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