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 | /* | |
50 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
51 | * initializer cleaner | |
52 | */ | |
c3d8c141 | 53 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 54 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 55 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 56 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b | 57 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 58 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 59 | long nr_swap_pages; |
8ad4b1fb | 60 | int percpu_pagelist_fraction; |
1da177e4 | 61 | |
d98c7a09 | 62 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 63 | |
1da177e4 LT |
64 | /* |
65 | * results with 256, 32 in the lowmem_reserve sysctl: | |
66 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
67 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
68 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
69 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
70 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
71 | * |
72 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
73 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 74 | */ |
2f1b6248 | 75 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 76 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 77 | 256, |
4b51d669 | 78 | #endif |
fb0e7942 | 79 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 80 | 256, |
fb0e7942 | 81 | #endif |
e53ef38d | 82 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 83 | 32 |
e53ef38d | 84 | #endif |
2f1b6248 | 85 | }; |
1da177e4 LT |
86 | |
87 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 88 | |
15ad7cdc | 89 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 90 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 91 | "DMA", |
4b51d669 | 92 | #endif |
fb0e7942 | 93 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 94 | "DMA32", |
fb0e7942 | 95 | #endif |
2f1b6248 | 96 | "Normal", |
e53ef38d | 97 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 98 | "HighMem" |
e53ef38d | 99 | #endif |
2f1b6248 CL |
100 | }; |
101 | ||
1da177e4 LT |
102 | int min_free_kbytes = 1024; |
103 | ||
86356ab1 YG |
104 | unsigned long __meminitdata nr_kernel_pages; |
105 | unsigned long __meminitdata nr_all_pages; | |
0e0b864e | 106 | static unsigned long __initdata dma_reserve; |
1da177e4 | 107 | |
c713216d MG |
108 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
109 | /* | |
110 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
111 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
112 | * Ranges passed to add_active_range() will be merged if possible | |
113 | * so the number of times add_active_range() can be called is | |
114 | * related to the number of nodes and the number of holes | |
115 | */ | |
116 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
117 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
118 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
119 | #else | |
120 | #if MAX_NUMNODES >= 32 | |
121 | /* If there can be many nodes, allow up to 50 holes per node */ | |
122 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
123 | #else | |
124 | /* By default, allow up to 256 distinct regions */ | |
125 | #define MAX_ACTIVE_REGIONS 256 | |
126 | #endif | |
127 | #endif | |
128 | ||
129 | struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS]; | |
130 | int __initdata nr_nodemap_entries; | |
131 | unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
132 | unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c MG |
133 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
134 | unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES]; | |
135 | unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES]; | |
136 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
137 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
138 | ||
13e7444b | 139 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 140 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 141 | { |
bdc8cb98 DH |
142 | int ret = 0; |
143 | unsigned seq; | |
144 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 145 | |
bdc8cb98 DH |
146 | do { |
147 | seq = zone_span_seqbegin(zone); | |
148 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
149 | ret = 1; | |
150 | else if (pfn < zone->zone_start_pfn) | |
151 | ret = 1; | |
152 | } while (zone_span_seqretry(zone, seq)); | |
153 | ||
154 | return ret; | |
c6a57e19 DH |
155 | } |
156 | ||
157 | static int page_is_consistent(struct zone *zone, struct page *page) | |
158 | { | |
1da177e4 LT |
159 | #ifdef CONFIG_HOLES_IN_ZONE |
160 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 161 | return 0; |
1da177e4 LT |
162 | #endif |
163 | if (zone != page_zone(page)) | |
c6a57e19 DH |
164 | return 0; |
165 | ||
166 | return 1; | |
167 | } | |
168 | /* | |
169 | * Temporary debugging check for pages not lying within a given zone. | |
170 | */ | |
171 | static int bad_range(struct zone *zone, struct page *page) | |
172 | { | |
173 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 174 | return 1; |
c6a57e19 DH |
175 | if (!page_is_consistent(zone, page)) |
176 | return 1; | |
177 | ||
1da177e4 LT |
178 | return 0; |
179 | } | |
13e7444b NP |
180 | #else |
181 | static inline int bad_range(struct zone *zone, struct page *page) | |
182 | { | |
183 | return 0; | |
184 | } | |
185 | #endif | |
186 | ||
224abf92 | 187 | static void bad_page(struct page *page) |
1da177e4 | 188 | { |
224abf92 | 189 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
190 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
191 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
192 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
193 | current->comm, page, (int)(2*sizeof(unsigned long)), |
194 | (unsigned long)page->flags, page->mapping, | |
195 | page_mapcount(page), page_count(page)); | |
1da177e4 | 196 | dump_stack(); |
334795ec HD |
197 | page->flags &= ~(1 << PG_lru | |
198 | 1 << PG_private | | |
1da177e4 | 199 | 1 << PG_locked | |
1da177e4 LT |
200 | 1 << PG_active | |
201 | 1 << PG_dirty | | |
334795ec HD |
202 | 1 << PG_reclaim | |
203 | 1 << PG_slab | | |
1da177e4 | 204 | 1 << PG_swapcache | |
676165a8 NP |
205 | 1 << PG_writeback | |
206 | 1 << PG_buddy ); | |
1da177e4 LT |
207 | set_page_count(page, 0); |
208 | reset_page_mapcount(page); | |
209 | page->mapping = NULL; | |
9f158333 | 210 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
211 | } |
212 | ||
1da177e4 LT |
213 | /* |
214 | * Higher-order pages are called "compound pages". They are structured thusly: | |
215 | * | |
216 | * The first PAGE_SIZE page is called the "head page". | |
217 | * | |
218 | * The remaining PAGE_SIZE pages are called "tail pages". | |
219 | * | |
220 | * All pages have PG_compound set. All pages have their ->private pointing at | |
221 | * the head page (even the head page has this). | |
222 | * | |
41d78ba5 HD |
223 | * The first tail page's ->lru.next holds the address of the compound page's |
224 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
225 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 226 | */ |
d98c7a09 HD |
227 | |
228 | static void free_compound_page(struct page *page) | |
229 | { | |
230 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
231 | } | |
232 | ||
1da177e4 LT |
233 | static void prep_compound_page(struct page *page, unsigned long order) |
234 | { | |
235 | int i; | |
236 | int nr_pages = 1 << order; | |
237 | ||
33f2ef89 | 238 | set_compound_page_dtor(page, free_compound_page); |
41d78ba5 | 239 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
240 | for (i = 0; i < nr_pages; i++) { |
241 | struct page *p = page + i; | |
242 | ||
5e9dace8 | 243 | __SetPageCompound(p); |
4c21e2f2 | 244 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
245 | } |
246 | } | |
247 | ||
248 | static void destroy_compound_page(struct page *page, unsigned long order) | |
249 | { | |
250 | int i; | |
251 | int nr_pages = 1 << order; | |
252 | ||
41d78ba5 | 253 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 254 | bad_page(page); |
1da177e4 LT |
255 | |
256 | for (i = 0; i < nr_pages; i++) { | |
257 | struct page *p = page + i; | |
258 | ||
224abf92 NP |
259 | if (unlikely(!PageCompound(p) | |
260 | (page_private(p) != (unsigned long)page))) | |
261 | bad_page(page); | |
5e9dace8 | 262 | __ClearPageCompound(p); |
1da177e4 LT |
263 | } |
264 | } | |
1da177e4 | 265 | |
17cf4406 NP |
266 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
267 | { | |
268 | int i; | |
269 | ||
725d704e | 270 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
271 | /* |
272 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
273 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
274 | */ | |
725d704e | 275 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
276 | for (i = 0; i < (1 << order); i++) |
277 | clear_highpage(page + i); | |
278 | } | |
279 | ||
1da177e4 LT |
280 | /* |
281 | * function for dealing with page's order in buddy system. | |
282 | * zone->lock is already acquired when we use these. | |
283 | * So, we don't need atomic page->flags operations here. | |
284 | */ | |
6aa3001b AM |
285 | static inline unsigned long page_order(struct page *page) |
286 | { | |
4c21e2f2 | 287 | return page_private(page); |
1da177e4 LT |
288 | } |
289 | ||
6aa3001b AM |
290 | static inline void set_page_order(struct page *page, int order) |
291 | { | |
4c21e2f2 | 292 | set_page_private(page, order); |
676165a8 | 293 | __SetPageBuddy(page); |
1da177e4 LT |
294 | } |
295 | ||
296 | static inline void rmv_page_order(struct page *page) | |
297 | { | |
676165a8 | 298 | __ClearPageBuddy(page); |
4c21e2f2 | 299 | set_page_private(page, 0); |
1da177e4 LT |
300 | } |
301 | ||
302 | /* | |
303 | * Locate the struct page for both the matching buddy in our | |
304 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
305 | * | |
306 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
307 | * the following equation: | |
308 | * B2 = B1 ^ (1 << O) | |
309 | * For example, if the starting buddy (buddy2) is #8 its order | |
310 | * 1 buddy is #10: | |
311 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
312 | * | |
313 | * 2) Any buddy B will have an order O+1 parent P which | |
314 | * satisfies the following equation: | |
315 | * P = B & ~(1 << O) | |
316 | * | |
d6e05edc | 317 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
318 | */ |
319 | static inline struct page * | |
320 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
321 | { | |
322 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
323 | ||
324 | return page + (buddy_idx - page_idx); | |
325 | } | |
326 | ||
327 | static inline unsigned long | |
328 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
329 | { | |
330 | return (page_idx & ~(1 << order)); | |
331 | } | |
332 | ||
333 | /* | |
334 | * This function checks whether a page is free && is the buddy | |
335 | * we can do coalesce a page and its buddy if | |
13e7444b | 336 | * (a) the buddy is not in a hole && |
676165a8 | 337 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
338 | * (c) a page and its buddy have the same order && |
339 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
340 | * |
341 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
342 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 343 | * |
676165a8 | 344 | * For recording page's order, we use page_private(page). |
1da177e4 | 345 | */ |
cb2b95e1 AW |
346 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
347 | int order) | |
1da177e4 | 348 | { |
13e7444b | 349 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 350 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
351 | return 0; |
352 | #endif | |
353 | ||
cb2b95e1 AW |
354 | if (page_zone_id(page) != page_zone_id(buddy)) |
355 | return 0; | |
356 | ||
357 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
358 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 359 | return 1; |
676165a8 | 360 | } |
6aa3001b | 361 | return 0; |
1da177e4 LT |
362 | } |
363 | ||
364 | /* | |
365 | * Freeing function for a buddy system allocator. | |
366 | * | |
367 | * The concept of a buddy system is to maintain direct-mapped table | |
368 | * (containing bit values) for memory blocks of various "orders". | |
369 | * The bottom level table contains the map for the smallest allocatable | |
370 | * units of memory (here, pages), and each level above it describes | |
371 | * pairs of units from the levels below, hence, "buddies". | |
372 | * At a high level, all that happens here is marking the table entry | |
373 | * at the bottom level available, and propagating the changes upward | |
374 | * as necessary, plus some accounting needed to play nicely with other | |
375 | * parts of the VM system. | |
376 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 377 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 378 | * order is recorded in page_private(page) field. |
1da177e4 LT |
379 | * So when we are allocating or freeing one, we can derive the state of the |
380 | * other. That is, if we allocate a small block, and both were | |
381 | * free, the remainder of the region must be split into blocks. | |
382 | * If a block is freed, and its buddy is also free, then this | |
383 | * triggers coalescing into a block of larger size. | |
384 | * | |
385 | * -- wli | |
386 | */ | |
387 | ||
48db57f8 | 388 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
389 | struct zone *zone, unsigned int order) |
390 | { | |
391 | unsigned long page_idx; | |
392 | int order_size = 1 << order; | |
393 | ||
224abf92 | 394 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
395 | destroy_compound_page(page, order); |
396 | ||
397 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
398 | ||
725d704e NP |
399 | VM_BUG_ON(page_idx & (order_size - 1)); |
400 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 401 | |
d23ad423 | 402 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
403 | while (order < MAX_ORDER-1) { |
404 | unsigned long combined_idx; | |
405 | struct free_area *area; | |
406 | struct page *buddy; | |
407 | ||
1da177e4 | 408 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 409 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 410 | break; /* Move the buddy up one level. */ |
13e7444b | 411 | |
1da177e4 LT |
412 | list_del(&buddy->lru); |
413 | area = zone->free_area + order; | |
414 | area->nr_free--; | |
415 | rmv_page_order(buddy); | |
13e7444b | 416 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
417 | page = page + (combined_idx - page_idx); |
418 | page_idx = combined_idx; | |
419 | order++; | |
420 | } | |
421 | set_page_order(page, order); | |
422 | list_add(&page->lru, &zone->free_area[order].free_list); | |
423 | zone->free_area[order].nr_free++; | |
424 | } | |
425 | ||
224abf92 | 426 | static inline int free_pages_check(struct page *page) |
1da177e4 | 427 | { |
92be2e33 NP |
428 | if (unlikely(page_mapcount(page) | |
429 | (page->mapping != NULL) | | |
430 | (page_count(page) != 0) | | |
1da177e4 LT |
431 | (page->flags & ( |
432 | 1 << PG_lru | | |
433 | 1 << PG_private | | |
434 | 1 << PG_locked | | |
435 | 1 << PG_active | | |
436 | 1 << PG_reclaim | | |
437 | 1 << PG_slab | | |
438 | 1 << PG_swapcache | | |
b5810039 | 439 | 1 << PG_writeback | |
676165a8 NP |
440 | 1 << PG_reserved | |
441 | 1 << PG_buddy )))) | |
224abf92 | 442 | bad_page(page); |
1da177e4 | 443 | if (PageDirty(page)) |
242e5468 | 444 | __ClearPageDirty(page); |
689bcebf HD |
445 | /* |
446 | * For now, we report if PG_reserved was found set, but do not | |
447 | * clear it, and do not free the page. But we shall soon need | |
448 | * to do more, for when the ZERO_PAGE count wraps negative. | |
449 | */ | |
450 | return PageReserved(page); | |
1da177e4 LT |
451 | } |
452 | ||
453 | /* | |
454 | * Frees a list of pages. | |
455 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 456 | * count is the number of pages to free. |
1da177e4 LT |
457 | * |
458 | * If the zone was previously in an "all pages pinned" state then look to | |
459 | * see if this freeing clears that state. | |
460 | * | |
461 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
462 | * pinned" detection logic. | |
463 | */ | |
48db57f8 NP |
464 | static void free_pages_bulk(struct zone *zone, int count, |
465 | struct list_head *list, int order) | |
1da177e4 | 466 | { |
c54ad30c | 467 | spin_lock(&zone->lock); |
1da177e4 LT |
468 | zone->all_unreclaimable = 0; |
469 | zone->pages_scanned = 0; | |
48db57f8 NP |
470 | while (count--) { |
471 | struct page *page; | |
472 | ||
725d704e | 473 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 474 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 475 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 476 | list_del(&page->lru); |
48db57f8 | 477 | __free_one_page(page, zone, order); |
1da177e4 | 478 | } |
c54ad30c | 479 | spin_unlock(&zone->lock); |
1da177e4 LT |
480 | } |
481 | ||
48db57f8 | 482 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 483 | { |
006d22d9 CL |
484 | spin_lock(&zone->lock); |
485 | zone->all_unreclaimable = 0; | |
486 | zone->pages_scanned = 0; | |
0798e519 | 487 | __free_one_page(page, zone, order); |
006d22d9 | 488 | spin_unlock(&zone->lock); |
48db57f8 NP |
489 | } |
490 | ||
491 | static void __free_pages_ok(struct page *page, unsigned int order) | |
492 | { | |
493 | unsigned long flags; | |
1da177e4 | 494 | int i; |
689bcebf | 495 | int reserved = 0; |
1da177e4 | 496 | |
1da177e4 | 497 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 498 | reserved += free_pages_check(page + i); |
689bcebf HD |
499 | if (reserved) |
500 | return; | |
501 | ||
9858db50 NP |
502 | if (!PageHighMem(page)) |
503 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 504 | arch_free_page(page, order); |
48db57f8 | 505 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 506 | |
c54ad30c | 507 | local_irq_save(flags); |
f8891e5e | 508 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 509 | free_one_page(page_zone(page), page, order); |
c54ad30c | 510 | local_irq_restore(flags); |
1da177e4 LT |
511 | } |
512 | ||
a226f6c8 DH |
513 | /* |
514 | * permit the bootmem allocator to evade page validation on high-order frees | |
515 | */ | |
516 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
517 | { | |
518 | if (order == 0) { | |
519 | __ClearPageReserved(page); | |
520 | set_page_count(page, 0); | |
7835e98b | 521 | set_page_refcounted(page); |
545b1ea9 | 522 | __free_page(page); |
a226f6c8 | 523 | } else { |
a226f6c8 DH |
524 | int loop; |
525 | ||
545b1ea9 | 526 | prefetchw(page); |
a226f6c8 DH |
527 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
528 | struct page *p = &page[loop]; | |
529 | ||
545b1ea9 NP |
530 | if (loop + 1 < BITS_PER_LONG) |
531 | prefetchw(p + 1); | |
a226f6c8 DH |
532 | __ClearPageReserved(p); |
533 | set_page_count(p, 0); | |
534 | } | |
535 | ||
7835e98b | 536 | set_page_refcounted(page); |
545b1ea9 | 537 | __free_pages(page, order); |
a226f6c8 DH |
538 | } |
539 | } | |
540 | ||
1da177e4 LT |
541 | |
542 | /* | |
543 | * The order of subdivision here is critical for the IO subsystem. | |
544 | * Please do not alter this order without good reasons and regression | |
545 | * testing. Specifically, as large blocks of memory are subdivided, | |
546 | * the order in which smaller blocks are delivered depends on the order | |
547 | * they're subdivided in this function. This is the primary factor | |
548 | * influencing the order in which pages are delivered to the IO | |
549 | * subsystem according to empirical testing, and this is also justified | |
550 | * by considering the behavior of a buddy system containing a single | |
551 | * large block of memory acted on by a series of small allocations. | |
552 | * This behavior is a critical factor in sglist merging's success. | |
553 | * | |
554 | * -- wli | |
555 | */ | |
085cc7d5 | 556 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
557 | int low, int high, struct free_area *area) |
558 | { | |
559 | unsigned long size = 1 << high; | |
560 | ||
561 | while (high > low) { | |
562 | area--; | |
563 | high--; | |
564 | size >>= 1; | |
725d704e | 565 | VM_BUG_ON(bad_range(zone, &page[size])); |
1da177e4 LT |
566 | list_add(&page[size].lru, &area->free_list); |
567 | area->nr_free++; | |
568 | set_page_order(&page[size], high); | |
569 | } | |
1da177e4 LT |
570 | } |
571 | ||
1da177e4 LT |
572 | /* |
573 | * This page is about to be returned from the page allocator | |
574 | */ | |
17cf4406 | 575 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 576 | { |
92be2e33 NP |
577 | if (unlikely(page_mapcount(page) | |
578 | (page->mapping != NULL) | | |
579 | (page_count(page) != 0) | | |
334795ec HD |
580 | (page->flags & ( |
581 | 1 << PG_lru | | |
1da177e4 LT |
582 | 1 << PG_private | |
583 | 1 << PG_locked | | |
1da177e4 LT |
584 | 1 << PG_active | |
585 | 1 << PG_dirty | | |
586 | 1 << PG_reclaim | | |
334795ec | 587 | 1 << PG_slab | |
1da177e4 | 588 | 1 << PG_swapcache | |
b5810039 | 589 | 1 << PG_writeback | |
676165a8 NP |
590 | 1 << PG_reserved | |
591 | 1 << PG_buddy )))) | |
224abf92 | 592 | bad_page(page); |
1da177e4 | 593 | |
689bcebf HD |
594 | /* |
595 | * For now, we report if PG_reserved was found set, but do not | |
596 | * clear it, and do not allocate the page: as a safety net. | |
597 | */ | |
598 | if (PageReserved(page)) | |
599 | return 1; | |
600 | ||
1da177e4 LT |
601 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
602 | 1 << PG_referenced | 1 << PG_arch_1 | | |
5409bae0 | 603 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 604 | set_page_private(page, 0); |
7835e98b | 605 | set_page_refcounted(page); |
cc102509 NP |
606 | |
607 | arch_alloc_page(page, order); | |
1da177e4 | 608 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
609 | |
610 | if (gfp_flags & __GFP_ZERO) | |
611 | prep_zero_page(page, order, gfp_flags); | |
612 | ||
613 | if (order && (gfp_flags & __GFP_COMP)) | |
614 | prep_compound_page(page, order); | |
615 | ||
689bcebf | 616 | return 0; |
1da177e4 LT |
617 | } |
618 | ||
619 | /* | |
620 | * Do the hard work of removing an element from the buddy allocator. | |
621 | * Call me with the zone->lock already held. | |
622 | */ | |
623 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
624 | { | |
625 | struct free_area * area; | |
626 | unsigned int current_order; | |
627 | struct page *page; | |
628 | ||
629 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
630 | area = zone->free_area + current_order; | |
631 | if (list_empty(&area->free_list)) | |
632 | continue; | |
633 | ||
634 | page = list_entry(area->free_list.next, struct page, lru); | |
635 | list_del(&page->lru); | |
636 | rmv_page_order(page); | |
637 | area->nr_free--; | |
d23ad423 | 638 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); |
085cc7d5 NP |
639 | expand(zone, page, order, current_order, area); |
640 | return page; | |
1da177e4 LT |
641 | } |
642 | ||
643 | return NULL; | |
644 | } | |
645 | ||
646 | /* | |
647 | * Obtain a specified number of elements from the buddy allocator, all under | |
648 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
649 | * Returns the number of new pages which were placed at *list. | |
650 | */ | |
651 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
652 | unsigned long count, struct list_head *list) | |
653 | { | |
1da177e4 | 654 | int i; |
1da177e4 | 655 | |
c54ad30c | 656 | spin_lock(&zone->lock); |
1da177e4 | 657 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
658 | struct page *page = __rmqueue(zone, order); |
659 | if (unlikely(page == NULL)) | |
1da177e4 | 660 | break; |
1da177e4 LT |
661 | list_add_tail(&page->lru, list); |
662 | } | |
c54ad30c | 663 | spin_unlock(&zone->lock); |
085cc7d5 | 664 | return i; |
1da177e4 LT |
665 | } |
666 | ||
74c7aa8b CL |
667 | #if MAX_NUMNODES > 1 |
668 | int nr_node_ids __read_mostly; | |
669 | EXPORT_SYMBOL(nr_node_ids); | |
670 | ||
671 | /* | |
672 | * Figure out the number of possible node ids. | |
673 | */ | |
674 | static void __init setup_nr_node_ids(void) | |
675 | { | |
676 | unsigned int node; | |
677 | unsigned int highest = 0; | |
678 | ||
679 | for_each_node_mask(node, node_possible_map) | |
680 | highest = node; | |
681 | nr_node_ids = highest + 1; | |
682 | } | |
683 | #else | |
684 | static void __init setup_nr_node_ids(void) {} | |
685 | #endif | |
686 | ||
4ae7c039 | 687 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
688 | /* |
689 | * Called from the slab reaper to drain pagesets on a particular node that | |
39bbcb8f | 690 | * belongs to the currently executing processor. |
879336c3 CL |
691 | * Note that this function must be called with the thread pinned to |
692 | * a single processor. | |
8fce4d8e CL |
693 | */ |
694 | void drain_node_pages(int nodeid) | |
4ae7c039 | 695 | { |
2f6726e5 CL |
696 | int i; |
697 | enum zone_type z; | |
4ae7c039 CL |
698 | unsigned long flags; |
699 | ||
8fce4d8e CL |
700 | for (z = 0; z < MAX_NR_ZONES; z++) { |
701 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
702 | struct per_cpu_pageset *pset; |
703 | ||
39bbcb8f CL |
704 | if (!populated_zone(zone)) |
705 | continue; | |
706 | ||
23316bc8 | 707 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
708 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
709 | struct per_cpu_pages *pcp; | |
710 | ||
711 | pcp = &pset->pcp[i]; | |
879336c3 | 712 | if (pcp->count) { |
bc4ba393 CL |
713 | int to_drain; |
714 | ||
879336c3 | 715 | local_irq_save(flags); |
bc4ba393 CL |
716 | if (pcp->count >= pcp->batch) |
717 | to_drain = pcp->batch; | |
718 | else | |
719 | to_drain = pcp->count; | |
720 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
721 | pcp->count -= to_drain; | |
879336c3 CL |
722 | local_irq_restore(flags); |
723 | } | |
4ae7c039 CL |
724 | } |
725 | } | |
4ae7c039 CL |
726 | } |
727 | #endif | |
728 | ||
1da177e4 LT |
729 | static void __drain_pages(unsigned int cpu) |
730 | { | |
c54ad30c | 731 | unsigned long flags; |
1da177e4 LT |
732 | struct zone *zone; |
733 | int i; | |
734 | ||
735 | for_each_zone(zone) { | |
736 | struct per_cpu_pageset *pset; | |
737 | ||
f2e12bb2 CL |
738 | if (!populated_zone(zone)) |
739 | continue; | |
740 | ||
e7c8d5c9 | 741 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
742 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
743 | struct per_cpu_pages *pcp; | |
744 | ||
745 | pcp = &pset->pcp[i]; | |
c54ad30c | 746 | local_irq_save(flags); |
48db57f8 NP |
747 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
748 | pcp->count = 0; | |
c54ad30c | 749 | local_irq_restore(flags); |
1da177e4 LT |
750 | } |
751 | } | |
752 | } | |
1da177e4 LT |
753 | |
754 | #ifdef CONFIG_PM | |
755 | ||
756 | void mark_free_pages(struct zone *zone) | |
757 | { | |
f623f0db RW |
758 | unsigned long pfn, max_zone_pfn; |
759 | unsigned long flags; | |
1da177e4 LT |
760 | int order; |
761 | struct list_head *curr; | |
762 | ||
763 | if (!zone->spanned_pages) | |
764 | return; | |
765 | ||
766 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
767 | |
768 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
769 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
770 | if (pfn_valid(pfn)) { | |
771 | struct page *page = pfn_to_page(pfn); | |
772 | ||
773 | if (!PageNosave(page)) | |
774 | ClearPageNosaveFree(page); | |
775 | } | |
1da177e4 LT |
776 | |
777 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
778 | list_for_each(curr, &zone->free_area[order].free_list) { | |
f623f0db | 779 | unsigned long i; |
1da177e4 | 780 | |
f623f0db RW |
781 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
782 | for (i = 0; i < (1UL << order); i++) | |
783 | SetPageNosaveFree(pfn_to_page(pfn + i)); | |
784 | } | |
1da177e4 | 785 | |
1da177e4 LT |
786 | spin_unlock_irqrestore(&zone->lock, flags); |
787 | } | |
788 | ||
789 | /* | |
790 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
791 | */ | |
792 | void drain_local_pages(void) | |
793 | { | |
794 | unsigned long flags; | |
795 | ||
796 | local_irq_save(flags); | |
797 | __drain_pages(smp_processor_id()); | |
798 | local_irq_restore(flags); | |
799 | } | |
800 | #endif /* CONFIG_PM */ | |
801 | ||
1da177e4 LT |
802 | /* |
803 | * Free a 0-order page | |
804 | */ | |
1da177e4 LT |
805 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
806 | { | |
807 | struct zone *zone = page_zone(page); | |
808 | struct per_cpu_pages *pcp; | |
809 | unsigned long flags; | |
810 | ||
1da177e4 LT |
811 | if (PageAnon(page)) |
812 | page->mapping = NULL; | |
224abf92 | 813 | if (free_pages_check(page)) |
689bcebf HD |
814 | return; |
815 | ||
9858db50 NP |
816 | if (!PageHighMem(page)) |
817 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 818 | arch_free_page(page, 0); |
689bcebf HD |
819 | kernel_map_pages(page, 1, 0); |
820 | ||
e7c8d5c9 | 821 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 822 | local_irq_save(flags); |
f8891e5e | 823 | __count_vm_event(PGFREE); |
1da177e4 LT |
824 | list_add(&page->lru, &pcp->list); |
825 | pcp->count++; | |
48db57f8 NP |
826 | if (pcp->count >= pcp->high) { |
827 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
828 | pcp->count -= pcp->batch; | |
829 | } | |
1da177e4 LT |
830 | local_irq_restore(flags); |
831 | put_cpu(); | |
832 | } | |
833 | ||
834 | void fastcall free_hot_page(struct page *page) | |
835 | { | |
836 | free_hot_cold_page(page, 0); | |
837 | } | |
838 | ||
839 | void fastcall free_cold_page(struct page *page) | |
840 | { | |
841 | free_hot_cold_page(page, 1); | |
842 | } | |
843 | ||
8dfcc9ba NP |
844 | /* |
845 | * split_page takes a non-compound higher-order page, and splits it into | |
846 | * n (1<<order) sub-pages: page[0..n] | |
847 | * Each sub-page must be freed individually. | |
848 | * | |
849 | * Note: this is probably too low level an operation for use in drivers. | |
850 | * Please consult with lkml before using this in your driver. | |
851 | */ | |
852 | void split_page(struct page *page, unsigned int order) | |
853 | { | |
854 | int i; | |
855 | ||
725d704e NP |
856 | VM_BUG_ON(PageCompound(page)); |
857 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
858 | for (i = 1; i < (1 << order); i++) |
859 | set_page_refcounted(page + i); | |
8dfcc9ba | 860 | } |
8dfcc9ba | 861 | |
1da177e4 LT |
862 | /* |
863 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
864 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
865 | * or two. | |
866 | */ | |
a74609fa NP |
867 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
868 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
869 | { |
870 | unsigned long flags; | |
689bcebf | 871 | struct page *page; |
1da177e4 | 872 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 873 | int cpu; |
1da177e4 | 874 | |
689bcebf | 875 | again: |
a74609fa | 876 | cpu = get_cpu(); |
48db57f8 | 877 | if (likely(order == 0)) { |
1da177e4 LT |
878 | struct per_cpu_pages *pcp; |
879 | ||
a74609fa | 880 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 881 | local_irq_save(flags); |
a74609fa | 882 | if (!pcp->count) { |
941c7105 | 883 | pcp->count = rmqueue_bulk(zone, 0, |
1da177e4 | 884 | pcp->batch, &pcp->list); |
a74609fa NP |
885 | if (unlikely(!pcp->count)) |
886 | goto failed; | |
1da177e4 | 887 | } |
a74609fa NP |
888 | page = list_entry(pcp->list.next, struct page, lru); |
889 | list_del(&page->lru); | |
890 | pcp->count--; | |
7fb1d9fc | 891 | } else { |
1da177e4 LT |
892 | spin_lock_irqsave(&zone->lock, flags); |
893 | page = __rmqueue(zone, order); | |
a74609fa NP |
894 | spin_unlock(&zone->lock); |
895 | if (!page) | |
896 | goto failed; | |
1da177e4 LT |
897 | } |
898 | ||
f8891e5e | 899 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 900 | zone_statistics(zonelist, zone); |
a74609fa NP |
901 | local_irq_restore(flags); |
902 | put_cpu(); | |
1da177e4 | 903 | |
725d704e | 904 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 905 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 906 | goto again; |
1da177e4 | 907 | return page; |
a74609fa NP |
908 | |
909 | failed: | |
910 | local_irq_restore(flags); | |
911 | put_cpu(); | |
912 | return NULL; | |
1da177e4 LT |
913 | } |
914 | ||
7fb1d9fc | 915 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
916 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
917 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
918 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
919 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
920 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
921 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 922 | |
933e312e AM |
923 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
924 | ||
925 | static struct fail_page_alloc_attr { | |
926 | struct fault_attr attr; | |
927 | ||
928 | u32 ignore_gfp_highmem; | |
929 | u32 ignore_gfp_wait; | |
930 | ||
931 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
932 | ||
933 | struct dentry *ignore_gfp_highmem_file; | |
934 | struct dentry *ignore_gfp_wait_file; | |
935 | ||
936 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
937 | ||
938 | } fail_page_alloc = { | |
939 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
940 | .ignore_gfp_wait = 1, |
941 | .ignore_gfp_highmem = 1, | |
933e312e AM |
942 | }; |
943 | ||
944 | static int __init setup_fail_page_alloc(char *str) | |
945 | { | |
946 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
947 | } | |
948 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
949 | ||
950 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
951 | { | |
952 | if (gfp_mask & __GFP_NOFAIL) | |
953 | return 0; | |
954 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
955 | return 0; | |
956 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
957 | return 0; | |
958 | ||
959 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
960 | } | |
961 | ||
962 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
963 | ||
964 | static int __init fail_page_alloc_debugfs(void) | |
965 | { | |
966 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
967 | struct dentry *dir; | |
968 | int err; | |
969 | ||
970 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
971 | "fail_page_alloc"); | |
972 | if (err) | |
973 | return err; | |
974 | dir = fail_page_alloc.attr.dentries.dir; | |
975 | ||
976 | fail_page_alloc.ignore_gfp_wait_file = | |
977 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
978 | &fail_page_alloc.ignore_gfp_wait); | |
979 | ||
980 | fail_page_alloc.ignore_gfp_highmem_file = | |
981 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
982 | &fail_page_alloc.ignore_gfp_highmem); | |
983 | ||
984 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
985 | !fail_page_alloc.ignore_gfp_highmem_file) { | |
986 | err = -ENOMEM; | |
987 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
988 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
989 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); | |
990 | } | |
991 | ||
992 | return err; | |
993 | } | |
994 | ||
995 | late_initcall(fail_page_alloc_debugfs); | |
996 | ||
997 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
998 | ||
999 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1000 | ||
1001 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1002 | { | |
1003 | return 0; | |
1004 | } | |
1005 | ||
1006 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1007 | ||
1da177e4 LT |
1008 | /* |
1009 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1010 | * of the allocation. | |
1011 | */ | |
1012 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1013 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1014 | { |
1015 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1016 | long min = mark; |
1017 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1018 | int o; |
1019 | ||
7fb1d9fc | 1020 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1021 | min -= min / 2; |
7fb1d9fc | 1022 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1023 | min -= min / 4; |
1024 | ||
1025 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1026 | return 0; | |
1027 | for (o = 0; o < order; o++) { | |
1028 | /* At the next order, this order's pages become unavailable */ | |
1029 | free_pages -= z->free_area[o].nr_free << o; | |
1030 | ||
1031 | /* Require fewer higher order pages to be free */ | |
1032 | min >>= 1; | |
1033 | ||
1034 | if (free_pages <= min) | |
1035 | return 0; | |
1036 | } | |
1037 | return 1; | |
1038 | } | |
1039 | ||
9276b1bc PJ |
1040 | #ifdef CONFIG_NUMA |
1041 | /* | |
1042 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1043 | * skip over zones that are not allowed by the cpuset, or that have | |
1044 | * been recently (in last second) found to be nearly full. See further | |
1045 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1046 | * that have to skip over alot of full or unallowed zones. | |
1047 | * | |
1048 | * If the zonelist cache is present in the passed in zonelist, then | |
1049 | * returns a pointer to the allowed node mask (either the current | |
1050 | * tasks mems_allowed, or node_online_map.) | |
1051 | * | |
1052 | * If the zonelist cache is not available for this zonelist, does | |
1053 | * nothing and returns NULL. | |
1054 | * | |
1055 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1056 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1057 | * | |
1058 | * We hold off even calling zlc_setup, until after we've checked the | |
1059 | * first zone in the zonelist, on the theory that most allocations will | |
1060 | * be satisfied from that first zone, so best to examine that zone as | |
1061 | * quickly as we can. | |
1062 | */ | |
1063 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1064 | { | |
1065 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1066 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1067 | ||
1068 | zlc = zonelist->zlcache_ptr; | |
1069 | if (!zlc) | |
1070 | return NULL; | |
1071 | ||
1072 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1073 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1074 | zlc->last_full_zap = jiffies; | |
1075 | } | |
1076 | ||
1077 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1078 | &cpuset_current_mems_allowed : | |
1079 | &node_online_map; | |
1080 | return allowednodes; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1085 | * if it is worth looking at further for free memory: | |
1086 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1087 | * bit set in the zonelist_cache fullzones BITMAP). | |
1088 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1089 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1090 | * Return true (non-zero) if zone is worth looking at further, or | |
1091 | * else return false (zero) if it is not. | |
1092 | * | |
1093 | * This check -ignores- the distinction between various watermarks, | |
1094 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1095 | * found to be full for any variation of these watermarks, it will | |
1096 | * be considered full for up to one second by all requests, unless | |
1097 | * we are so low on memory on all allowed nodes that we are forced | |
1098 | * into the second scan of the zonelist. | |
1099 | * | |
1100 | * In the second scan we ignore this zonelist cache and exactly | |
1101 | * apply the watermarks to all zones, even it is slower to do so. | |
1102 | * We are low on memory in the second scan, and should leave no stone | |
1103 | * unturned looking for a free page. | |
1104 | */ | |
1105 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1106 | nodemask_t *allowednodes) | |
1107 | { | |
1108 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1109 | int i; /* index of *z in zonelist zones */ | |
1110 | int n; /* node that zone *z is on */ | |
1111 | ||
1112 | zlc = zonelist->zlcache_ptr; | |
1113 | if (!zlc) | |
1114 | return 1; | |
1115 | ||
1116 | i = z - zonelist->zones; | |
1117 | n = zlc->z_to_n[i]; | |
1118 | ||
1119 | /* This zone is worth trying if it is allowed but not full */ | |
1120 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1125 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1126 | * from that zone don't waste time re-examining it. | |
1127 | */ | |
1128 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1129 | { | |
1130 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1131 | int i; /* index of *z in zonelist zones */ | |
1132 | ||
1133 | zlc = zonelist->zlcache_ptr; | |
1134 | if (!zlc) | |
1135 | return; | |
1136 | ||
1137 | i = z - zonelist->zones; | |
1138 | ||
1139 | set_bit(i, zlc->fullzones); | |
1140 | } | |
1141 | ||
1142 | #else /* CONFIG_NUMA */ | |
1143 | ||
1144 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1145 | { | |
1146 | return NULL; | |
1147 | } | |
1148 | ||
1149 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1150 | nodemask_t *allowednodes) | |
1151 | { | |
1152 | return 1; | |
1153 | } | |
1154 | ||
1155 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1156 | { | |
1157 | } | |
1158 | #endif /* CONFIG_NUMA */ | |
1159 | ||
7fb1d9fc | 1160 | /* |
0798e519 | 1161 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1162 | * a page. |
1163 | */ | |
1164 | static struct page * | |
1165 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1166 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1167 | { |
9276b1bc | 1168 | struct zone **z; |
7fb1d9fc | 1169 | struct page *page = NULL; |
9276b1bc | 1170 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1171 | struct zone *zone; |
9276b1bc PJ |
1172 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1173 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1174 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
7fb1d9fc | 1175 | |
9276b1bc | 1176 | zonelist_scan: |
7fb1d9fc | 1177 | /* |
9276b1bc | 1178 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1179 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1180 | */ | |
9276b1bc PJ |
1181 | z = zonelist->zones; |
1182 | ||
7fb1d9fc | 1183 | do { |
9276b1bc PJ |
1184 | if (NUMA_BUILD && zlc_active && |
1185 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1186 | continue; | |
1192d526 | 1187 | zone = *z; |
08e0f6a9 | 1188 | if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) && |
1192d526 | 1189 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) |
9b819d20 | 1190 | break; |
7fb1d9fc | 1191 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1192 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1193 | goto try_next_zone; |
7fb1d9fc RS |
1194 | |
1195 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1196 | unsigned long mark; |
1197 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1198 | mark = zone->pages_min; |
3148890b | 1199 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1200 | mark = zone->pages_low; |
3148890b | 1201 | else |
1192d526 | 1202 | mark = zone->pages_high; |
0798e519 PJ |
1203 | if (!zone_watermark_ok(zone, order, mark, |
1204 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1205 | if (!zone_reclaim_mode || |
1192d526 | 1206 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1207 | goto this_zone_full; |
0798e519 | 1208 | } |
7fb1d9fc RS |
1209 | } |
1210 | ||
1192d526 | 1211 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1212 | if (page) |
7fb1d9fc | 1213 | break; |
9276b1bc PJ |
1214 | this_zone_full: |
1215 | if (NUMA_BUILD) | |
1216 | zlc_mark_zone_full(zonelist, z); | |
1217 | try_next_zone: | |
1218 | if (NUMA_BUILD && !did_zlc_setup) { | |
1219 | /* we do zlc_setup after the first zone is tried */ | |
1220 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1221 | zlc_active = 1; | |
1222 | did_zlc_setup = 1; | |
1223 | } | |
7fb1d9fc | 1224 | } while (*(++z) != NULL); |
9276b1bc PJ |
1225 | |
1226 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1227 | /* Disable zlc cache for second zonelist scan */ | |
1228 | zlc_active = 0; | |
1229 | goto zonelist_scan; | |
1230 | } | |
7fb1d9fc | 1231 | return page; |
753ee728 MH |
1232 | } |
1233 | ||
1da177e4 LT |
1234 | /* |
1235 | * This is the 'heart' of the zoned buddy allocator. | |
1236 | */ | |
1237 | struct page * fastcall | |
dd0fc66f | 1238 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1239 | struct zonelist *zonelist) |
1240 | { | |
260b2367 | 1241 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1242 | struct zone **z; |
1da177e4 LT |
1243 | struct page *page; |
1244 | struct reclaim_state reclaim_state; | |
1245 | struct task_struct *p = current; | |
1da177e4 | 1246 | int do_retry; |
7fb1d9fc | 1247 | int alloc_flags; |
1da177e4 LT |
1248 | int did_some_progress; |
1249 | ||
1250 | might_sleep_if(wait); | |
1251 | ||
933e312e AM |
1252 | if (should_fail_alloc_page(gfp_mask, order)) |
1253 | return NULL; | |
1254 | ||
6b1de916 | 1255 | restart: |
7fb1d9fc | 1256 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1257 | |
7fb1d9fc | 1258 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
1259 | /* Should this ever happen?? */ |
1260 | return NULL; | |
1261 | } | |
6b1de916 | 1262 | |
7fb1d9fc | 1263 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1264 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1265 | if (page) |
1266 | goto got_pg; | |
1da177e4 | 1267 | |
952f3b51 CL |
1268 | /* |
1269 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1270 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1271 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1272 | * using a larger set of nodes after it has established that the | |
1273 | * allowed per node queues are empty and that nodes are | |
1274 | * over allocated. | |
1275 | */ | |
1276 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1277 | goto nopage; | |
1278 | ||
0798e519 | 1279 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1280 | wakeup_kswapd(*z, order); |
1da177e4 | 1281 | |
9bf2229f | 1282 | /* |
7fb1d9fc RS |
1283 | * OK, we're below the kswapd watermark and have kicked background |
1284 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1285 | * to how we want to proceed. | |
1286 | * | |
1287 | * The caller may dip into page reserves a bit more if the caller | |
1288 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1289 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1290 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1291 | */ |
3148890b | 1292 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1293 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1294 | alloc_flags |= ALLOC_HARDER; | |
1295 | if (gfp_mask & __GFP_HIGH) | |
1296 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1297 | if (wait) |
1298 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1299 | |
1300 | /* | |
1301 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1302 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1303 | * |
1304 | * This is the last chance, in general, before the goto nopage. | |
1305 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1306 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1307 | */ |
7fb1d9fc RS |
1308 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1309 | if (page) | |
1310 | goto got_pg; | |
1da177e4 LT |
1311 | |
1312 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1313 | |
b43a57bb | 1314 | rebalance: |
b84a35be NP |
1315 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1316 | && !in_interrupt()) { | |
1317 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1318 | nofail_alloc: |
b84a35be | 1319 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1320 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1321 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1322 | if (page) |
1323 | goto got_pg; | |
885036d3 | 1324 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1325 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1326 | goto nofail_alloc; |
1327 | } | |
1da177e4 LT |
1328 | } |
1329 | goto nopage; | |
1330 | } | |
1331 | ||
1332 | /* Atomic allocations - we can't balance anything */ | |
1333 | if (!wait) | |
1334 | goto nopage; | |
1335 | ||
1da177e4 LT |
1336 | cond_resched(); |
1337 | ||
1338 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1339 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1340 | p->flags |= PF_MEMALLOC; |
1341 | reclaim_state.reclaimed_slab = 0; | |
1342 | p->reclaim_state = &reclaim_state; | |
1343 | ||
7fb1d9fc | 1344 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1345 | |
1346 | p->reclaim_state = NULL; | |
1347 | p->flags &= ~PF_MEMALLOC; | |
1348 | ||
1349 | cond_resched(); | |
1350 | ||
1351 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1352 | page = get_page_from_freelist(gfp_mask, order, |
1353 | zonelist, alloc_flags); | |
1354 | if (page) | |
1355 | goto got_pg; | |
1da177e4 LT |
1356 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1357 | /* | |
1358 | * Go through the zonelist yet one more time, keep | |
1359 | * very high watermark here, this is only to catch | |
1360 | * a parallel oom killing, we must fail if we're still | |
1361 | * under heavy pressure. | |
1362 | */ | |
7fb1d9fc | 1363 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1364 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1365 | if (page) |
1366 | goto got_pg; | |
1da177e4 | 1367 | |
9b0f8b04 | 1368 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1369 | goto restart; |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * Don't let big-order allocations loop unless the caller explicitly | |
1374 | * requests that. Wait for some write requests to complete then retry. | |
1375 | * | |
1376 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1377 | * <= 3, but that may not be true in other implementations. | |
1378 | */ | |
1379 | do_retry = 0; | |
1380 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1381 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1382 | do_retry = 1; | |
1383 | if (gfp_mask & __GFP_NOFAIL) | |
1384 | do_retry = 1; | |
1385 | } | |
1386 | if (do_retry) { | |
3fcfab16 | 1387 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1388 | goto rebalance; |
1389 | } | |
1390 | ||
1391 | nopage: | |
1392 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1393 | printk(KERN_WARNING "%s: page allocation failure." | |
1394 | " order:%d, mode:0x%x\n", | |
1395 | p->comm, order, gfp_mask); | |
1396 | dump_stack(); | |
578c2fd6 | 1397 | show_mem(); |
1da177e4 | 1398 | } |
1da177e4 | 1399 | got_pg: |
1da177e4 LT |
1400 | return page; |
1401 | } | |
1402 | ||
1403 | EXPORT_SYMBOL(__alloc_pages); | |
1404 | ||
1405 | /* | |
1406 | * Common helper functions. | |
1407 | */ | |
dd0fc66f | 1408 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1409 | { |
1410 | struct page * page; | |
1411 | page = alloc_pages(gfp_mask, order); | |
1412 | if (!page) | |
1413 | return 0; | |
1414 | return (unsigned long) page_address(page); | |
1415 | } | |
1416 | ||
1417 | EXPORT_SYMBOL(__get_free_pages); | |
1418 | ||
dd0fc66f | 1419 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1420 | { |
1421 | struct page * page; | |
1422 | ||
1423 | /* | |
1424 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1425 | * a highmem page | |
1426 | */ | |
725d704e | 1427 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1428 | |
1429 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1430 | if (page) | |
1431 | return (unsigned long) page_address(page); | |
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | EXPORT_SYMBOL(get_zeroed_page); | |
1436 | ||
1437 | void __pagevec_free(struct pagevec *pvec) | |
1438 | { | |
1439 | int i = pagevec_count(pvec); | |
1440 | ||
1441 | while (--i >= 0) | |
1442 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1443 | } | |
1444 | ||
1445 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1446 | { | |
b5810039 | 1447 | if (put_page_testzero(page)) { |
1da177e4 LT |
1448 | if (order == 0) |
1449 | free_hot_page(page); | |
1450 | else | |
1451 | __free_pages_ok(page, order); | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | EXPORT_SYMBOL(__free_pages); | |
1456 | ||
1457 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1458 | { | |
1459 | if (addr != 0) { | |
725d704e | 1460 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1461 | __free_pages(virt_to_page((void *)addr), order); |
1462 | } | |
1463 | } | |
1464 | ||
1465 | EXPORT_SYMBOL(free_pages); | |
1466 | ||
1da177e4 LT |
1467 | static unsigned int nr_free_zone_pages(int offset) |
1468 | { | |
e310fd43 MB |
1469 | /* Just pick one node, since fallback list is circular */ |
1470 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1471 | unsigned int sum = 0; |
1472 | ||
e310fd43 MB |
1473 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1474 | struct zone **zonep = zonelist->zones; | |
1475 | struct zone *zone; | |
1da177e4 | 1476 | |
e310fd43 MB |
1477 | for (zone = *zonep++; zone; zone = *zonep++) { |
1478 | unsigned long size = zone->present_pages; | |
1479 | unsigned long high = zone->pages_high; | |
1480 | if (size > high) | |
1481 | sum += size - high; | |
1da177e4 LT |
1482 | } |
1483 | ||
1484 | return sum; | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1489 | */ | |
1490 | unsigned int nr_free_buffer_pages(void) | |
1491 | { | |
af4ca457 | 1492 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1493 | } |
1494 | ||
1495 | /* | |
1496 | * Amount of free RAM allocatable within all zones | |
1497 | */ | |
1498 | unsigned int nr_free_pagecache_pages(void) | |
1499 | { | |
af4ca457 | 1500 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 | 1501 | } |
08e0f6a9 CL |
1502 | |
1503 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1504 | { |
08e0f6a9 | 1505 | if (NUMA_BUILD) |
25ba77c1 | 1506 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1507 | } |
1da177e4 | 1508 | |
1da177e4 LT |
1509 | void si_meminfo(struct sysinfo *val) |
1510 | { | |
1511 | val->totalram = totalram_pages; | |
1512 | val->sharedram = 0; | |
d23ad423 | 1513 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1514 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1515 | val->totalhigh = totalhigh_pages; |
1516 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1517 | val->mem_unit = PAGE_SIZE; |
1518 | } | |
1519 | ||
1520 | EXPORT_SYMBOL(si_meminfo); | |
1521 | ||
1522 | #ifdef CONFIG_NUMA | |
1523 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1524 | { | |
1525 | pg_data_t *pgdat = NODE_DATA(nid); | |
1526 | ||
1527 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1528 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1529 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1530 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1531 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1532 | NR_FREE_PAGES); | |
98d2b0eb CL |
1533 | #else |
1534 | val->totalhigh = 0; | |
1535 | val->freehigh = 0; | |
1536 | #endif | |
1da177e4 LT |
1537 | val->mem_unit = PAGE_SIZE; |
1538 | } | |
1539 | #endif | |
1540 | ||
1541 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1542 | ||
1543 | /* | |
1544 | * Show free area list (used inside shift_scroll-lock stuff) | |
1545 | * We also calculate the percentage fragmentation. We do this by counting the | |
1546 | * memory on each free list with the exception of the first item on the list. | |
1547 | */ | |
1548 | void show_free_areas(void) | |
1549 | { | |
c7241913 | 1550 | int cpu; |
1da177e4 LT |
1551 | struct zone *zone; |
1552 | ||
1553 | for_each_zone(zone) { | |
c7241913 | 1554 | if (!populated_zone(zone)) |
1da177e4 | 1555 | continue; |
c7241913 JS |
1556 | |
1557 | show_node(zone); | |
1558 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1559 | |
6b482c67 | 1560 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1561 | struct per_cpu_pageset *pageset; |
1562 | ||
e7c8d5c9 | 1563 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1564 | |
c7241913 JS |
1565 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1566 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1567 | cpu, pageset->pcp[0].high, | |
1568 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1569 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1570 | pageset->pcp[1].count); | |
1da177e4 LT |
1571 | } |
1572 | } | |
1573 | ||
a25700a5 | 1574 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1575 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1576 | global_page_state(NR_ACTIVE), |
1577 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1578 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1579 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1580 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1581 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1582 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1583 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1584 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1585 | global_page_state(NR_PAGETABLE), |
1586 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1587 | |
1588 | for_each_zone(zone) { | |
1589 | int i; | |
1590 | ||
c7241913 JS |
1591 | if (!populated_zone(zone)) |
1592 | continue; | |
1593 | ||
1da177e4 LT |
1594 | show_node(zone); |
1595 | printk("%s" | |
1596 | " free:%lukB" | |
1597 | " min:%lukB" | |
1598 | " low:%lukB" | |
1599 | " high:%lukB" | |
1600 | " active:%lukB" | |
1601 | " inactive:%lukB" | |
1602 | " present:%lukB" | |
1603 | " pages_scanned:%lu" | |
1604 | " all_unreclaimable? %s" | |
1605 | "\n", | |
1606 | zone->name, | |
d23ad423 | 1607 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1608 | K(zone->pages_min), |
1609 | K(zone->pages_low), | |
1610 | K(zone->pages_high), | |
c8785385 CL |
1611 | K(zone_page_state(zone, NR_ACTIVE)), |
1612 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1613 | K(zone->present_pages), |
1614 | zone->pages_scanned, | |
1615 | (zone->all_unreclaimable ? "yes" : "no") | |
1616 | ); | |
1617 | printk("lowmem_reserve[]:"); | |
1618 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1619 | printk(" %lu", zone->lowmem_reserve[i]); | |
1620 | printk("\n"); | |
1621 | } | |
1622 | ||
1623 | for_each_zone(zone) { | |
8f9de51a | 1624 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1625 | |
c7241913 JS |
1626 | if (!populated_zone(zone)) |
1627 | continue; | |
1628 | ||
1da177e4 LT |
1629 | show_node(zone); |
1630 | printk("%s: ", zone->name); | |
1da177e4 LT |
1631 | |
1632 | spin_lock_irqsave(&zone->lock, flags); | |
1633 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1634 | nr[order] = zone->free_area[order].nr_free; |
1635 | total += nr[order] << order; | |
1da177e4 LT |
1636 | } |
1637 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1638 | for (order = 0; order < MAX_ORDER; order++) |
1639 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1640 | printk("= %lukB\n", K(total)); |
1641 | } | |
1642 | ||
1643 | show_swap_cache_info(); | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1648 | * |
1649 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1650 | */ |
86356ab1 | 1651 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
2f6726e5 | 1652 | struct zonelist *zonelist, int nr_zones, enum zone_type zone_type) |
1da177e4 | 1653 | { |
1a93205b CL |
1654 | struct zone *zone; |
1655 | ||
98d2b0eb | 1656 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1657 | zone_type++; |
02a68a5e CL |
1658 | |
1659 | do { | |
2f6726e5 | 1660 | zone_type--; |
070f8032 | 1661 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1662 | if (populated_zone(zone)) { |
070f8032 CL |
1663 | zonelist->zones[nr_zones++] = zone; |
1664 | check_highest_zone(zone_type); | |
1da177e4 | 1665 | } |
02a68a5e | 1666 | |
2f6726e5 | 1667 | } while (zone_type); |
070f8032 | 1668 | return nr_zones; |
1da177e4 LT |
1669 | } |
1670 | ||
1671 | #ifdef CONFIG_NUMA | |
1672 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1673 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1674 | /** |
4dc3b16b | 1675 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1676 | * @node: node whose fallback list we're appending |
1677 | * @used_node_mask: nodemask_t of already used nodes | |
1678 | * | |
1679 | * We use a number of factors to determine which is the next node that should | |
1680 | * appear on a given node's fallback list. The node should not have appeared | |
1681 | * already in @node's fallback list, and it should be the next closest node | |
1682 | * according to the distance array (which contains arbitrary distance values | |
1683 | * from each node to each node in the system), and should also prefer nodes | |
1684 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1685 | * on them otherwise. | |
1686 | * It returns -1 if no node is found. | |
1687 | */ | |
86356ab1 | 1688 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1689 | { |
4cf808eb | 1690 | int n, val; |
1da177e4 LT |
1691 | int min_val = INT_MAX; |
1692 | int best_node = -1; | |
1693 | ||
4cf808eb LT |
1694 | /* Use the local node if we haven't already */ |
1695 | if (!node_isset(node, *used_node_mask)) { | |
1696 | node_set(node, *used_node_mask); | |
1697 | return node; | |
1698 | } | |
1da177e4 | 1699 | |
4cf808eb LT |
1700 | for_each_online_node(n) { |
1701 | cpumask_t tmp; | |
1da177e4 LT |
1702 | |
1703 | /* Don't want a node to appear more than once */ | |
1704 | if (node_isset(n, *used_node_mask)) | |
1705 | continue; | |
1706 | ||
1da177e4 LT |
1707 | /* Use the distance array to find the distance */ |
1708 | val = node_distance(node, n); | |
1709 | ||
4cf808eb LT |
1710 | /* Penalize nodes under us ("prefer the next node") */ |
1711 | val += (n < node); | |
1712 | ||
1da177e4 LT |
1713 | /* Give preference to headless and unused nodes */ |
1714 | tmp = node_to_cpumask(n); | |
1715 | if (!cpus_empty(tmp)) | |
1716 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1717 | ||
1718 | /* Slight preference for less loaded node */ | |
1719 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1720 | val += node_load[n]; | |
1721 | ||
1722 | if (val < min_val) { | |
1723 | min_val = val; | |
1724 | best_node = n; | |
1725 | } | |
1726 | } | |
1727 | ||
1728 | if (best_node >= 0) | |
1729 | node_set(best_node, *used_node_mask); | |
1730 | ||
1731 | return best_node; | |
1732 | } | |
1733 | ||
86356ab1 | 1734 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1735 | { |
19655d34 CL |
1736 | int j, node, local_node; |
1737 | enum zone_type i; | |
1da177e4 LT |
1738 | int prev_node, load; |
1739 | struct zonelist *zonelist; | |
1740 | nodemask_t used_mask; | |
1741 | ||
1742 | /* initialize zonelists */ | |
19655d34 | 1743 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1744 | zonelist = pgdat->node_zonelists + i; |
1745 | zonelist->zones[0] = NULL; | |
1746 | } | |
1747 | ||
1748 | /* NUMA-aware ordering of nodes */ | |
1749 | local_node = pgdat->node_id; | |
1750 | load = num_online_nodes(); | |
1751 | prev_node = local_node; | |
1752 | nodes_clear(used_mask); | |
1753 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1754 | int distance = node_distance(local_node, node); |
1755 | ||
1756 | /* | |
1757 | * If another node is sufficiently far away then it is better | |
1758 | * to reclaim pages in a zone before going off node. | |
1759 | */ | |
1760 | if (distance > RECLAIM_DISTANCE) | |
1761 | zone_reclaim_mode = 1; | |
1762 | ||
1da177e4 LT |
1763 | /* |
1764 | * We don't want to pressure a particular node. | |
1765 | * So adding penalty to the first node in same | |
1766 | * distance group to make it round-robin. | |
1767 | */ | |
9eeff239 CL |
1768 | |
1769 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1770 | node_load[node] += load; |
1771 | prev_node = node; | |
1772 | load--; | |
19655d34 | 1773 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1774 | zonelist = pgdat->node_zonelists + i; |
1775 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1776 | ||
19655d34 | 1777 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1778 | zonelist->zones[j] = NULL; |
1779 | } | |
1780 | } | |
1781 | } | |
1782 | ||
9276b1bc PJ |
1783 | /* Construct the zonelist performance cache - see further mmzone.h */ |
1784 | static void __meminit build_zonelist_cache(pg_data_t *pgdat) | |
1785 | { | |
1786 | int i; | |
1787 | ||
1788 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1789 | struct zonelist *zonelist; | |
1790 | struct zonelist_cache *zlc; | |
1791 | struct zone **z; | |
1792 | ||
1793 | zonelist = pgdat->node_zonelists + i; | |
1794 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
1795 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1796 | for (z = zonelist->zones; *z; z++) | |
1797 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
1798 | } | |
1799 | } | |
1800 | ||
1da177e4 LT |
1801 | #else /* CONFIG_NUMA */ |
1802 | ||
86356ab1 | 1803 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1804 | { |
19655d34 CL |
1805 | int node, local_node; |
1806 | enum zone_type i,j; | |
1da177e4 LT |
1807 | |
1808 | local_node = pgdat->node_id; | |
19655d34 | 1809 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1810 | struct zonelist *zonelist; |
1811 | ||
1812 | zonelist = pgdat->node_zonelists + i; | |
1813 | ||
19655d34 | 1814 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
1815 | /* |
1816 | * Now we build the zonelist so that it contains the zones | |
1817 | * of all the other nodes. | |
1818 | * We don't want to pressure a particular node, so when | |
1819 | * building the zones for node N, we make sure that the | |
1820 | * zones coming right after the local ones are those from | |
1821 | * node N+1 (modulo N) | |
1822 | */ | |
1823 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1824 | if (!node_online(node)) | |
1825 | continue; | |
19655d34 | 1826 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1827 | } |
1828 | for (node = 0; node < local_node; node++) { | |
1829 | if (!node_online(node)) | |
1830 | continue; | |
19655d34 | 1831 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1832 | } |
1833 | ||
1834 | zonelist->zones[j] = NULL; | |
1835 | } | |
1836 | } | |
1837 | ||
9276b1bc PJ |
1838 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
1839 | static void __meminit build_zonelist_cache(pg_data_t *pgdat) | |
1840 | { | |
1841 | int i; | |
1842 | ||
1843 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1844 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
1845 | } | |
1846 | ||
1da177e4 LT |
1847 | #endif /* CONFIG_NUMA */ |
1848 | ||
6811378e YG |
1849 | /* return values int ....just for stop_machine_run() */ |
1850 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1851 | { |
6811378e | 1852 | int nid; |
9276b1bc PJ |
1853 | |
1854 | for_each_online_node(nid) { | |
6811378e | 1855 | build_zonelists(NODE_DATA(nid)); |
9276b1bc PJ |
1856 | build_zonelist_cache(NODE_DATA(nid)); |
1857 | } | |
6811378e YG |
1858 | return 0; |
1859 | } | |
1860 | ||
1861 | void __meminit build_all_zonelists(void) | |
1862 | { | |
1863 | if (system_state == SYSTEM_BOOTING) { | |
423b41d7 | 1864 | __build_all_zonelists(NULL); |
6811378e YG |
1865 | cpuset_init_current_mems_allowed(); |
1866 | } else { | |
1867 | /* we have to stop all cpus to guaranntee there is no user | |
1868 | of zonelist */ | |
1869 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1870 | /* cpuset refresh routine should be here */ | |
1871 | } | |
bd1e22b8 AM |
1872 | vm_total_pages = nr_free_pagecache_pages(); |
1873 | printk("Built %i zonelists. Total pages: %ld\n", | |
1874 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1875 | } |
1876 | ||
1877 | /* | |
1878 | * Helper functions to size the waitqueue hash table. | |
1879 | * Essentially these want to choose hash table sizes sufficiently | |
1880 | * large so that collisions trying to wait on pages are rare. | |
1881 | * But in fact, the number of active page waitqueues on typical | |
1882 | * systems is ridiculously low, less than 200. So this is even | |
1883 | * conservative, even though it seems large. | |
1884 | * | |
1885 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1886 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1887 | */ | |
1888 | #define PAGES_PER_WAITQUEUE 256 | |
1889 | ||
cca448fe | 1890 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1891 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1892 | { |
1893 | unsigned long size = 1; | |
1894 | ||
1895 | pages /= PAGES_PER_WAITQUEUE; | |
1896 | ||
1897 | while (size < pages) | |
1898 | size <<= 1; | |
1899 | ||
1900 | /* | |
1901 | * Once we have dozens or even hundreds of threads sleeping | |
1902 | * on IO we've got bigger problems than wait queue collision. | |
1903 | * Limit the size of the wait table to a reasonable size. | |
1904 | */ | |
1905 | size = min(size, 4096UL); | |
1906 | ||
1907 | return max(size, 4UL); | |
1908 | } | |
cca448fe YG |
1909 | #else |
1910 | /* | |
1911 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1912 | * a suitable size for its wait_table. So we use the maximum size now. | |
1913 | * | |
1914 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1915 | * | |
1916 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1917 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1918 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1919 | * | |
1920 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1921 | * or more by the traditional way. (See above). It equals: | |
1922 | * | |
1923 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1924 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1925 | * powerpc (64K page size) : = (32G +16M)byte. | |
1926 | */ | |
1927 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1928 | { | |
1929 | return 4096UL; | |
1930 | } | |
1931 | #endif | |
1da177e4 LT |
1932 | |
1933 | /* | |
1934 | * This is an integer logarithm so that shifts can be used later | |
1935 | * to extract the more random high bits from the multiplicative | |
1936 | * hash function before the remainder is taken. | |
1937 | */ | |
1938 | static inline unsigned long wait_table_bits(unsigned long size) | |
1939 | { | |
1940 | return ffz(~size); | |
1941 | } | |
1942 | ||
1943 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1944 | ||
1da177e4 LT |
1945 | /* |
1946 | * Initially all pages are reserved - free ones are freed | |
1947 | * up by free_all_bootmem() once the early boot process is | |
1948 | * done. Non-atomic initialization, single-pass. | |
1949 | */ | |
c09b4240 | 1950 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 1951 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 1952 | { |
1da177e4 | 1953 | struct page *page; |
29751f69 AW |
1954 | unsigned long end_pfn = start_pfn + size; |
1955 | unsigned long pfn; | |
1da177e4 | 1956 | |
cbe8dd4a | 1957 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
1958 | /* |
1959 | * There can be holes in boot-time mem_map[]s | |
1960 | * handed to this function. They do not | |
1961 | * exist on hotplugged memory. | |
1962 | */ | |
1963 | if (context == MEMMAP_EARLY) { | |
1964 | if (!early_pfn_valid(pfn)) | |
1965 | continue; | |
1966 | if (!early_pfn_in_nid(pfn, nid)) | |
1967 | continue; | |
1968 | } | |
d41dee36 AW |
1969 | page = pfn_to_page(pfn); |
1970 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1971 | init_page_count(page); |
1da177e4 LT |
1972 | reset_page_mapcount(page); |
1973 | SetPageReserved(page); | |
1974 | INIT_LIST_HEAD(&page->lru); | |
1975 | #ifdef WANT_PAGE_VIRTUAL | |
1976 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1977 | if (!is_highmem_idx(zone)) | |
3212c6be | 1978 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1979 | #endif |
1da177e4 LT |
1980 | } |
1981 | } | |
1982 | ||
1983 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1984 | unsigned long size) | |
1985 | { | |
1986 | int order; | |
1987 | for (order = 0; order < MAX_ORDER ; order++) { | |
1988 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1989 | zone->free_area[order].nr_free = 0; | |
1990 | } | |
1991 | } | |
1992 | ||
1993 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
1994 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 1995 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
1996 | #endif |
1997 | ||
6292d9aa | 1998 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1999 | { |
2000 | int batch; | |
2001 | ||
2002 | /* | |
2003 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2004 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2005 | * |
2006 | * OK, so we don't know how big the cache is. So guess. | |
2007 | */ | |
2008 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2009 | if (batch * PAGE_SIZE > 512 * 1024) |
2010 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2011 | batch /= 4; /* We effectively *= 4 below */ |
2012 | if (batch < 1) | |
2013 | batch = 1; | |
2014 | ||
2015 | /* | |
0ceaacc9 NP |
2016 | * Clamp the batch to a 2^n - 1 value. Having a power |
2017 | * of 2 value was found to be more likely to have | |
2018 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2019 | * |
0ceaacc9 NP |
2020 | * For example if 2 tasks are alternately allocating |
2021 | * batches of pages, one task can end up with a lot | |
2022 | * of pages of one half of the possible page colors | |
2023 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2024 | */ |
0ceaacc9 | 2025 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2026 | |
e7c8d5c9 CL |
2027 | return batch; |
2028 | } | |
2029 | ||
2caaad41 CL |
2030 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2031 | { | |
2032 | struct per_cpu_pages *pcp; | |
2033 | ||
1c6fe946 MD |
2034 | memset(p, 0, sizeof(*p)); |
2035 | ||
2caaad41 CL |
2036 | pcp = &p->pcp[0]; /* hot */ |
2037 | pcp->count = 0; | |
2caaad41 CL |
2038 | pcp->high = 6 * batch; |
2039 | pcp->batch = max(1UL, 1 * batch); | |
2040 | INIT_LIST_HEAD(&pcp->list); | |
2041 | ||
2042 | pcp = &p->pcp[1]; /* cold*/ | |
2043 | pcp->count = 0; | |
2caaad41 | 2044 | pcp->high = 2 * batch; |
e46a5e28 | 2045 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
2046 | INIT_LIST_HEAD(&pcp->list); |
2047 | } | |
2048 | ||
8ad4b1fb RS |
2049 | /* |
2050 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2051 | * to the value high for the pageset p. | |
2052 | */ | |
2053 | ||
2054 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2055 | unsigned long high) | |
2056 | { | |
2057 | struct per_cpu_pages *pcp; | |
2058 | ||
2059 | pcp = &p->pcp[0]; /* hot list */ | |
2060 | pcp->high = high; | |
2061 | pcp->batch = max(1UL, high/4); | |
2062 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2063 | pcp->batch = PAGE_SHIFT * 8; | |
2064 | } | |
2065 | ||
2066 | ||
e7c8d5c9 CL |
2067 | #ifdef CONFIG_NUMA |
2068 | /* | |
2caaad41 CL |
2069 | * Boot pageset table. One per cpu which is going to be used for all |
2070 | * zones and all nodes. The parameters will be set in such a way | |
2071 | * that an item put on a list will immediately be handed over to | |
2072 | * the buddy list. This is safe since pageset manipulation is done | |
2073 | * with interrupts disabled. | |
2074 | * | |
2075 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2076 | * |
2077 | * The boot_pagesets must be kept even after bootup is complete for | |
2078 | * unused processors and/or zones. They do play a role for bootstrapping | |
2079 | * hotplugged processors. | |
2080 | * | |
2081 | * zoneinfo_show() and maybe other functions do | |
2082 | * not check if the processor is online before following the pageset pointer. | |
2083 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2084 | */ |
88a2a4ac | 2085 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2086 | |
2087 | /* | |
2088 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2089 | * per cpu pageset array in struct zone. |
2090 | */ | |
6292d9aa | 2091 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2092 | { |
2093 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
2094 | |
2095 | for_each_zone(zone) { | |
e7c8d5c9 | 2096 | |
66a55030 CL |
2097 | if (!populated_zone(zone)) |
2098 | continue; | |
2099 | ||
23316bc8 | 2100 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 2101 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 2102 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2103 | goto bad; |
e7c8d5c9 | 2104 | |
23316bc8 | 2105 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2106 | |
2107 | if (percpu_pagelist_fraction) | |
2108 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2109 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2110 | } |
2111 | ||
2112 | return 0; | |
2113 | bad: | |
2114 | for_each_zone(dzone) { | |
2115 | if (dzone == zone) | |
2116 | break; | |
23316bc8 NP |
2117 | kfree(zone_pcp(dzone, cpu)); |
2118 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2119 | } |
2120 | return -ENOMEM; | |
2121 | } | |
2122 | ||
2123 | static inline void free_zone_pagesets(int cpu) | |
2124 | { | |
e7c8d5c9 CL |
2125 | struct zone *zone; |
2126 | ||
2127 | for_each_zone(zone) { | |
2128 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2129 | ||
f3ef9ead DR |
2130 | /* Free per_cpu_pageset if it is slab allocated */ |
2131 | if (pset != &boot_pageset[cpu]) | |
2132 | kfree(pset); | |
e7c8d5c9 | 2133 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2134 | } |
e7c8d5c9 CL |
2135 | } |
2136 | ||
9c7b216d | 2137 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2138 | unsigned long action, |
2139 | void *hcpu) | |
2140 | { | |
2141 | int cpu = (long)hcpu; | |
2142 | int ret = NOTIFY_OK; | |
2143 | ||
2144 | switch (action) { | |
ce421c79 AW |
2145 | case CPU_UP_PREPARE: |
2146 | if (process_zones(cpu)) | |
2147 | ret = NOTIFY_BAD; | |
2148 | break; | |
2149 | case CPU_UP_CANCELED: | |
2150 | case CPU_DEAD: | |
2151 | free_zone_pagesets(cpu); | |
2152 | break; | |
2153 | default: | |
2154 | break; | |
e7c8d5c9 CL |
2155 | } |
2156 | return ret; | |
2157 | } | |
2158 | ||
74b85f37 | 2159 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2160 | { &pageset_cpuup_callback, NULL, 0 }; |
2161 | ||
78d9955b | 2162 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2163 | { |
2164 | int err; | |
2165 | ||
2166 | /* Initialize per_cpu_pageset for cpu 0. | |
2167 | * A cpuup callback will do this for every cpu | |
2168 | * as it comes online | |
2169 | */ | |
2170 | err = process_zones(smp_processor_id()); | |
2171 | BUG_ON(err); | |
2172 | register_cpu_notifier(&pageset_notifier); | |
2173 | } | |
2174 | ||
2175 | #endif | |
2176 | ||
c09b4240 | 2177 | static __meminit |
cca448fe | 2178 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2179 | { |
2180 | int i; | |
2181 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2182 | size_t alloc_size; |
ed8ece2e DH |
2183 | |
2184 | /* | |
2185 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2186 | * per zone. | |
2187 | */ | |
02b694de YG |
2188 | zone->wait_table_hash_nr_entries = |
2189 | wait_table_hash_nr_entries(zone_size_pages); | |
2190 | zone->wait_table_bits = | |
2191 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2192 | alloc_size = zone->wait_table_hash_nr_entries |
2193 | * sizeof(wait_queue_head_t); | |
2194 | ||
2195 | if (system_state == SYSTEM_BOOTING) { | |
2196 | zone->wait_table = (wait_queue_head_t *) | |
2197 | alloc_bootmem_node(pgdat, alloc_size); | |
2198 | } else { | |
2199 | /* | |
2200 | * This case means that a zone whose size was 0 gets new memory | |
2201 | * via memory hot-add. | |
2202 | * But it may be the case that a new node was hot-added. In | |
2203 | * this case vmalloc() will not be able to use this new node's | |
2204 | * memory - this wait_table must be initialized to use this new | |
2205 | * node itself as well. | |
2206 | * To use this new node's memory, further consideration will be | |
2207 | * necessary. | |
2208 | */ | |
2209 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
2210 | } | |
2211 | if (!zone->wait_table) | |
2212 | return -ENOMEM; | |
ed8ece2e | 2213 | |
02b694de | 2214 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2215 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2216 | |
2217 | return 0; | |
ed8ece2e DH |
2218 | } |
2219 | ||
c09b4240 | 2220 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2221 | { |
2222 | int cpu; | |
2223 | unsigned long batch = zone_batchsize(zone); | |
2224 | ||
2225 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2226 | #ifdef CONFIG_NUMA | |
2227 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2228 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2229 | setup_pageset(&boot_pageset[cpu],0); |
2230 | #else | |
2231 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2232 | #endif | |
2233 | } | |
f5335c0f AB |
2234 | if (zone->present_pages) |
2235 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2236 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2237 | } |
2238 | ||
718127cc YG |
2239 | __meminit int init_currently_empty_zone(struct zone *zone, |
2240 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2241 | unsigned long size, |
2242 | enum memmap_context context) | |
ed8ece2e DH |
2243 | { |
2244 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2245 | int ret; |
2246 | ret = zone_wait_table_init(zone, size); | |
2247 | if (ret) | |
2248 | return ret; | |
ed8ece2e DH |
2249 | pgdat->nr_zones = zone_idx(zone) + 1; |
2250 | ||
ed8ece2e DH |
2251 | zone->zone_start_pfn = zone_start_pfn; |
2252 | ||
2253 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2254 | ||
2255 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2256 | |
2257 | return 0; | |
ed8ece2e DH |
2258 | } |
2259 | ||
c713216d MG |
2260 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2261 | /* | |
2262 | * Basic iterator support. Return the first range of PFNs for a node | |
2263 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2264 | */ | |
2265 | static int __init first_active_region_index_in_nid(int nid) | |
2266 | { | |
2267 | int i; | |
2268 | ||
2269 | for (i = 0; i < nr_nodemap_entries; i++) | |
2270 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2271 | return i; | |
2272 | ||
2273 | return -1; | |
2274 | } | |
2275 | ||
2276 | /* | |
2277 | * Basic iterator support. Return the next active range of PFNs for a node | |
2278 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2279 | */ | |
2280 | static int __init next_active_region_index_in_nid(int index, int nid) | |
2281 | { | |
2282 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2283 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2284 | return index; | |
2285 | ||
2286 | return -1; | |
2287 | } | |
2288 | ||
2289 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2290 | /* | |
2291 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2292 | * Architectures may implement their own version but if add_active_range() | |
2293 | * was used and there are no special requirements, this is a convenient | |
2294 | * alternative | |
2295 | */ | |
2296 | int __init early_pfn_to_nid(unsigned long pfn) | |
2297 | { | |
2298 | int i; | |
2299 | ||
2300 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2301 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2302 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2303 | ||
2304 | if (start_pfn <= pfn && pfn < end_pfn) | |
2305 | return early_node_map[i].nid; | |
2306 | } | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2311 | ||
2312 | /* Basic iterator support to walk early_node_map[] */ | |
2313 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2314 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2315 | i = next_active_region_index_in_nid(i, nid)) | |
2316 | ||
2317 | /** | |
2318 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2319 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2320 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2321 | * |
2322 | * If an architecture guarantees that all ranges registered with | |
2323 | * add_active_ranges() contain no holes and may be freed, this | |
2324 | * this function may be used instead of calling free_bootmem() manually. | |
2325 | */ | |
2326 | void __init free_bootmem_with_active_regions(int nid, | |
2327 | unsigned long max_low_pfn) | |
2328 | { | |
2329 | int i; | |
2330 | ||
2331 | for_each_active_range_index_in_nid(i, nid) { | |
2332 | unsigned long size_pages = 0; | |
2333 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2334 | ||
2335 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2336 | continue; | |
2337 | ||
2338 | if (end_pfn > max_low_pfn) | |
2339 | end_pfn = max_low_pfn; | |
2340 | ||
2341 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2342 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2343 | PFN_PHYS(early_node_map[i].start_pfn), | |
2344 | size_pages << PAGE_SHIFT); | |
2345 | } | |
2346 | } | |
2347 | ||
2348 | /** | |
2349 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2350 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2351 | * |
2352 | * If an architecture guarantees that all ranges registered with | |
2353 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2354 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2355 | */ |
2356 | void __init sparse_memory_present_with_active_regions(int nid) | |
2357 | { | |
2358 | int i; | |
2359 | ||
2360 | for_each_active_range_index_in_nid(i, nid) | |
2361 | memory_present(early_node_map[i].nid, | |
2362 | early_node_map[i].start_pfn, | |
2363 | early_node_map[i].end_pfn); | |
2364 | } | |
2365 | ||
fb01439c MG |
2366 | /** |
2367 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2368 | * @nid: The nid of the node to push the boundary for | |
2369 | * @start_pfn: The start pfn of the node | |
2370 | * @end_pfn: The end pfn of the node | |
2371 | * | |
2372 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2373 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2374 | * be hotplugged even though no physical memory exists. This function allows | |
2375 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2376 | * be used later. | |
2377 | */ | |
2378 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2379 | void __init push_node_boundaries(unsigned int nid, | |
2380 | unsigned long start_pfn, unsigned long end_pfn) | |
2381 | { | |
2382 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2383 | nid, start_pfn, end_pfn); | |
2384 | ||
2385 | /* Initialise the boundary for this node if necessary */ | |
2386 | if (node_boundary_end_pfn[nid] == 0) | |
2387 | node_boundary_start_pfn[nid] = -1UL; | |
2388 | ||
2389 | /* Update the boundaries */ | |
2390 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2391 | node_boundary_start_pfn[nid] = start_pfn; | |
2392 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2393 | node_boundary_end_pfn[nid] = end_pfn; | |
2394 | } | |
2395 | ||
2396 | /* If necessary, push the node boundary out for reserve hotadd */ | |
2397 | static void __init account_node_boundary(unsigned int nid, | |
2398 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2399 | { | |
2400 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2401 | nid, *start_pfn, *end_pfn); | |
2402 | ||
2403 | /* Return if boundary information has not been provided */ | |
2404 | if (node_boundary_end_pfn[nid] == 0) | |
2405 | return; | |
2406 | ||
2407 | /* Check the boundaries and update if necessary */ | |
2408 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2409 | *start_pfn = node_boundary_start_pfn[nid]; | |
2410 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2411 | *end_pfn = node_boundary_end_pfn[nid]; | |
2412 | } | |
2413 | #else | |
2414 | void __init push_node_boundaries(unsigned int nid, | |
2415 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2416 | ||
2417 | static void __init account_node_boundary(unsigned int nid, | |
2418 | unsigned long *start_pfn, unsigned long *end_pfn) {} | |
2419 | #endif | |
2420 | ||
2421 | ||
c713216d MG |
2422 | /** |
2423 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2424 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2425 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2426 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2427 | * |
2428 | * It returns the start and end page frame of a node based on information | |
2429 | * provided by an arch calling add_active_range(). If called for a node | |
2430 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2431 | * PFNs will be 0. |
c713216d MG |
2432 | */ |
2433 | void __init get_pfn_range_for_nid(unsigned int nid, | |
2434 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2435 | { | |
2436 | int i; | |
2437 | *start_pfn = -1UL; | |
2438 | *end_pfn = 0; | |
2439 | ||
2440 | for_each_active_range_index_in_nid(i, nid) { | |
2441 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2442 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2443 | } | |
2444 | ||
2445 | if (*start_pfn == -1UL) { | |
2446 | printk(KERN_WARNING "Node %u active with no memory\n", nid); | |
2447 | *start_pfn = 0; | |
2448 | } | |
fb01439c MG |
2449 | |
2450 | /* Push the node boundaries out if requested */ | |
2451 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
2452 | } |
2453 | ||
2454 | /* | |
2455 | * Return the number of pages a zone spans in a node, including holes | |
2456 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2457 | */ | |
2458 | unsigned long __init zone_spanned_pages_in_node(int nid, | |
2459 | unsigned long zone_type, | |
2460 | unsigned long *ignored) | |
2461 | { | |
2462 | unsigned long node_start_pfn, node_end_pfn; | |
2463 | unsigned long zone_start_pfn, zone_end_pfn; | |
2464 | ||
2465 | /* Get the start and end of the node and zone */ | |
2466 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2467 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2468 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2469 | ||
2470 | /* Check that this node has pages within the zone's required range */ | |
2471 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2472 | return 0; | |
2473 | ||
2474 | /* Move the zone boundaries inside the node if necessary */ | |
2475 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2476 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2477 | ||
2478 | /* Return the spanned pages */ | |
2479 | return zone_end_pfn - zone_start_pfn; | |
2480 | } | |
2481 | ||
2482 | /* | |
2483 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 2484 | * then all holes in the requested range will be accounted for. |
c713216d MG |
2485 | */ |
2486 | unsigned long __init __absent_pages_in_range(int nid, | |
2487 | unsigned long range_start_pfn, | |
2488 | unsigned long range_end_pfn) | |
2489 | { | |
2490 | int i = 0; | |
2491 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2492 | unsigned long start_pfn; | |
2493 | ||
2494 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2495 | i = first_active_region_index_in_nid(nid); | |
2496 | if (i == -1) | |
2497 | return 0; | |
2498 | ||
9c7cd687 MG |
2499 | /* Account for ranges before physical memory on this node */ |
2500 | if (early_node_map[i].start_pfn > range_start_pfn) | |
2501 | hole_pages = early_node_map[i].start_pfn - range_start_pfn; | |
2502 | ||
c713216d MG |
2503 | prev_end_pfn = early_node_map[i].start_pfn; |
2504 | ||
2505 | /* Find all holes for the zone within the node */ | |
2506 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2507 | ||
2508 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2509 | if (prev_end_pfn >= range_end_pfn) | |
2510 | break; | |
2511 | ||
2512 | /* Make sure the end of the zone is not within the hole */ | |
2513 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2514 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2515 | ||
2516 | /* Update the hole size cound and move on */ | |
2517 | if (start_pfn > range_start_pfn) { | |
2518 | BUG_ON(prev_end_pfn > start_pfn); | |
2519 | hole_pages += start_pfn - prev_end_pfn; | |
2520 | } | |
2521 | prev_end_pfn = early_node_map[i].end_pfn; | |
2522 | } | |
2523 | ||
9c7cd687 MG |
2524 | /* Account for ranges past physical memory on this node */ |
2525 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 2526 | hole_pages += range_end_pfn - |
9c7cd687 MG |
2527 | max(range_start_pfn, prev_end_pfn); |
2528 | ||
c713216d MG |
2529 | return hole_pages; |
2530 | } | |
2531 | ||
2532 | /** | |
2533 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2534 | * @start_pfn: The start PFN to start searching for holes | |
2535 | * @end_pfn: The end PFN to stop searching for holes | |
2536 | * | |
88ca3b94 | 2537 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
2538 | */ |
2539 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2540 | unsigned long end_pfn) | |
2541 | { | |
2542 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2543 | } | |
2544 | ||
2545 | /* Return the number of page frames in holes in a zone on a node */ | |
2546 | unsigned long __init zone_absent_pages_in_node(int nid, | |
2547 | unsigned long zone_type, | |
2548 | unsigned long *ignored) | |
2549 | { | |
9c7cd687 MG |
2550 | unsigned long node_start_pfn, node_end_pfn; |
2551 | unsigned long zone_start_pfn, zone_end_pfn; | |
2552 | ||
2553 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2554 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
2555 | node_start_pfn); | |
2556 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
2557 | node_end_pfn); | |
2558 | ||
2559 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); | |
c713216d | 2560 | } |
0e0b864e | 2561 | |
c713216d MG |
2562 | #else |
2563 | static inline unsigned long zone_spanned_pages_in_node(int nid, | |
2564 | unsigned long zone_type, | |
2565 | unsigned long *zones_size) | |
2566 | { | |
2567 | return zones_size[zone_type]; | |
2568 | } | |
2569 | ||
2570 | static inline unsigned long zone_absent_pages_in_node(int nid, | |
2571 | unsigned long zone_type, | |
2572 | unsigned long *zholes_size) | |
2573 | { | |
2574 | if (!zholes_size) | |
2575 | return 0; | |
2576 | ||
2577 | return zholes_size[zone_type]; | |
2578 | } | |
0e0b864e | 2579 | |
c713216d MG |
2580 | #endif |
2581 | ||
2582 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, | |
2583 | unsigned long *zones_size, unsigned long *zholes_size) | |
2584 | { | |
2585 | unsigned long realtotalpages, totalpages = 0; | |
2586 | enum zone_type i; | |
2587 | ||
2588 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2589 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
2590 | zones_size); | |
2591 | pgdat->node_spanned_pages = totalpages; | |
2592 | ||
2593 | realtotalpages = totalpages; | |
2594 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2595 | realtotalpages -= | |
2596 | zone_absent_pages_in_node(pgdat->node_id, i, | |
2597 | zholes_size); | |
2598 | pgdat->node_present_pages = realtotalpages; | |
2599 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
2600 | realtotalpages); | |
2601 | } | |
2602 | ||
1da177e4 LT |
2603 | /* |
2604 | * Set up the zone data structures: | |
2605 | * - mark all pages reserved | |
2606 | * - mark all memory queues empty | |
2607 | * - clear the memory bitmaps | |
2608 | */ | |
86356ab1 | 2609 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
2610 | unsigned long *zones_size, unsigned long *zholes_size) |
2611 | { | |
2f1b6248 | 2612 | enum zone_type j; |
ed8ece2e | 2613 | int nid = pgdat->node_id; |
1da177e4 | 2614 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 2615 | int ret; |
1da177e4 | 2616 | |
208d54e5 | 2617 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2618 | pgdat->nr_zones = 0; |
2619 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2620 | pgdat->kswapd_max_order = 0; | |
2621 | ||
2622 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2623 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 2624 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 2625 | |
c713216d MG |
2626 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
2627 | realsize = size - zone_absent_pages_in_node(nid, j, | |
2628 | zholes_size); | |
1da177e4 | 2629 | |
0e0b864e MG |
2630 | /* |
2631 | * Adjust realsize so that it accounts for how much memory | |
2632 | * is used by this zone for memmap. This affects the watermark | |
2633 | * and per-cpu initialisations | |
2634 | */ | |
2635 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
2636 | if (realsize >= memmap_pages) { | |
2637 | realsize -= memmap_pages; | |
2638 | printk(KERN_DEBUG | |
2639 | " %s zone: %lu pages used for memmap\n", | |
2640 | zone_names[j], memmap_pages); | |
2641 | } else | |
2642 | printk(KERN_WARNING | |
2643 | " %s zone: %lu pages exceeds realsize %lu\n", | |
2644 | zone_names[j], memmap_pages, realsize); | |
2645 | ||
6267276f CL |
2646 | /* Account for reserved pages */ |
2647 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 2648 | realsize -= dma_reserve; |
6267276f CL |
2649 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
2650 | zone_names[0], dma_reserve); | |
0e0b864e MG |
2651 | } |
2652 | ||
98d2b0eb | 2653 | if (!is_highmem_idx(j)) |
1da177e4 LT |
2654 | nr_kernel_pages += realsize; |
2655 | nr_all_pages += realsize; | |
2656 | ||
2657 | zone->spanned_pages = size; | |
2658 | zone->present_pages = realsize; | |
9614634f | 2659 | #ifdef CONFIG_NUMA |
d5f541ed | 2660 | zone->node = nid; |
8417bba4 | 2661 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 2662 | / 100; |
0ff38490 | 2663 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 2664 | #endif |
1da177e4 LT |
2665 | zone->name = zone_names[j]; |
2666 | spin_lock_init(&zone->lock); | |
2667 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2668 | zone_seqlock_init(zone); |
1da177e4 | 2669 | zone->zone_pgdat = pgdat; |
1da177e4 | 2670 | |
3bb1a852 | 2671 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 2672 | |
ed8ece2e | 2673 | zone_pcp_init(zone); |
1da177e4 LT |
2674 | INIT_LIST_HEAD(&zone->active_list); |
2675 | INIT_LIST_HEAD(&zone->inactive_list); | |
2676 | zone->nr_scan_active = 0; | |
2677 | zone->nr_scan_inactive = 0; | |
2244b95a | 2678 | zap_zone_vm_stats(zone); |
53e9a615 | 2679 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2680 | if (!size) |
2681 | continue; | |
2682 | ||
a2f3aa02 DH |
2683 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
2684 | size, MEMMAP_EARLY); | |
718127cc | 2685 | BUG_ON(ret); |
1da177e4 | 2686 | zone_start_pfn += size; |
1da177e4 LT |
2687 | } |
2688 | } | |
2689 | ||
2690 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2691 | { | |
1da177e4 LT |
2692 | /* Skip empty nodes */ |
2693 | if (!pgdat->node_spanned_pages) | |
2694 | return; | |
2695 | ||
d41dee36 | 2696 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2697 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2698 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2699 | unsigned long size, start, end; |
d41dee36 AW |
2700 | struct page *map; |
2701 | ||
e984bb43 BP |
2702 | /* |
2703 | * The zone's endpoints aren't required to be MAX_ORDER | |
2704 | * aligned but the node_mem_map endpoints must be in order | |
2705 | * for the buddy allocator to function correctly. | |
2706 | */ | |
2707 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2708 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2709 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2710 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2711 | map = alloc_remap(pgdat->node_id, size); |
2712 | if (!map) | |
2713 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2714 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2715 | } |
d41dee36 | 2716 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2717 | /* |
2718 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2719 | */ | |
c713216d | 2720 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 2721 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
2722 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2723 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
2724 | mem_map -= pgdat->node_start_pfn; | |
2725 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2726 | } | |
1da177e4 | 2727 | #endif |
d41dee36 | 2728 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2729 | } |
2730 | ||
86356ab1 | 2731 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2732 | unsigned long *zones_size, unsigned long node_start_pfn, |
2733 | unsigned long *zholes_size) | |
2734 | { | |
2735 | pgdat->node_id = nid; | |
2736 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 2737 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
2738 | |
2739 | alloc_node_mem_map(pgdat); | |
2740 | ||
2741 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2742 | } | |
2743 | ||
c713216d MG |
2744 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2745 | /** | |
2746 | * add_active_range - Register a range of PFNs backed by physical memory | |
2747 | * @nid: The node ID the range resides on | |
2748 | * @start_pfn: The start PFN of the available physical memory | |
2749 | * @end_pfn: The end PFN of the available physical memory | |
2750 | * | |
2751 | * These ranges are stored in an early_node_map[] and later used by | |
2752 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
2753 | * range spans a memory hole, it is up to the architecture to ensure | |
2754 | * the memory is not freed by the bootmem allocator. If possible | |
2755 | * the range being registered will be merged with existing ranges. | |
2756 | */ | |
2757 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
2758 | unsigned long end_pfn) | |
2759 | { | |
2760 | int i; | |
2761 | ||
2762 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
2763 | "%d entries of %d used\n", | |
2764 | nid, start_pfn, end_pfn, | |
2765 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
2766 | ||
2767 | /* Merge with existing active regions if possible */ | |
2768 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2769 | if (early_node_map[i].nid != nid) | |
2770 | continue; | |
2771 | ||
2772 | /* Skip if an existing region covers this new one */ | |
2773 | if (start_pfn >= early_node_map[i].start_pfn && | |
2774 | end_pfn <= early_node_map[i].end_pfn) | |
2775 | return; | |
2776 | ||
2777 | /* Merge forward if suitable */ | |
2778 | if (start_pfn <= early_node_map[i].end_pfn && | |
2779 | end_pfn > early_node_map[i].end_pfn) { | |
2780 | early_node_map[i].end_pfn = end_pfn; | |
2781 | return; | |
2782 | } | |
2783 | ||
2784 | /* Merge backward if suitable */ | |
2785 | if (start_pfn < early_node_map[i].end_pfn && | |
2786 | end_pfn >= early_node_map[i].start_pfn) { | |
2787 | early_node_map[i].start_pfn = start_pfn; | |
2788 | return; | |
2789 | } | |
2790 | } | |
2791 | ||
2792 | /* Check that early_node_map is large enough */ | |
2793 | if (i >= MAX_ACTIVE_REGIONS) { | |
2794 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
2795 | MAX_ACTIVE_REGIONS); | |
2796 | return; | |
2797 | } | |
2798 | ||
2799 | early_node_map[i].nid = nid; | |
2800 | early_node_map[i].start_pfn = start_pfn; | |
2801 | early_node_map[i].end_pfn = end_pfn; | |
2802 | nr_nodemap_entries = i + 1; | |
2803 | } | |
2804 | ||
2805 | /** | |
2806 | * shrink_active_range - Shrink an existing registered range of PFNs | |
2807 | * @nid: The node id the range is on that should be shrunk | |
2808 | * @old_end_pfn: The old end PFN of the range | |
2809 | * @new_end_pfn: The new PFN of the range | |
2810 | * | |
2811 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
2812 | * The map is kept at the end physical page range that has already been | |
2813 | * registered with add_active_range(). This function allows an arch to shrink | |
2814 | * an existing registered range. | |
2815 | */ | |
2816 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
2817 | unsigned long new_end_pfn) | |
2818 | { | |
2819 | int i; | |
2820 | ||
2821 | /* Find the old active region end and shrink */ | |
2822 | for_each_active_range_index_in_nid(i, nid) | |
2823 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
2824 | early_node_map[i].end_pfn = new_end_pfn; | |
2825 | break; | |
2826 | } | |
2827 | } | |
2828 | ||
2829 | /** | |
2830 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 2831 | * |
c713216d MG |
2832 | * During discovery, it may be found that a table like SRAT is invalid |
2833 | * and an alternative discovery method must be used. This function removes | |
2834 | * all currently registered regions. | |
2835 | */ | |
88ca3b94 | 2836 | void __init remove_all_active_ranges(void) |
c713216d MG |
2837 | { |
2838 | memset(early_node_map, 0, sizeof(early_node_map)); | |
2839 | nr_nodemap_entries = 0; | |
fb01439c MG |
2840 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
2841 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
2842 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
2843 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
2844 | } |
2845 | ||
2846 | /* Compare two active node_active_regions */ | |
2847 | static int __init cmp_node_active_region(const void *a, const void *b) | |
2848 | { | |
2849 | struct node_active_region *arange = (struct node_active_region *)a; | |
2850 | struct node_active_region *brange = (struct node_active_region *)b; | |
2851 | ||
2852 | /* Done this way to avoid overflows */ | |
2853 | if (arange->start_pfn > brange->start_pfn) | |
2854 | return 1; | |
2855 | if (arange->start_pfn < brange->start_pfn) | |
2856 | return -1; | |
2857 | ||
2858 | return 0; | |
2859 | } | |
2860 | ||
2861 | /* sort the node_map by start_pfn */ | |
2862 | static void __init sort_node_map(void) | |
2863 | { | |
2864 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
2865 | sizeof(struct node_active_region), | |
2866 | cmp_node_active_region, NULL); | |
2867 | } | |
2868 | ||
a6af2bc3 | 2869 | /* Find the lowest pfn for a node */ |
c713216d MG |
2870 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
2871 | { | |
2872 | int i; | |
a6af2bc3 | 2873 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 2874 | |
c713216d MG |
2875 | /* Assuming a sorted map, the first range found has the starting pfn */ |
2876 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 2877 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 2878 | |
a6af2bc3 MG |
2879 | if (min_pfn == ULONG_MAX) { |
2880 | printk(KERN_WARNING | |
2881 | "Could not find start_pfn for node %lu\n", nid); | |
2882 | return 0; | |
2883 | } | |
2884 | ||
2885 | return min_pfn; | |
c713216d MG |
2886 | } |
2887 | ||
2888 | /** | |
2889 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
2890 | * | |
2891 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 2892 | * add_active_range(). |
c713216d MG |
2893 | */ |
2894 | unsigned long __init find_min_pfn_with_active_regions(void) | |
2895 | { | |
2896 | return find_min_pfn_for_node(MAX_NUMNODES); | |
2897 | } | |
2898 | ||
2899 | /** | |
2900 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
2901 | * | |
2902 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 2903 | * add_active_range(). |
c713216d MG |
2904 | */ |
2905 | unsigned long __init find_max_pfn_with_active_regions(void) | |
2906 | { | |
2907 | int i; | |
2908 | unsigned long max_pfn = 0; | |
2909 | ||
2910 | for (i = 0; i < nr_nodemap_entries; i++) | |
2911 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
2912 | ||
2913 | return max_pfn; | |
2914 | } | |
2915 | ||
2916 | /** | |
2917 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 2918 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
2919 | * |
2920 | * This will call free_area_init_node() for each active node in the system. | |
2921 | * Using the page ranges provided by add_active_range(), the size of each | |
2922 | * zone in each node and their holes is calculated. If the maximum PFN | |
2923 | * between two adjacent zones match, it is assumed that the zone is empty. | |
2924 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
2925 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
2926 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
2927 | * at arch_max_dma_pfn. | |
2928 | */ | |
2929 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
2930 | { | |
2931 | unsigned long nid; | |
2932 | enum zone_type i; | |
2933 | ||
a6af2bc3 MG |
2934 | /* Sort early_node_map as initialisation assumes it is sorted */ |
2935 | sort_node_map(); | |
2936 | ||
c713216d MG |
2937 | /* Record where the zone boundaries are */ |
2938 | memset(arch_zone_lowest_possible_pfn, 0, | |
2939 | sizeof(arch_zone_lowest_possible_pfn)); | |
2940 | memset(arch_zone_highest_possible_pfn, 0, | |
2941 | sizeof(arch_zone_highest_possible_pfn)); | |
2942 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
2943 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
2944 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2945 | arch_zone_lowest_possible_pfn[i] = | |
2946 | arch_zone_highest_possible_pfn[i-1]; | |
2947 | arch_zone_highest_possible_pfn[i] = | |
2948 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
2949 | } | |
2950 | ||
c713216d MG |
2951 | /* Print out the zone ranges */ |
2952 | printk("Zone PFN ranges:\n"); | |
2953 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2954 | printk(" %-8s %8lu -> %8lu\n", | |
2955 | zone_names[i], | |
2956 | arch_zone_lowest_possible_pfn[i], | |
2957 | arch_zone_highest_possible_pfn[i]); | |
2958 | ||
2959 | /* Print out the early_node_map[] */ | |
2960 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
2961 | for (i = 0; i < nr_nodemap_entries; i++) | |
2962 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
2963 | early_node_map[i].start_pfn, | |
2964 | early_node_map[i].end_pfn); | |
2965 | ||
2966 | /* Initialise every node */ | |
8ef82866 | 2967 | setup_nr_node_ids(); |
c713216d MG |
2968 | for_each_online_node(nid) { |
2969 | pg_data_t *pgdat = NODE_DATA(nid); | |
2970 | free_area_init_node(nid, pgdat, NULL, | |
2971 | find_min_pfn_for_node(nid), NULL); | |
2972 | } | |
2973 | } | |
2974 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2975 | ||
0e0b864e | 2976 | /** |
88ca3b94 RD |
2977 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
2978 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
2979 | * |
2980 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
2981 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
2982 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
2983 | * function may optionally be used to account for unfreeable pages in the |
2984 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
2985 | * smaller per-cpu batchsize. | |
0e0b864e MG |
2986 | */ |
2987 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
2988 | { | |
2989 | dma_reserve = new_dma_reserve; | |
2990 | } | |
2991 | ||
93b7504e | 2992 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2993 | static bootmem_data_t contig_bootmem_data; |
2994 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2995 | ||
2996 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2997 | #endif |
1da177e4 LT |
2998 | |
2999 | void __init free_area_init(unsigned long *zones_size) | |
3000 | { | |
93b7504e | 3001 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3002 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3003 | } | |
1da177e4 | 3004 | |
1da177e4 LT |
3005 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3006 | unsigned long action, void *hcpu) | |
3007 | { | |
3008 | int cpu = (unsigned long)hcpu; | |
1da177e4 LT |
3009 | |
3010 | if (action == CPU_DEAD) { | |
1da177e4 LT |
3011 | local_irq_disable(); |
3012 | __drain_pages(cpu); | |
f8891e5e | 3013 | vm_events_fold_cpu(cpu); |
1da177e4 | 3014 | local_irq_enable(); |
2244b95a | 3015 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3016 | } |
3017 | return NOTIFY_OK; | |
3018 | } | |
1da177e4 LT |
3019 | |
3020 | void __init page_alloc_init(void) | |
3021 | { | |
3022 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3023 | } | |
3024 | ||
cb45b0e9 HA |
3025 | /* |
3026 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3027 | * or min_free_kbytes changes. | |
3028 | */ | |
3029 | static void calculate_totalreserve_pages(void) | |
3030 | { | |
3031 | struct pglist_data *pgdat; | |
3032 | unsigned long reserve_pages = 0; | |
2f6726e5 | 3033 | enum zone_type i, j; |
cb45b0e9 HA |
3034 | |
3035 | for_each_online_pgdat(pgdat) { | |
3036 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3037 | struct zone *zone = pgdat->node_zones + i; | |
3038 | unsigned long max = 0; | |
3039 | ||
3040 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3041 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3042 | if (zone->lowmem_reserve[j] > max) | |
3043 | max = zone->lowmem_reserve[j]; | |
3044 | } | |
3045 | ||
3046 | /* we treat pages_high as reserved pages. */ | |
3047 | max += zone->pages_high; | |
3048 | ||
3049 | if (max > zone->present_pages) | |
3050 | max = zone->present_pages; | |
3051 | reserve_pages += max; | |
3052 | } | |
3053 | } | |
3054 | totalreserve_pages = reserve_pages; | |
3055 | } | |
3056 | ||
1da177e4 LT |
3057 | /* |
3058 | * setup_per_zone_lowmem_reserve - called whenever | |
3059 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3060 | * has a correct pages reserved value, so an adequate number of | |
3061 | * pages are left in the zone after a successful __alloc_pages(). | |
3062 | */ | |
3063 | static void setup_per_zone_lowmem_reserve(void) | |
3064 | { | |
3065 | struct pglist_data *pgdat; | |
2f6726e5 | 3066 | enum zone_type j, idx; |
1da177e4 | 3067 | |
ec936fc5 | 3068 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
3069 | for (j = 0; j < MAX_NR_ZONES; j++) { |
3070 | struct zone *zone = pgdat->node_zones + j; | |
3071 | unsigned long present_pages = zone->present_pages; | |
3072 | ||
3073 | zone->lowmem_reserve[j] = 0; | |
3074 | ||
2f6726e5 CL |
3075 | idx = j; |
3076 | while (idx) { | |
1da177e4 LT |
3077 | struct zone *lower_zone; |
3078 | ||
2f6726e5 CL |
3079 | idx--; |
3080 | ||
1da177e4 LT |
3081 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
3082 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
3083 | ||
3084 | lower_zone = pgdat->node_zones + idx; | |
3085 | lower_zone->lowmem_reserve[j] = present_pages / | |
3086 | sysctl_lowmem_reserve_ratio[idx]; | |
3087 | present_pages += lower_zone->present_pages; | |
3088 | } | |
3089 | } | |
3090 | } | |
cb45b0e9 HA |
3091 | |
3092 | /* update totalreserve_pages */ | |
3093 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3094 | } |
3095 | ||
88ca3b94 RD |
3096 | /** |
3097 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
3098 | * | |
3099 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
3100 | * with respect to min_free_kbytes. | |
1da177e4 | 3101 | */ |
3947be19 | 3102 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
3103 | { |
3104 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
3105 | unsigned long lowmem_pages = 0; | |
3106 | struct zone *zone; | |
3107 | unsigned long flags; | |
3108 | ||
3109 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
3110 | for_each_zone(zone) { | |
3111 | if (!is_highmem(zone)) | |
3112 | lowmem_pages += zone->present_pages; | |
3113 | } | |
3114 | ||
3115 | for_each_zone(zone) { | |
ac924c60 AM |
3116 | u64 tmp; |
3117 | ||
1da177e4 | 3118 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
3119 | tmp = (u64)pages_min * zone->present_pages; |
3120 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
3121 | if (is_highmem(zone)) { |
3122 | /* | |
669ed175 NP |
3123 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
3124 | * need highmem pages, so cap pages_min to a small | |
3125 | * value here. | |
3126 | * | |
3127 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
3128 | * deltas controls asynch page reclaim, and so should | |
3129 | * not be capped for highmem. | |
1da177e4 LT |
3130 | */ |
3131 | int min_pages; | |
3132 | ||
3133 | min_pages = zone->present_pages / 1024; | |
3134 | if (min_pages < SWAP_CLUSTER_MAX) | |
3135 | min_pages = SWAP_CLUSTER_MAX; | |
3136 | if (min_pages > 128) | |
3137 | min_pages = 128; | |
3138 | zone->pages_min = min_pages; | |
3139 | } else { | |
669ed175 NP |
3140 | /* |
3141 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
3142 | * proportionate to the zone's size. |
3143 | */ | |
669ed175 | 3144 | zone->pages_min = tmp; |
1da177e4 LT |
3145 | } |
3146 | ||
ac924c60 AM |
3147 | zone->pages_low = zone->pages_min + (tmp >> 2); |
3148 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
3149 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
3150 | } | |
cb45b0e9 HA |
3151 | |
3152 | /* update totalreserve_pages */ | |
3153 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3154 | } |
3155 | ||
3156 | /* | |
3157 | * Initialise min_free_kbytes. | |
3158 | * | |
3159 | * For small machines we want it small (128k min). For large machines | |
3160 | * we want it large (64MB max). But it is not linear, because network | |
3161 | * bandwidth does not increase linearly with machine size. We use | |
3162 | * | |
3163 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
3164 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
3165 | * | |
3166 | * which yields | |
3167 | * | |
3168 | * 16MB: 512k | |
3169 | * 32MB: 724k | |
3170 | * 64MB: 1024k | |
3171 | * 128MB: 1448k | |
3172 | * 256MB: 2048k | |
3173 | * 512MB: 2896k | |
3174 | * 1024MB: 4096k | |
3175 | * 2048MB: 5792k | |
3176 | * 4096MB: 8192k | |
3177 | * 8192MB: 11584k | |
3178 | * 16384MB: 16384k | |
3179 | */ | |
3180 | static int __init init_per_zone_pages_min(void) | |
3181 | { | |
3182 | unsigned long lowmem_kbytes; | |
3183 | ||
3184 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
3185 | ||
3186 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
3187 | if (min_free_kbytes < 128) | |
3188 | min_free_kbytes = 128; | |
3189 | if (min_free_kbytes > 65536) | |
3190 | min_free_kbytes = 65536; | |
3191 | setup_per_zone_pages_min(); | |
3192 | setup_per_zone_lowmem_reserve(); | |
3193 | return 0; | |
3194 | } | |
3195 | module_init(init_per_zone_pages_min) | |
3196 | ||
3197 | /* | |
3198 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
3199 | * that we can call two helper functions whenever min_free_kbytes | |
3200 | * changes. | |
3201 | */ | |
3202 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
3203 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3204 | { | |
3205 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3206 | setup_per_zone_pages_min(); | |
3207 | return 0; | |
3208 | } | |
3209 | ||
9614634f CL |
3210 | #ifdef CONFIG_NUMA |
3211 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
3212 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3213 | { | |
3214 | struct zone *zone; | |
3215 | int rc; | |
3216 | ||
3217 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3218 | if (rc) | |
3219 | return rc; | |
3220 | ||
3221 | for_each_zone(zone) | |
8417bba4 | 3222 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
3223 | sysctl_min_unmapped_ratio) / 100; |
3224 | return 0; | |
3225 | } | |
0ff38490 CL |
3226 | |
3227 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
3228 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3229 | { | |
3230 | struct zone *zone; | |
3231 | int rc; | |
3232 | ||
3233 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3234 | if (rc) | |
3235 | return rc; | |
3236 | ||
3237 | for_each_zone(zone) | |
3238 | zone->min_slab_pages = (zone->present_pages * | |
3239 | sysctl_min_slab_ratio) / 100; | |
3240 | return 0; | |
3241 | } | |
9614634f CL |
3242 | #endif |
3243 | ||
1da177e4 LT |
3244 | /* |
3245 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
3246 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
3247 | * whenever sysctl_lowmem_reserve_ratio changes. | |
3248 | * | |
3249 | * The reserve ratio obviously has absolutely no relation with the | |
3250 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
3251 | * if in function of the boot time zone sizes. | |
3252 | */ | |
3253 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
3254 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3255 | { | |
3256 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3257 | setup_per_zone_lowmem_reserve(); | |
3258 | return 0; | |
3259 | } | |
3260 | ||
8ad4b1fb RS |
3261 | /* |
3262 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
3263 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
3264 | * can have before it gets flushed back to buddy allocator. | |
3265 | */ | |
3266 | ||
3267 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
3268 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3269 | { | |
3270 | struct zone *zone; | |
3271 | unsigned int cpu; | |
3272 | int ret; | |
3273 | ||
3274 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3275 | if (!write || (ret == -EINVAL)) | |
3276 | return ret; | |
3277 | for_each_zone(zone) { | |
3278 | for_each_online_cpu(cpu) { | |
3279 | unsigned long high; | |
3280 | high = zone->present_pages / percpu_pagelist_fraction; | |
3281 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
3282 | } | |
3283 | } | |
3284 | return 0; | |
3285 | } | |
3286 | ||
f034b5d4 | 3287 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
3288 | |
3289 | #ifdef CONFIG_NUMA | |
3290 | static int __init set_hashdist(char *str) | |
3291 | { | |
3292 | if (!str) | |
3293 | return 0; | |
3294 | hashdist = simple_strtoul(str, &str, 0); | |
3295 | return 1; | |
3296 | } | |
3297 | __setup("hashdist=", set_hashdist); | |
3298 | #endif | |
3299 | ||
3300 | /* | |
3301 | * allocate a large system hash table from bootmem | |
3302 | * - it is assumed that the hash table must contain an exact power-of-2 | |
3303 | * quantity of entries | |
3304 | * - limit is the number of hash buckets, not the total allocation size | |
3305 | */ | |
3306 | void *__init alloc_large_system_hash(const char *tablename, | |
3307 | unsigned long bucketsize, | |
3308 | unsigned long numentries, | |
3309 | int scale, | |
3310 | int flags, | |
3311 | unsigned int *_hash_shift, | |
3312 | unsigned int *_hash_mask, | |
3313 | unsigned long limit) | |
3314 | { | |
3315 | unsigned long long max = limit; | |
3316 | unsigned long log2qty, size; | |
3317 | void *table = NULL; | |
3318 | ||
3319 | /* allow the kernel cmdline to have a say */ | |
3320 | if (!numentries) { | |
3321 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 3322 | numentries = nr_kernel_pages; |
1da177e4 LT |
3323 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
3324 | numentries >>= 20 - PAGE_SHIFT; | |
3325 | numentries <<= 20 - PAGE_SHIFT; | |
3326 | ||
3327 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
3328 | if (scale > PAGE_SHIFT) | |
3329 | numentries >>= (scale - PAGE_SHIFT); | |
3330 | else | |
3331 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
3332 | |
3333 | /* Make sure we've got at least a 0-order allocation.. */ | |
3334 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
3335 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 3336 | } |
6e692ed3 | 3337 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
3338 | |
3339 | /* limit allocation size to 1/16 total memory by default */ | |
3340 | if (max == 0) { | |
3341 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
3342 | do_div(max, bucketsize); | |
3343 | } | |
3344 | ||
3345 | if (numentries > max) | |
3346 | numentries = max; | |
3347 | ||
f0d1b0b3 | 3348 | log2qty = ilog2(numentries); |
1da177e4 LT |
3349 | |
3350 | do { | |
3351 | size = bucketsize << log2qty; | |
3352 | if (flags & HASH_EARLY) | |
3353 | table = alloc_bootmem(size); | |
3354 | else if (hashdist) | |
3355 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
3356 | else { | |
3357 | unsigned long order; | |
3358 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
3359 | ; | |
3360 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
3361 | } | |
3362 | } while (!table && size > PAGE_SIZE && --log2qty); | |
3363 | ||
3364 | if (!table) | |
3365 | panic("Failed to allocate %s hash table\n", tablename); | |
3366 | ||
3367 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
3368 | tablename, | |
3369 | (1U << log2qty), | |
f0d1b0b3 | 3370 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
3371 | size); |
3372 | ||
3373 | if (_hash_shift) | |
3374 | *_hash_shift = log2qty; | |
3375 | if (_hash_mask) | |
3376 | *_hash_mask = (1 << log2qty) - 1; | |
3377 | ||
3378 | return table; | |
3379 | } | |
a117e66e KH |
3380 | |
3381 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
3382 | struct page *pfn_to_page(unsigned long pfn) |
3383 | { | |
67de6482 | 3384 | return __pfn_to_page(pfn); |
a117e66e KH |
3385 | } |
3386 | unsigned long page_to_pfn(struct page *page) | |
3387 | { | |
67de6482 | 3388 | return __page_to_pfn(page); |
a117e66e | 3389 | } |
a117e66e KH |
3390 | EXPORT_SYMBOL(pfn_to_page); |
3391 | EXPORT_SYMBOL(page_to_pfn); | |
3392 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 3393 | |
6220ec78 | 3394 |