for the zone within the node */ | |
2927 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2928 | ||
2929 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2930 | if (prev_end_pfn >= range_end_pfn) | |
2931 | break; | |
2932 | ||
2933 | /* Make sure the end of the zone is not within the hole */ | |
2934 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2935 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2936 | ||
2937 | /* Update the hole size cound and move on */ | |
2938 | if (start_pfn > range_start_pfn) { | |
2939 | BUG_ON(prev_end_pfn > start_pfn); | |
2940 | hole_pages += start_pfn - prev_end_pfn; | |
2941 | } | |
2942 | prev_end_pfn = early_node_map[i].end_pfn; | |
2943 | } | |
2944 | ||
9c7cd687 MG |
2945 | /* Account for ranges past physical memory on this node */ |
2946 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 2947 | hole_pages += range_end_pfn - |
9c7cd687 MG |
2948 | max(range_start_pfn, prev_end_pfn); |
2949 | ||
c713216d MG |
2950 | return hole_pages; |
2951 | } | |
2952 | ||
2953 | /** | |
2954 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2955 | * @start_pfn: The start PFN to start searching for holes | |
2956 | * @end_pfn: The end PFN to stop searching for holes | |
2957 | * | |
88ca3b94 | 2958 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
2959 | */ |
2960 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2961 | unsigned long end_pfn) | |
2962 | { | |
2963 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2964 | } | |
2965 | ||
2966 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 2967 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
2968 | unsigned long zone_type, |
2969 | unsigned long *ignored) | |
2970 | { | |
9c7cd687 MG |
2971 | unsigned long node_start_pfn, node_end_pfn; |
2972 | unsigned long zone_start_pfn, zone_end_pfn; | |
2973 | ||
2974 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2975 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
2976 | node_start_pfn); | |
2977 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
2978 | node_end_pfn); | |
2979 | ||
2a1e274a MG |
2980 | adjust_zone_range_for_zone_movable(nid, zone_type, |
2981 | node_start_pfn, node_end_pfn, | |
2982 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 2983 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 2984 | } |
0e0b864e | 2985 | |
c713216d | 2986 | #else |
6ea6e688 | 2987 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
2988 | unsigned long zone_type, |
2989 | unsigned long *zones_size) | |
2990 | { | |
2991 | return zones_size[zone_type]; | |
2992 | } | |
2993 | ||
6ea6e688 | 2994 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
2995 | unsigned long zone_type, |
2996 | unsigned long *zholes_size) | |
2997 | { | |
2998 | if (!zholes_size) | |
2999 | return 0; | |
3000 | ||
3001 | return zholes_size[zone_type]; | |
3002 | } | |
0e0b864e | 3003 | |
c713216d MG |
3004 | #endif |
3005 | ||
a3142c8e | 3006 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3007 | unsigned long *zones_size, unsigned long *zholes_size) |
3008 | { | |
3009 | unsigned long realtotalpages, totalpages = 0; | |
3010 | enum zone_type i; | |
3011 | ||
3012 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3013 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3014 | zones_size); | |
3015 | pgdat->node_spanned_pages = totalpages; | |
3016 | ||
3017 | realtotalpages = totalpages; | |
3018 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3019 | realtotalpages -= | |
3020 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3021 | zholes_size); | |
3022 | pgdat->node_present_pages = realtotalpages; | |
3023 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3024 | realtotalpages); | |
3025 | } | |
3026 | ||
835c134e MG |
3027 | #ifndef CONFIG_SPARSEMEM |
3028 | /* | |
3029 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
3030 | * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up | |
3031 | * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally | |
3032 | * round what is now in bits to nearest long in bits, then return it in | |
3033 | * bytes. | |
3034 | */ | |
3035 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3036 | { | |
3037 | unsigned long usemapsize; | |
3038 | ||
3039 | usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES); | |
3040 | usemapsize = usemapsize >> (MAX_ORDER-1); | |
3041 | usemapsize *= NR_PAGEBLOCK_BITS; | |
3042 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3043 | ||
3044 | return usemapsize / 8; | |
3045 | } | |
3046 | ||
3047 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3048 | struct zone *zone, unsigned long zonesize) | |
3049 | { | |
3050 | unsigned long usemapsize = usemap_size(zonesize); | |
3051 | zone->pageblock_flags = NULL; | |
3052 | if (usemapsize) { | |
3053 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3054 | memset(zone->pageblock_flags, 0, usemapsize); | |
3055 | } | |
3056 | } | |
3057 | #else | |
3058 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3059 | struct zone *zone, unsigned long zonesize) {} | |
3060 | #endif /* CONFIG_SPARSEMEM */ | |
3061 | ||
1da177e4 LT |
3062 | /* |
3063 | * Set up the zone data structures: | |
3064 | * - mark all pages reserved | |
3065 | * - mark all memory queues empty | |
3066 | * - clear the memory bitmaps | |
3067 | */ | |
86356ab1 | 3068 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3069 | unsigned long *zones_size, unsigned long *zholes_size) |
3070 | { | |
2f1b6248 | 3071 | enum zone_type j; |
ed8ece2e | 3072 | int nid = pgdat->node_id; |
1da177e4 | 3073 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3074 | int ret; |
1da177e4 | 3075 | |
208d54e5 | 3076 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3077 | pgdat->nr_zones = 0; |
3078 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3079 | pgdat->kswapd_max_order = 0; | |
3080 | ||
3081 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3082 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3083 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 3084 | |
c713216d MG |
3085 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3086 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3087 | zholes_size); | |
1da177e4 | 3088 | |
0e0b864e MG |
3089 | /* |
3090 | * Adjust realsize so that it accounts for how much memory | |
3091 | * is used by this zone for memmap. This affects the watermark | |
3092 | * and per-cpu initialisations | |
3093 | */ | |
3094 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
3095 | if (realsize >= memmap_pages) { | |
3096 | realsize -= memmap_pages; | |
3097 | printk(KERN_DEBUG | |
3098 | " %s zone: %lu pages used for memmap\n", | |
3099 | zone_names[j], memmap_pages); | |
3100 | } else | |
3101 | printk(KERN_WARNING | |
3102 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3103 | zone_names[j], memmap_pages, realsize); | |
3104 | ||
6267276f CL |
3105 | /* Account for reserved pages */ |
3106 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3107 | realsize -= dma_reserve; |
6267276f CL |
3108 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
3109 | zone_names[0], dma_reserve); | |
0e0b864e MG |
3110 | } |
3111 | ||
98d2b0eb | 3112 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3113 | nr_kernel_pages += realsize; |
3114 | nr_all_pages += realsize; | |
3115 | ||
3116 | zone->spanned_pages = size; | |
3117 | zone->present_pages = realsize; | |
9614634f | 3118 | #ifdef CONFIG_NUMA |
d5f541ed | 3119 | zone->node = nid; |
8417bba4 | 3120 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3121 | / 100; |
0ff38490 | 3122 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3123 | #endif |
1da177e4 LT |
3124 | zone->name = zone_names[j]; |
3125 | spin_lock_init(&zone->lock); | |
3126 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3127 | zone_seqlock_init(zone); |
1da177e4 | 3128 | zone->zone_pgdat = pgdat; |
1da177e4 | 3129 | |
3bb1a852 | 3130 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3131 | |
ed8ece2e | 3132 | zone_pcp_init(zone); |
1da177e4 LT |
3133 | INIT_LIST_HEAD(&zone->active_list); |
3134 | INIT_LIST_HEAD(&zone->inactive_list); | |
3135 | zone->nr_scan_active = 0; | |
3136 | zone->nr_scan_inactive = 0; | |
2244b95a | 3137 | zap_zone_vm_stats(zone); |
53e9a615 | 3138 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
3139 | if (!size) |
3140 | continue; | |
3141 | ||
835c134e | 3142 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3143 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3144 | size, MEMMAP_EARLY); | |
718127cc | 3145 | BUG_ON(ret); |
1da177e4 | 3146 | zone_start_pfn += size; |
1da177e4 LT |
3147 | } |
3148 | } | |
3149 | ||
577a32f6 | 3150 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3151 | { |
1da177e4 LT |
3152 | /* Skip empty nodes */ |
3153 | if (!pgdat->node_spanned_pages) | |
3154 | return; | |
3155 | ||
d41dee36 | 3156 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3157 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3158 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3159 | unsigned long size, start, end; |
d41dee36 AW |
3160 | struct page *map; |
3161 | ||
e984bb43 BP |
3162 | /* |
3163 | * The zone's endpoints aren't required to be MAX_ORDER | |
3164 | * aligned but the node_mem_map endpoints must be in order | |
3165 | * for the buddy allocator to function correctly. | |
3166 | */ | |
3167 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3168 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3169 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3170 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3171 | map = alloc_remap(pgdat->node_id, size); |
3172 | if (!map) | |
3173 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3174 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3175 | } |
12d810c1 | 3176 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3177 | /* |
3178 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3179 | */ | |
c713216d | 3180 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3181 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3182 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3183 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
3184 | mem_map -= pgdat->node_start_pfn; | |
3185 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
3186 | } | |
1da177e4 | 3187 | #endif |
d41dee36 | 3188 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3189 | } |
3190 | ||
86356ab1 | 3191 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
3192 | unsigned long *zones_size, unsigned long node_start_pfn, |
3193 | unsigned long *zholes_size) | |
3194 | { | |
3195 | pgdat->node_id = nid; | |
3196 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3197 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3198 | |
3199 | alloc_node_mem_map(pgdat); | |
3200 | ||
3201 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3202 | } | |
3203 | ||
c713216d | 3204 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3205 | |
3206 | #if MAX_NUMNODES > 1 | |
3207 | /* | |
3208 | * Figure out the number of possible node ids. | |
3209 | */ | |
3210 | static void __init setup_nr_node_ids(void) | |
3211 | { | |
3212 | unsigned int node; | |
3213 | unsigned int highest = 0; | |
3214 | ||
3215 | for_each_node_mask(node, node_possible_map) | |
3216 | highest = node; | |
3217 | nr_node_ids = highest + 1; | |
3218 | } | |
3219 | #else | |
3220 | static inline void setup_nr_node_ids(void) | |
3221 | { | |
3222 | } | |
3223 | #endif | |
3224 | ||
c713216d MG |
3225 | /** |
3226 | * add_active_range - Register a range of PFNs backed by physical memory | |
3227 | * @nid: The node ID the range resides on | |
3228 | * @start_pfn: The start PFN of the available physical memory | |
3229 | * @end_pfn: The end PFN of the available physical memory | |
3230 | * | |
3231 | * These ranges are stored in an early_node_map[] and later used by | |
3232 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3233 | * range spans a memory hole, it is up to the architecture to ensure | |
3234 | * the memory is not freed by the bootmem allocator. If possible | |
3235 | * the range being registered will be merged with existing ranges. | |
3236 | */ | |
3237 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3238 | unsigned long end_pfn) | |
3239 | { | |
3240 | int i; | |
3241 | ||
3242 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
3243 | "%d entries of %d used\n", | |
3244 | nid, start_pfn, end_pfn, | |
3245 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3246 | ||
3247 | /* Merge with existing active regions if possible */ | |
3248 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3249 | if (early_node_map[i].nid != nid) | |
3250 | continue; | |
3251 | ||
3252 | /* Skip if an existing region covers this new one */ | |
3253 | if (start_pfn >= early_node_map[i].start_pfn && | |
3254 | end_pfn <= early_node_map[i].end_pfn) | |
3255 | return; | |
3256 | ||
3257 | /* Merge forward if suitable */ | |
3258 | if (start_pfn <= early_node_map[i].end_pfn && | |
3259 | end_pfn > early_node_map[i].end_pfn) { | |
3260 | early_node_map[i].end_pfn = end_pfn; | |
3261 | return; | |
3262 | } | |
3263 | ||
3264 | /* Merge backward if suitable */ | |
3265 | if (start_pfn < early_node_map[i].end_pfn && | |
3266 | end_pfn >= early_node_map[i].start_pfn) { | |
3267 | early_node_map[i].start_pfn = start_pfn; | |
3268 | return; | |
3269 | } | |
3270 | } | |
3271 | ||
3272 | /* Check that early_node_map is large enough */ | |
3273 | if (i >= MAX_ACTIVE_REGIONS) { | |
3274 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3275 | MAX_ACTIVE_REGIONS); | |
3276 | return; | |
3277 | } | |
3278 | ||
3279 | early_node_map[i].nid = nid; | |
3280 | early_node_map[i].start_pfn = start_pfn; | |
3281 | early_node_map[i].end_pfn = end_pfn; | |
3282 | nr_nodemap_entries = i + 1; | |
3283 | } | |
3284 | ||
3285 | /** | |
3286 | * shrink_active_range - Shrink an existing registered range of PFNs | |
3287 | * @nid: The node id the range is on that should be shrunk | |
3288 | * @old_end_pfn: The old end PFN of the range | |
3289 | * @new_end_pfn: The new PFN of the range | |
3290 | * | |
3291 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
3292 | * The map is kept at the end physical page range that has already been | |
3293 | * registered with add_active_range(). This function allows an arch to shrink | |
3294 | * an existing registered range. | |
3295 | */ | |
3296 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
3297 | unsigned long new_end_pfn) | |
3298 | { | |
3299 | int i; | |
3300 | ||
3301 | /* Find the old active region end and shrink */ | |
3302 | for_each_active_range_index_in_nid(i, nid) | |
3303 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
3304 | early_node_map[i].end_pfn = new_end_pfn; | |
3305 | break; | |
3306 | } | |
3307 | } | |
3308 | ||
3309 | /** | |
3310 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3311 | * |
c713216d MG |
3312 | * During discovery, it may be found that a table like SRAT is invalid |
3313 | * and an alternative discovery method must be used. This function removes | |
3314 | * all currently registered regions. | |
3315 | */ | |
88ca3b94 | 3316 | void __init remove_all_active_ranges(void) |
c713216d MG |
3317 | { |
3318 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3319 | nr_nodemap_entries = 0; | |
fb01439c MG |
3320 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3321 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3322 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3323 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3324 | } |
3325 | ||
3326 | /* Compare two active node_active_regions */ | |
3327 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3328 | { | |
3329 | struct node_active_region *arange = (struct node_active_region *)a; | |
3330 | struct node_active_region *brange = (struct node_active_region *)b; | |
3331 | ||
3332 | /* Done this way to avoid overflows */ | |
3333 | if (arange->start_pfn > brange->start_pfn) | |
3334 | return 1; | |
3335 | if (arange->start_pfn < brange->start_pfn) | |
3336 | return -1; | |
3337 | ||
3338 | return 0; | |
3339 | } | |
3340 | ||
3341 | /* sort the node_map by start_pfn */ | |
3342 | static void __init sort_node_map(void) | |
3343 | { | |
3344 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3345 | sizeof(struct node_active_region), | |
3346 | cmp_node_active_region, NULL); | |
3347 | } | |
3348 | ||
a6af2bc3 | 3349 | /* Find the lowest pfn for a node */ |
c713216d MG |
3350 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
3351 | { | |
3352 | int i; | |
a6af2bc3 | 3353 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3354 | |
c713216d MG |
3355 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3356 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3357 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3358 | |
a6af2bc3 MG |
3359 | if (min_pfn == ULONG_MAX) { |
3360 | printk(KERN_WARNING | |
3361 | "Could not find start_pfn for node %lu\n", nid); | |
3362 | return 0; | |
3363 | } | |
3364 | ||
3365 | return min_pfn; | |
c713216d MG |
3366 | } |
3367 | ||
3368 | /** | |
3369 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3370 | * | |
3371 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3372 | * add_active_range(). |
c713216d MG |
3373 | */ |
3374 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3375 | { | |
3376 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3377 | } | |
3378 | ||
3379 | /** | |
3380 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3381 | * | |
3382 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3383 | * add_active_range(). |
c713216d MG |
3384 | */ |
3385 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3386 | { | |
3387 | int i; | |
3388 | unsigned long max_pfn = 0; | |
3389 | ||
3390 | for (i = 0; i < nr_nodemap_entries; i++) | |
3391 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3392 | ||
3393 | return max_pfn; | |
3394 | } | |
3395 | ||
37b07e41 LS |
3396 | /* |
3397 | * early_calculate_totalpages() | |
3398 | * Sum pages in active regions for movable zone. | |
3399 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3400 | */ | |
7e63efef MG |
3401 | unsigned long __init early_calculate_totalpages(void) |
3402 | { | |
3403 | int i; | |
3404 | unsigned long totalpages = 0; | |
3405 | ||
37b07e41 LS |
3406 | for (i = 0; i < nr_nodemap_entries; i++) { |
3407 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3408 | early_node_map[i].start_pfn; |
37b07e41 LS |
3409 | totalpages += pages; |
3410 | if (pages) | |
3411 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3412 | } | |
3413 | return totalpages; | |
7e63efef MG |
3414 | } |
3415 | ||
2a1e274a MG |
3416 | /* |
3417 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3418 | * is spread evenly between nodes as long as the nodes have enough | |
3419 | * memory. When they don't, some nodes will have more kernelcore than | |
3420 | * others | |
3421 | */ | |
3422 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3423 | { | |
3424 | int i, nid; | |
3425 | unsigned long usable_startpfn; | |
3426 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3427 | unsigned long totalpages = early_calculate_totalpages(); |
3428 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3429 | |
7e63efef MG |
3430 | /* |
3431 | * If movablecore was specified, calculate what size of | |
3432 | * kernelcore that corresponds so that memory usable for | |
3433 | * any allocation type is evenly spread. If both kernelcore | |
3434 | * and movablecore are specified, then the value of kernelcore | |
3435 | * will be used for required_kernelcore if it's greater than | |
3436 | * what movablecore would have allowed. | |
3437 | */ | |
3438 | if (required_movablecore) { | |
7e63efef MG |
3439 | unsigned long corepages; |
3440 | ||
3441 | /* | |
3442 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3443 | * was requested by the user | |
3444 | */ | |
3445 | required_movablecore = | |
3446 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3447 | corepages = totalpages - required_movablecore; | |
3448 | ||
3449 | required_kernelcore = max(required_kernelcore, corepages); | |
3450 | } | |
3451 | ||
2a1e274a MG |
3452 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3453 | if (!required_kernelcore) | |
3454 | return; | |
3455 | ||
3456 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3457 | find_usable_zone_for_movable(); | |
3458 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3459 | ||
3460 | restart: | |
3461 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3462 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3463 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3464 | /* |
3465 | * Recalculate kernelcore_node if the division per node | |
3466 | * now exceeds what is necessary to satisfy the requested | |
3467 | * amount of memory for the kernel | |
3468 | */ | |
3469 | if (required_kernelcore < kernelcore_node) | |
3470 | kernelcore_node = required_kernelcore / usable_nodes; | |
3471 | ||
3472 | /* | |
3473 | * As the map is walked, we track how much memory is usable | |
3474 | * by the kernel using kernelcore_remaining. When it is | |
3475 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3476 | */ | |
3477 | kernelcore_remaining = kernelcore_node; | |
3478 | ||
3479 | /* Go through each range of PFNs within this node */ | |
3480 | for_each_active_range_index_in_nid(i, nid) { | |
3481 | unsigned long start_pfn, end_pfn; | |
3482 | unsigned long size_pages; | |
3483 | ||
3484 | start_pfn = max(early_node_map[i].start_pfn, | |
3485 | zone_movable_pfn[nid]); | |
3486 | end_pfn = early_node_map[i].end_pfn; | |
3487 | if (start_pfn >= end_pfn) | |
3488 | continue; | |
3489 | ||
3490 | /* Account for what is only usable for kernelcore */ | |
3491 | if (start_pfn < usable_startpfn) { | |
3492 | unsigned long kernel_pages; | |
3493 | kernel_pages = min(end_pfn, usable_startpfn) | |
3494 | - start_pfn; | |
3495 | ||
3496 | kernelcore_remaining -= min(kernel_pages, | |
3497 | kernelcore_remaining); | |
3498 | required_kernelcore -= min(kernel_pages, | |
3499 | required_kernelcore); | |
3500 | ||
3501 | /* Continue if range is now fully accounted */ | |
3502 | if (end_pfn <= usable_startpfn) { | |
3503 | ||
3504 | /* | |
3505 | * Push zone_movable_pfn to the end so | |
3506 | * that if we have to rebalance | |
3507 | * kernelcore across nodes, we will | |
3508 | * not double account here | |
3509 | */ | |
3510 | zone_movable_pfn[nid] = end_pfn; | |
3511 | continue; | |
3512 | } | |
3513 | start_pfn = usable_startpfn; | |
3514 | } | |
3515 | ||
3516 | /* | |
3517 | * The usable PFN range for ZONE_MOVABLE is from | |
3518 | * start_pfn->end_pfn. Calculate size_pages as the | |
3519 | * number of pages used as kernelcore | |
3520 | */ | |
3521 | size_pages = end_pfn - start_pfn; | |
3522 | if (size_pages > kernelcore_remaining) | |
3523 | size_pages = kernelcore_remaining; | |
3524 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3525 | ||
3526 | /* | |
3527 | * Some kernelcore has been met, update counts and | |
3528 | * break if the kernelcore for this node has been | |
3529 | * satisified | |
3530 | */ | |
3531 | required_kernelcore -= min(required_kernelcore, | |
3532 | size_pages); | |
3533 | kernelcore_remaining -= size_pages; | |
3534 | if (!kernelcore_remaining) | |
3535 | break; | |
3536 | } | |
3537 | } | |
3538 | ||
3539 | /* | |
3540 | * If there is still required_kernelcore, we do another pass with one | |
3541 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3542 | * along on the nodes that still have memory until kernelcore is | |
3543 | * satisified | |
3544 | */ | |
3545 | usable_nodes--; | |
3546 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3547 | goto restart; | |
3548 | ||
3549 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3550 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3551 | zone_movable_pfn[nid] = | |
3552 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3553 | } | |
3554 | ||
37b07e41 LS |
3555 | /* Any regular memory on that node ? */ |
3556 | static void check_for_regular_memory(pg_data_t *pgdat) | |
3557 | { | |
3558 | #ifdef CONFIG_HIGHMEM | |
3559 | enum zone_type zone_type; | |
3560 | ||
3561 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
3562 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3563 | if (zone->present_pages) | |
3564 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
3565 | } | |
3566 | #endif | |
3567 | } | |
3568 | ||
c713216d MG |
3569 | /** |
3570 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3571 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3572 | * |
3573 | * This will call free_area_init_node() for each active node in the system. | |
3574 | * Using the page ranges provided by add_active_range(), the size of each | |
3575 | * zone in each node and their holes is calculated. If the maximum PFN | |
3576 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3577 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3578 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3579 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3580 | * at arch_max_dma_pfn. | |
3581 | */ | |
3582 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3583 | { | |
3584 | unsigned long nid; | |
3585 | enum zone_type i; | |
3586 | ||
a6af2bc3 MG |
3587 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3588 | sort_node_map(); | |
3589 | ||
c713216d MG |
3590 | /* Record where the zone boundaries are */ |
3591 | memset(arch_zone_lowest_possible_pfn, 0, | |
3592 | sizeof(arch_zone_lowest_possible_pfn)); | |
3593 | memset(arch_zone_highest_possible_pfn, 0, | |
3594 | sizeof(arch_zone_highest_possible_pfn)); | |
3595 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3596 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3597 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3598 | if (i == ZONE_MOVABLE) |
3599 | continue; | |
c713216d MG |
3600 | arch_zone_lowest_possible_pfn[i] = |
3601 | arch_zone_highest_possible_pfn[i-1]; | |
3602 | arch_zone_highest_possible_pfn[i] = | |
3603 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3604 | } | |
2a1e274a MG |
3605 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3606 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3607 | ||
3608 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3609 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3610 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3611 | |
c713216d MG |
3612 | /* Print out the zone ranges */ |
3613 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3614 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3615 | if (i == ZONE_MOVABLE) | |
3616 | continue; | |
c713216d MG |
3617 | printk(" %-8s %8lu -> %8lu\n", |
3618 | zone_names[i], | |
3619 | arch_zone_lowest_possible_pfn[i], | |
3620 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3621 | } |
3622 | ||
3623 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3624 | printk("Movable zone start PFN for each node\n"); | |
3625 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3626 | if (zone_movable_pfn[i]) | |
3627 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3628 | } | |
c713216d MG |
3629 | |
3630 | /* Print out the early_node_map[] */ | |
3631 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3632 | for (i = 0; i < nr_nodemap_entries; i++) | |
3633 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
3634 | early_node_map[i].start_pfn, | |
3635 | early_node_map[i].end_pfn); | |
3636 | ||
3637 | /* Initialise every node */ | |
8ef82866 | 3638 | setup_nr_node_ids(); |
c713216d MG |
3639 | for_each_online_node(nid) { |
3640 | pg_data_t *pgdat = NODE_DATA(nid); | |
3641 | free_area_init_node(nid, pgdat, NULL, | |
3642 | find_min_pfn_for_node(nid), NULL); | |
37b07e41 LS |
3643 | |
3644 | /* Any memory on that node */ | |
3645 | if (pgdat->node_present_pages) | |
3646 | node_set_state(nid, N_HIGH_MEMORY); | |
3647 | check_for_regular_memory(pgdat); | |
c713216d MG |
3648 | } |
3649 | } | |
2a1e274a | 3650 | |
7e63efef | 3651 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3652 | { |
3653 | unsigned long long coremem; | |
3654 | if (!p) | |
3655 | return -EINVAL; | |
3656 | ||
3657 | coremem = memparse(p, &p); | |
7e63efef | 3658 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3659 | |
7e63efef | 3660 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3661 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3662 | ||
3663 | return 0; | |
3664 | } | |
ed7ed365 | 3665 | |
7e63efef MG |
3666 | /* |
3667 | * kernelcore=size sets the amount of memory for use for allocations that | |
3668 | * cannot be reclaimed or migrated. | |
3669 | */ | |
3670 | static int __init cmdline_parse_kernelcore(char *p) | |
3671 | { | |
3672 | return cmdline_parse_core(p, &required_kernelcore); | |
3673 | } | |
3674 | ||
3675 | /* | |
3676 | * movablecore=size sets the amount of memory for use for allocations that | |
3677 | * can be reclaimed or migrated. | |
3678 | */ | |
3679 | static int __init cmdline_parse_movablecore(char *p) | |
3680 | { | |
3681 | return cmdline_parse_core(p, &required_movablecore); | |
3682 | } | |
3683 | ||
ed7ed365 | 3684 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 3685 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 3686 | |
c713216d MG |
3687 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3688 | ||
0e0b864e | 3689 | /** |
88ca3b94 RD |
3690 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
3691 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
3692 | * |
3693 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
3694 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
3695 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
3696 | * function may optionally be used to account for unfreeable pages in the |
3697 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
3698 | * smaller per-cpu batchsize. | |
0e0b864e MG |
3699 | */ |
3700 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
3701 | { | |
3702 | dma_reserve = new_dma_reserve; | |
3703 | } | |
3704 | ||
93b7504e | 3705 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3706 | static bootmem_data_t contig_bootmem_data; |
3707 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
3708 | ||
3709 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 3710 | #endif |
1da177e4 LT |
3711 | |
3712 | void __init free_area_init(unsigned long *zones_size) | |
3713 | { | |
93b7504e | 3714 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3715 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3716 | } | |
1da177e4 | 3717 | |
1da177e4 LT |
3718 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3719 | unsigned long action, void *hcpu) | |
3720 | { | |
3721 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 3722 | |
8bb78442 | 3723 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
3724 | local_irq_disable(); |
3725 | __drain_pages(cpu); | |
f8891e5e | 3726 | vm_events_fold_cpu(cpu); |
1da177e4 | 3727 | local_irq_enable(); |
2244b95a | 3728 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3729 | } |
3730 | return NOTIFY_OK; | |
3731 | } | |
1da177e4 LT |
3732 | |
3733 | void __init page_alloc_init(void) | |
3734 | { | |
3735 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3736 | } | |
3737 | ||
cb45b0e9 HA |
3738 | /* |
3739 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3740 | * or min_free_kbytes changes. | |
3741 | */ | |
3742 | static void calculate_totalreserve_pages(void) | |
3743 | { | |
3744 | struct pglist_data *pgdat; | |
3745 | unsigned long reserve_pages = 0; | |
2f6726e5 | 3746 | enum zone_type i, j; |
cb45b0e9 HA |
3747 | |
3748 | for_each_online_pgdat(pgdat) { | |
3749 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3750 | struct zone *zone = pgdat->node_zones + i; | |
3751 | unsigned long max = 0; | |
3752 | ||
3753 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3754 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3755 | if (zone->lowmem_reserve[j] > max) | |
3756 | max = zone->lowmem_reserve[j]; | |
3757 | } | |
3758 | ||
3759 | /* we treat pages_high as reserved pages. */ | |
3760 | max += zone->pages_high; | |
3761 | ||
3762 | if (max > zone->present_pages) | |
3763 | max = zone->present_pages; | |
3764 | reserve_pages += max; | |
3765 | } | |
3766 | } | |
3767 | totalreserve_pages = reserve_pages; | |
3768 | } | |
3769 | ||
1da177e4 LT |
3770 | /* |
3771 | * setup_per_zone_lowmem_reserve - called whenever | |
3772 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3773 | * has a correct pages reserved value, so an adequate number of | |
3774 | * pages are left in the zone after a successful __alloc_pages(). | |
3775 | */ | |
3776 | static void setup_per_zone_lowmem_reserve(void) | |
3777 | { | |
3778 | struct pglist_data *pgdat; | |
2f6726e5 | 3779 | enum zone_type j, idx; |
1da177e4 | 3780 | |
ec936fc5 | 3781 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
3782 | for (j = 0; j < MAX_NR_ZONES; j++) { |
3783 | struct zone *zone = pgdat->node_zones + j; | |
3784 | unsigned long present_pages = zone->present_pages; | |
3785 | ||
3786 | zone->lowmem_reserve[j] = 0; | |
3787 | ||
2f6726e5 CL |
3788 | idx = j; |
3789 | while (idx) { | |
1da177e4 LT |
3790 | struct zone *lower_zone; |
3791 | ||
2f6726e5 CL |
3792 | idx--; |
3793 | ||
1da177e4 LT |
3794 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
3795 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
3796 | ||
3797 | lower_zone = pgdat->node_zones + idx; | |
3798 | lower_zone->lowmem_reserve[j] = present_pages / | |
3799 | sysctl_lowmem_reserve_ratio[idx]; | |
3800 | present_pages += lower_zone->present_pages; | |
3801 | } | |
3802 | } | |
3803 | } | |
cb45b0e9 HA |
3804 | |
3805 | /* update totalreserve_pages */ | |
3806 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3807 | } |
3808 | ||
88ca3b94 RD |
3809 | /** |
3810 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
3811 | * | |
3812 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
3813 | * with respect to min_free_kbytes. | |
1da177e4 | 3814 | */ |
3947be19 | 3815 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
3816 | { |
3817 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
3818 | unsigned long lowmem_pages = 0; | |
3819 | struct zone *zone; | |
3820 | unsigned long flags; | |
3821 | ||
3822 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
3823 | for_each_zone(zone) { | |
3824 | if (!is_highmem(zone)) | |
3825 | lowmem_pages += zone->present_pages; | |
3826 | } | |
3827 | ||
3828 | for_each_zone(zone) { | |
ac924c60 AM |
3829 | u64 tmp; |
3830 | ||
1da177e4 | 3831 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
3832 | tmp = (u64)pages_min * zone->present_pages; |
3833 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
3834 | if (is_highmem(zone)) { |
3835 | /* | |
669ed175 NP |
3836 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
3837 | * need highmem pages, so cap pages_min to a small | |
3838 | * value here. | |
3839 | * | |
3840 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
3841 | * deltas controls asynch page reclaim, and so should | |
3842 | * not be capped for highmem. | |
1da177e4 LT |
3843 | */ |
3844 | int min_pages; | |
3845 | ||
3846 | min_pages = zone->present_pages / 1024; | |
3847 | if (min_pages < SWAP_CLUSTER_MAX) | |
3848 | min_pages = SWAP_CLUSTER_MAX; | |
3849 | if (min_pages > 128) | |
3850 | min_pages = 128; | |
3851 | zone->pages_min = min_pages; | |
3852 | } else { | |
669ed175 NP |
3853 | /* |
3854 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
3855 | * proportionate to the zone's size. |
3856 | */ | |
669ed175 | 3857 | zone->pages_min = tmp; |
1da177e4 LT |
3858 | } |
3859 | ||
ac924c60 AM |
3860 | zone->pages_low = zone->pages_min + (tmp >> 2); |
3861 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
3862 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
3863 | } | |
cb45b0e9 HA |
3864 | |
3865 | /* update totalreserve_pages */ | |
3866 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3867 | } |
3868 | ||
3869 | /* | |
3870 | * Initialise min_free_kbytes. | |
3871 | * | |
3872 | * For small machines we want it small (128k min). For large machines | |
3873 | * we want it large (64MB max). But it is not linear, because network | |
3874 | * bandwidth does not increase linearly with machine size. We use | |
3875 | * | |
3876 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
3877 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
3878 | * | |
3879 | * which yields | |
3880 | * | |
3881 | * 16MB: 512k | |
3882 | * 32MB: 724k | |
3883 | * 64MB: 1024k | |
3884 | * 128MB: 1448k | |
3885 | * 256MB: 2048k | |
3886 | * 512MB: 2896k | |
3887 | * 1024MB: 4096k | |
3888 | * 2048MB: 5792k | |
3889 | * 4096MB: 8192k | |
3890 | * 8192MB: 11584k | |
3891 | * 16384MB: 16384k | |
3892 | */ | |
3893 | static int __init init_per_zone_pages_min(void) | |
3894 | { | |
3895 | unsigned long lowmem_kbytes; | |
3896 | ||
3897 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
3898 | ||
3899 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
3900 | if (min_free_kbytes < 128) | |
3901 | min_free_kbytes = 128; | |
3902 | if (min_free_kbytes > 65536) | |
3903 | min_free_kbytes = 65536; | |
3904 | setup_per_zone_pages_min(); | |
3905 | setup_per_zone_lowmem_reserve(); | |
3906 | return 0; | |
3907 | } | |
3908 | module_init(init_per_zone_pages_min) | |
3909 | ||
3910 | /* | |
3911 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
3912 | * that we can call two helper functions whenever min_free_kbytes | |
3913 | * changes. | |
3914 | */ | |
3915 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
3916 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3917 | { | |
3918 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
3919 | if (write) |
3920 | setup_per_zone_pages_min(); | |
1da177e4 LT |
3921 | return 0; |
3922 | } | |
3923 | ||
9614634f CL |
3924 | #ifdef CONFIG_NUMA |
3925 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
3926 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3927 | { | |
3928 | struct zone *zone; | |
3929 | int rc; | |
3930 | ||
3931 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3932 | if (rc) | |
3933 | return rc; | |
3934 | ||
3935 | for_each_zone(zone) | |
8417bba4 | 3936 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
3937 | sysctl_min_unmapped_ratio) / 100; |
3938 | return 0; | |
3939 | } | |
0ff38490 CL |
3940 | |
3941 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
3942 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3943 | { | |
3944 | struct zone *zone; | |
3945 | int rc; | |
3946 | ||
3947 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3948 | if (rc) | |
3949 | return rc; | |
3950 | ||
3951 | for_each_zone(zone) | |
3952 | zone->min_slab_pages = (zone->present_pages * | |
3953 | sysctl_min_slab_ratio) / 100; | |
3954 | return 0; | |
3955 | } | |
9614634f CL |
3956 | #endif |
3957 | ||
1da177e4 LT |
3958 | /* |
3959 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
3960 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
3961 | * whenever sysctl_lowmem_reserve_ratio changes. | |
3962 | * | |
3963 | * The reserve ratio obviously has absolutely no relation with the | |
3964 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
3965 | * if in function of the boot time zone sizes. | |
3966 | */ | |
3967 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
3968 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3969 | { | |
3970 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3971 | setup_per_zone_lowmem_reserve(); | |
3972 | return 0; | |
3973 | } | |
3974 | ||
8ad4b1fb RS |
3975 | /* |
3976 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
3977 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
3978 | * can have before it gets flushed back to buddy allocator. | |
3979 | */ | |
3980 | ||
3981 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
3982 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3983 | { | |
3984 | struct zone *zone; | |
3985 | unsigned int cpu; | |
3986 | int ret; | |
3987 | ||
3988 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3989 | if (!write || (ret == -EINVAL)) | |
3990 | return ret; | |
3991 | for_each_zone(zone) { | |
3992 | for_each_online_cpu(cpu) { | |
3993 | unsigned long high; | |
3994 | high = zone->present_pages / percpu_pagelist_fraction; | |
3995 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
3996 | } | |
3997 | } | |
3998 | return 0; | |
3999 | } | |
4000 | ||
f034b5d4 | 4001 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4002 | |
4003 | #ifdef CONFIG_NUMA | |
4004 | static int __init set_hashdist(char *str) | |
4005 | { | |
4006 | if (!str) | |
4007 | return 0; | |
4008 | hashdist = simple_strtoul(str, &str, 0); | |
4009 | return 1; | |
4010 | } | |
4011 | __setup("hashdist=", set_hashdist); | |
4012 | #endif | |
4013 | ||
4014 | /* | |
4015 | * allocate a large system hash table from bootmem | |
4016 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4017 | * quantity of entries | |
4018 | * - limit is the number of hash buckets, not the total allocation size | |
4019 | */ | |
4020 | void *__init alloc_large_system_hash(const char *tablename, | |
4021 | unsigned long bucketsize, | |
4022 | unsigned long numentries, | |
4023 | int scale, | |
4024 | int flags, | |
4025 | unsigned int *_hash_shift, | |
4026 | unsigned int *_hash_mask, | |
4027 | unsigned long limit) | |
4028 | { | |
4029 | unsigned long long max = limit; | |
4030 | unsigned long log2qty, size; | |
4031 | void *table = NULL; | |
4032 | ||
4033 | /* allow the kernel cmdline to have a say */ | |
4034 | if (!numentries) { | |
4035 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4036 | numentries = nr_kernel_pages; |
1da177e4 LT |
4037 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4038 | numentries >>= 20 - PAGE_SHIFT; | |
4039 | numentries <<= 20 - PAGE_SHIFT; | |
4040 | ||
4041 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4042 | if (scale > PAGE_SHIFT) | |
4043 | numentries >>= (scale - PAGE_SHIFT); | |
4044 | else | |
4045 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4046 | |
4047 | /* Make sure we've got at least a 0-order allocation.. */ | |
4048 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4049 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4050 | } |
6e692ed3 | 4051 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4052 | |
4053 | /* limit allocation size to 1/16 total memory by default */ | |
4054 | if (max == 0) { | |
4055 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4056 | do_div(max, bucketsize); | |
4057 | } | |
4058 | ||
4059 | if (numentries > max) | |
4060 | numentries = max; | |
4061 | ||
f0d1b0b3 | 4062 | log2qty = ilog2(numentries); |
1da177e4 LT |
4063 | |
4064 | do { | |
4065 | size = bucketsize << log2qty; | |
4066 | if (flags & HASH_EARLY) | |
4067 | table = alloc_bootmem(size); | |
4068 | else if (hashdist) | |
4069 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4070 | else { | |
4071 | unsigned long order; | |
4072 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
4073 | ; | |
4074 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
1037b83b ED |
4075 | /* |
4076 | * If bucketsize is not a power-of-two, we may free | |
4077 | * some pages at the end of hash table. | |
4078 | */ | |
4079 | if (table) { | |
4080 | unsigned long alloc_end = (unsigned long)table + | |
4081 | (PAGE_SIZE << order); | |
4082 | unsigned long used = (unsigned long)table + | |
4083 | PAGE_ALIGN(size); | |
4084 | split_page(virt_to_page(table), order); | |
4085 | while (used < alloc_end) { | |
4086 | free_page(used); | |
4087 | used += PAGE_SIZE; | |
4088 | } | |
4089 | } | |
1da177e4 LT |
4090 | } |
4091 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4092 | ||
4093 | if (!table) | |
4094 | panic("Failed to allocate %s hash table\n", tablename); | |
4095 | ||
b49ad484 | 4096 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4097 | tablename, |
4098 | (1U << log2qty), | |
f0d1b0b3 | 4099 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4100 | size); |
4101 | ||
4102 | if (_hash_shift) | |
4103 | *_hash_shift = log2qty; | |
4104 | if (_hash_mask) | |
4105 | *_hash_mask = (1 << log2qty) - 1; | |
4106 | ||
4107 | return table; | |
4108 | } | |
a117e66e KH |
4109 | |
4110 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
4111 | struct page *pfn_to_page(unsigned long pfn) |
4112 | { | |
67de6482 | 4113 | return __pfn_to_page(pfn); |
a117e66e KH |
4114 | } |
4115 | unsigned long page_to_pfn(struct page *page) | |
4116 | { | |
67de6482 | 4117 | return __page_to_pfn(page); |
a117e66e | 4118 | } |
a117e66e KH |
4119 | EXPORT_SYMBOL(pfn_to_page); |
4120 | EXPORT_SYMBOL(page_to_pfn); | |
4121 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 4122 | |
835c134e MG |
4123 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4124 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4125 | unsigned long pfn) | |
4126 | { | |
4127 | #ifdef CONFIG_SPARSEMEM | |
4128 | return __pfn_to_section(pfn)->pageblock_flags; | |
4129 | #else | |
4130 | return zone->pageblock_flags; | |
4131 | #endif /* CONFIG_SPARSEMEM */ | |
4132 | } | |
4133 | ||
4134 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4135 | { | |
4136 | #ifdef CONFIG_SPARSEMEM | |
4137 | pfn &= (PAGES_PER_SECTION-1); | |
4138 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4139 | #else | |
4140 | pfn = pfn - zone->zone_start_pfn; | |
4141 | return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS; | |
4142 | #endif /* CONFIG_SPARSEMEM */ | |
4143 | } | |
4144 | ||
4145 | /** | |
4146 | * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages | |
4147 | * @page: The page within the block of interest | |
4148 | * @start_bitidx: The first bit of interest to retrieve | |
4149 | * @end_bitidx: The last bit of interest | |
4150 | * returns pageblock_bits flags | |
4151 | */ | |
4152 | unsigned long get_pageblock_flags_group(struct page *page, | |
4153 | int start_bitidx, int end_bitidx) | |
4154 | { | |
4155 | struct zone *zone; | |
4156 | unsigned long *bitmap; | |
4157 | unsigned long pfn, bitidx; | |
4158 | unsigned long flags = 0; | |
4159 | unsigned long value = 1; | |
4160 | ||
4161 | zone = page_zone(page); | |
4162 | pfn = page_to_pfn(page); | |
4163 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4164 | bitidx = pfn_to_bitidx(zone, pfn); | |
4165 | ||
4166 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4167 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4168 | flags |= value; | |
6220ec78 | 4169 | |
835c134e MG |
4170 | return flags; |
4171 | } | |
4172 | ||
4173 | /** | |
4174 | * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages | |
4175 | * @page: The page within the block of interest | |
4176 | * @start_bitidx: The first bit of interest | |
4177 | * @end_bitidx: The last bit of interest | |
4178 | * @flags: The flags to set | |
4179 | */ | |
4180 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4181 | int start_bitidx, int end_bitidx) | |
4182 | { | |
4183 | struct zone *zone; | |
4184 | unsigned long *bitmap; | |
4185 | unsigned long pfn, bitidx; | |
4186 | unsigned long value = 1; | |
4187 | ||
4188 | zone = page_zone(page); | |
4189 | pfn = page_to_pfn(page); | |
4190 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4191 | bitidx = pfn_to_bitidx(zone, pfn); | |
4192 | ||
4193 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4194 | if (flags & value) | |
4195 | __set_bit(bitidx + start_bitidx, bitmap); | |
4196 | else | |
4197 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4198 | } |