]>
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 |
2a1e274a | 83 | 32, |
e53ef38d | 84 | #endif |
2a1e274a | 85 | 32, |
2f1b6248 | 86 | }; |
1da177e4 LT |
87 | |
88 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 89 | |
15ad7cdc | 90 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 91 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 92 | "DMA", |
4b51d669 | 93 | #endif |
fb0e7942 | 94 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 95 | "DMA32", |
fb0e7942 | 96 | #endif |
2f1b6248 | 97 | "Normal", |
e53ef38d | 98 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 99 | "HighMem", |
e53ef38d | 100 | #endif |
2a1e274a | 101 | "Movable", |
2f1b6248 CL |
102 | }; |
103 | ||
1da177e4 LT |
104 | int min_free_kbytes = 1024; |
105 | ||
86356ab1 YG |
106 | unsigned long __meminitdata nr_kernel_pages; |
107 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 108 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 109 | |
c713216d MG |
110 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
111 | /* | |
112 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
113 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
114 | * Ranges passed to add_active_range() will be merged if possible | |
115 | * so the number of times add_active_range() can be called is | |
116 | * related to the number of nodes and the number of holes | |
117 | */ | |
118 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
119 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
120 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
121 | #else | |
122 | #if MAX_NUMNODES >= 32 | |
123 | /* If there can be many nodes, allow up to 50 holes per node */ | |
124 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
125 | #else | |
126 | /* By default, allow up to 256 distinct regions */ | |
127 | #define MAX_ACTIVE_REGIONS 256 | |
128 | #endif | |
129 | #endif | |
130 | ||
98011f56 JB |
131 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
132 | static int __meminitdata nr_nodemap_entries; | |
133 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
134 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c | 135 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
98011f56 JB |
136 | static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES]; |
137 | static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES]; | |
fb01439c | 138 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ |
2a1e274a | 139 | unsigned long __initdata required_kernelcore; |
7e63efef | 140 | unsigned long __initdata required_movablecore; |
2a1e274a MG |
141 | unsigned long __initdata zone_movable_pfn[MAX_NUMNODES]; |
142 | ||
143 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
144 | int movable_zone; | |
145 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
146 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
147 | ||
418508c1 MS |
148 | #if MAX_NUMNODES > 1 |
149 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
150 | EXPORT_SYMBOL(nr_node_ids); | |
151 | #endif | |
152 | ||
13e7444b | 153 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 154 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 155 | { |
bdc8cb98 DH |
156 | int ret = 0; |
157 | unsigned seq; | |
158 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 159 | |
bdc8cb98 DH |
160 | do { |
161 | seq = zone_span_seqbegin(zone); | |
162 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
163 | ret = 1; | |
164 | else if (pfn < zone->zone_start_pfn) | |
165 | ret = 1; | |
166 | } while (zone_span_seqretry(zone, seq)); | |
167 | ||
168 | return ret; | |
c6a57e19 DH |
169 | } |
170 | ||
171 | static int page_is_consistent(struct zone *zone, struct page *page) | |
172 | { | |
14e07298 | 173 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 174 | return 0; |
1da177e4 | 175 | if (zone != page_zone(page)) |
c6a57e19 DH |
176 | return 0; |
177 | ||
178 | return 1; | |
179 | } | |
180 | /* | |
181 | * Temporary debugging check for pages not lying within a given zone. | |
182 | */ | |
183 | static int bad_range(struct zone *zone, struct page *page) | |
184 | { | |
185 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 186 | return 1; |
c6a57e19 DH |
187 | if (!page_is_consistent(zone, page)) |
188 | return 1; | |
189 | ||
1da177e4 LT |
190 | return 0; |
191 | } | |
13e7444b NP |
192 | #else |
193 | static inline int bad_range(struct zone *zone, struct page *page) | |
194 | { | |
195 | return 0; | |
196 | } | |
197 | #endif | |
198 | ||
224abf92 | 199 | static void bad_page(struct page *page) |
1da177e4 | 200 | { |
224abf92 | 201 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
202 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
203 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
204 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
205 | current->comm, page, (int)(2*sizeof(unsigned long)), |
206 | (unsigned long)page->flags, page->mapping, | |
207 | page_mapcount(page), page_count(page)); | |
1da177e4 | 208 | dump_stack(); |
334795ec HD |
209 | page->flags &= ~(1 << PG_lru | |
210 | 1 << PG_private | | |
1da177e4 | 211 | 1 << PG_locked | |
1da177e4 LT |
212 | 1 << PG_active | |
213 | 1 << PG_dirty | | |
334795ec HD |
214 | 1 << PG_reclaim | |
215 | 1 << PG_slab | | |
1da177e4 | 216 | 1 << PG_swapcache | |
676165a8 NP |
217 | 1 << PG_writeback | |
218 | 1 << PG_buddy ); | |
1da177e4 LT |
219 | set_page_count(page, 0); |
220 | reset_page_mapcount(page); | |
221 | page->mapping = NULL; | |
9f158333 | 222 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
223 | } |
224 | ||
1da177e4 LT |
225 | /* |
226 | * Higher-order pages are called "compound pages". They are structured thusly: | |
227 | * | |
228 | * The first PAGE_SIZE page is called the "head page". | |
229 | * | |
230 | * The remaining PAGE_SIZE pages are called "tail pages". | |
231 | * | |
232 | * All pages have PG_compound set. All pages have their ->private pointing at | |
233 | * the head page (even the head page has this). | |
234 | * | |
41d78ba5 HD |
235 | * The first tail page's ->lru.next holds the address of the compound page's |
236 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
237 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 238 | */ |
d98c7a09 HD |
239 | |
240 | static void free_compound_page(struct page *page) | |
241 | { | |
d85f3385 | 242 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
243 | } |
244 | ||
1da177e4 LT |
245 | static void prep_compound_page(struct page *page, unsigned long order) |
246 | { | |
247 | int i; | |
248 | int nr_pages = 1 << order; | |
249 | ||
33f2ef89 | 250 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 251 | set_compound_order(page, order); |
6d777953 | 252 | __SetPageHead(page); |
d85f3385 | 253 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
254 | struct page *p = page + i; |
255 | ||
d85f3385 | 256 | __SetPageTail(p); |
d85f3385 | 257 | p->first_page = page; |
1da177e4 LT |
258 | } |
259 | } | |
260 | ||
261 | static void destroy_compound_page(struct page *page, unsigned long order) | |
262 | { | |
263 | int i; | |
264 | int nr_pages = 1 << order; | |
265 | ||
d85f3385 | 266 | if (unlikely(compound_order(page) != order)) |
224abf92 | 267 | bad_page(page); |
1da177e4 | 268 | |
6d777953 | 269 | if (unlikely(!PageHead(page))) |
d85f3385 | 270 | bad_page(page); |
6d777953 | 271 | __ClearPageHead(page); |
d85f3385 | 272 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
273 | struct page *p = page + i; |
274 | ||
6d777953 | 275 | if (unlikely(!PageTail(p) | |
d85f3385 | 276 | (p->first_page != page))) |
224abf92 | 277 | bad_page(page); |
d85f3385 | 278 | __ClearPageTail(p); |
1da177e4 LT |
279 | } |
280 | } | |
1da177e4 | 281 | |
17cf4406 NP |
282 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
283 | { | |
284 | int i; | |
285 | ||
725d704e | 286 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
287 | /* |
288 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
289 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
290 | */ | |
725d704e | 291 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
292 | for (i = 0; i < (1 << order); i++) |
293 | clear_highpage(page + i); | |
294 | } | |
295 | ||
1da177e4 LT |
296 | /* |
297 | * function for dealing with page's order in buddy system. | |
298 | * zone->lock is already acquired when we use these. | |
299 | * So, we don't need atomic page->flags operations here. | |
300 | */ | |
6aa3001b AM |
301 | static inline unsigned long page_order(struct page *page) |
302 | { | |
4c21e2f2 | 303 | return page_private(page); |
1da177e4 LT |
304 | } |
305 | ||
6aa3001b AM |
306 | static inline void set_page_order(struct page *page, int order) |
307 | { | |
4c21e2f2 | 308 | set_page_private(page, order); |
676165a8 | 309 | __SetPageBuddy(page); |
1da177e4 LT |
310 | } |
311 | ||
312 | static inline void rmv_page_order(struct page *page) | |
313 | { | |
676165a8 | 314 | __ClearPageBuddy(page); |
4c21e2f2 | 315 | set_page_private(page, 0); |
1da177e4 LT |
316 | } |
317 | ||
318 | /* | |
319 | * Locate the struct page for both the matching buddy in our | |
320 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
321 | * | |
322 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
323 | * the following equation: | |
324 | * B2 = B1 ^ (1 << O) | |
325 | * For example, if the starting buddy (buddy2) is #8 its order | |
326 | * 1 buddy is #10: | |
327 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
328 | * | |
329 | * 2) Any buddy B will have an order O+1 parent P which | |
330 | * satisfies the following equation: | |
331 | * P = B & ~(1 << O) | |
332 | * | |
d6e05edc | 333 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
334 | */ |
335 | static inline struct page * | |
336 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
337 | { | |
338 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
339 | ||
340 | return page + (buddy_idx - page_idx); | |
341 | } | |
342 | ||
343 | static inline unsigned long | |
344 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
345 | { | |
346 | return (page_idx & ~(1 << order)); | |
347 | } | |
348 | ||
349 | /* | |
350 | * This function checks whether a page is free && is the buddy | |
351 | * we can do coalesce a page and its buddy if | |
13e7444b | 352 | * (a) the buddy is not in a hole && |
676165a8 | 353 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
354 | * (c) a page and its buddy have the same order && |
355 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
356 | * |
357 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
358 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 359 | * |
676165a8 | 360 | * For recording page's order, we use page_private(page). |
1da177e4 | 361 | */ |
cb2b95e1 AW |
362 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
363 | int order) | |
1da177e4 | 364 | { |
14e07298 | 365 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 366 | return 0; |
13e7444b | 367 | |
cb2b95e1 AW |
368 | if (page_zone_id(page) != page_zone_id(buddy)) |
369 | return 0; | |
370 | ||
371 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
372 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 373 | return 1; |
676165a8 | 374 | } |
6aa3001b | 375 | return 0; |
1da177e4 LT |
376 | } |
377 | ||
378 | /* | |
379 | * Freeing function for a buddy system allocator. | |
380 | * | |
381 | * The concept of a buddy system is to maintain direct-mapped table | |
382 | * (containing bit values) for memory blocks of various "orders". | |
383 | * The bottom level table contains the map for the smallest allocatable | |
384 | * units of memory (here, pages), and each level above it describes | |
385 | * pairs of units from the levels below, hence, "buddies". | |
386 | * At a high level, all that happens here is marking the table entry | |
387 | * at the bottom level available, and propagating the changes upward | |
388 | * as necessary, plus some accounting needed to play nicely with other | |
389 | * parts of the VM system. | |
390 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 391 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 392 | * order is recorded in page_private(page) field. |
1da177e4 LT |
393 | * So when we are allocating or freeing one, we can derive the state of the |
394 | * other. That is, if we allocate a small block, and both were | |
395 | * free, the remainder of the region must be split into blocks. | |
396 | * If a block is freed, and its buddy is also free, then this | |
397 | * triggers coalescing into a block of larger size. | |
398 | * | |
399 | * -- wli | |
400 | */ | |
401 | ||
48db57f8 | 402 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
403 | struct zone *zone, unsigned int order) |
404 | { | |
405 | unsigned long page_idx; | |
406 | int order_size = 1 << order; | |
407 | ||
224abf92 | 408 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
409 | destroy_compound_page(page, order); |
410 | ||
411 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
412 | ||
725d704e NP |
413 | VM_BUG_ON(page_idx & (order_size - 1)); |
414 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 415 | |
d23ad423 | 416 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
417 | while (order < MAX_ORDER-1) { |
418 | unsigned long combined_idx; | |
419 | struct free_area *area; | |
420 | struct page *buddy; | |
421 | ||
1da177e4 | 422 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 423 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 424 | break; /* Move the buddy up one level. */ |
13e7444b | 425 | |
1da177e4 LT |
426 | list_del(&buddy->lru); |
427 | area = zone->free_area + order; | |
428 | area->nr_free--; | |
429 | rmv_page_order(buddy); | |
13e7444b | 430 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
431 | page = page + (combined_idx - page_idx); |
432 | page_idx = combined_idx; | |
433 | order++; | |
434 | } | |
435 | set_page_order(page, order); | |
436 | list_add(&page->lru, &zone->free_area[order].free_list); | |
437 | zone->free_area[order].nr_free++; | |
438 | } | |
439 | ||
224abf92 | 440 | static inline int free_pages_check(struct page *page) |
1da177e4 | 441 | { |
92be2e33 NP |
442 | if (unlikely(page_mapcount(page) | |
443 | (page->mapping != NULL) | | |
444 | (page_count(page) != 0) | | |
1da177e4 LT |
445 | (page->flags & ( |
446 | 1 << PG_lru | | |
447 | 1 << PG_private | | |
448 | 1 << PG_locked | | |
449 | 1 << PG_active | | |
1da177e4 LT |
450 | 1 << PG_slab | |
451 | 1 << PG_swapcache | | |
b5810039 | 452 | 1 << PG_writeback | |
676165a8 NP |
453 | 1 << PG_reserved | |
454 | 1 << PG_buddy )))) | |
224abf92 | 455 | bad_page(page); |
d85f3385 CL |
456 | /* |
457 | * PageReclaim == PageTail. It is only an error | |
458 | * for PageReclaim to be set if PageCompound is clear. | |
459 | */ | |
460 | if (unlikely(!PageCompound(page) && PageReclaim(page))) | |
461 | bad_page(page); | |
1da177e4 | 462 | if (PageDirty(page)) |
242e5468 | 463 | __ClearPageDirty(page); |
689bcebf HD |
464 | /* |
465 | * For now, we report if PG_reserved was found set, but do not | |
466 | * clear it, and do not free the page. But we shall soon need | |
467 | * to do more, for when the ZERO_PAGE count wraps negative. | |
468 | */ | |
469 | return PageReserved(page); | |
1da177e4 LT |
470 | } |
471 | ||
472 | /* | |
473 | * Frees a list of pages. | |
474 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 475 | * count is the number of pages to free. |
1da177e4 LT |
476 | * |
477 | * If the zone was previously in an "all pages pinned" state then look to | |
478 | * see if this freeing clears that state. | |
479 | * | |
480 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
481 | * pinned" detection logic. | |
482 | */ | |
48db57f8 NP |
483 | static void free_pages_bulk(struct zone *zone, int count, |
484 | struct list_head *list, int order) | |
1da177e4 | 485 | { |
c54ad30c | 486 | spin_lock(&zone->lock); |
1da177e4 LT |
487 | zone->all_unreclaimable = 0; |
488 | zone->pages_scanned = 0; | |
48db57f8 NP |
489 | while (count--) { |
490 | struct page *page; | |
491 | ||
725d704e | 492 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 493 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 494 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 495 | list_del(&page->lru); |
48db57f8 | 496 | __free_one_page(page, zone, order); |
1da177e4 | 497 | } |
c54ad30c | 498 | spin_unlock(&zone->lock); |
1da177e4 LT |
499 | } |
500 | ||
48db57f8 | 501 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 502 | { |
006d22d9 CL |
503 | spin_lock(&zone->lock); |
504 | zone->all_unreclaimable = 0; | |
505 | zone->pages_scanned = 0; | |
0798e519 | 506 | __free_one_page(page, zone, order); |
006d22d9 | 507 | spin_unlock(&zone->lock); |
48db57f8 NP |
508 | } |
509 | ||
510 | static void __free_pages_ok(struct page *page, unsigned int order) | |
511 | { | |
512 | unsigned long flags; | |
1da177e4 | 513 | int i; |
689bcebf | 514 | int reserved = 0; |
1da177e4 | 515 | |
1da177e4 | 516 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 517 | reserved += free_pages_check(page + i); |
689bcebf HD |
518 | if (reserved) |
519 | return; | |
520 | ||
9858db50 NP |
521 | if (!PageHighMem(page)) |
522 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 523 | arch_free_page(page, order); |
48db57f8 | 524 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 525 | |
c54ad30c | 526 | local_irq_save(flags); |
f8891e5e | 527 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 528 | free_one_page(page_zone(page), page, order); |
c54ad30c | 529 | local_irq_restore(flags); |
1da177e4 LT |
530 | } |
531 | ||
a226f6c8 DH |
532 | /* |
533 | * permit the bootmem allocator to evade page validation on high-order frees | |
534 | */ | |
535 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
536 | { | |
537 | if (order == 0) { | |
538 | __ClearPageReserved(page); | |
539 | set_page_count(page, 0); | |
7835e98b | 540 | set_page_refcounted(page); |
545b1ea9 | 541 | __free_page(page); |
a226f6c8 | 542 | } else { |
a226f6c8 DH |
543 | int loop; |
544 | ||
545b1ea9 | 545 | prefetchw(page); |
a226f6c8 DH |
546 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
547 | struct page *p = &page[loop]; | |
548 | ||
545b1ea9 NP |
549 | if (loop + 1 < BITS_PER_LONG) |
550 | prefetchw(p + 1); | |
a226f6c8 DH |
551 | __ClearPageReserved(p); |
552 | set_page_count(p, 0); | |
553 | } | |
554 | ||
7835e98b | 555 | set_page_refcounted(page); |
545b1ea9 | 556 | __free_pages(page, order); |
a226f6c8 DH |
557 | } |
558 | } | |
559 | ||
1da177e4 LT |
560 | |
561 | /* | |
562 | * The order of subdivision here is critical for the IO subsystem. | |
563 | * Please do not alter this order without good reasons and regression | |
564 | * testing. Specifically, as large blocks of memory are subdivided, | |
565 | * the order in which smaller blocks are delivered depends on the order | |
566 | * they're subdivided in this function. This is the primary factor | |
567 | * influencing the order in which pages are delivered to the IO | |
568 | * subsystem according to empirical testing, and this is also justified | |
569 | * by considering the behavior of a buddy system containing a single | |
570 | * large block of memory acted on by a series of small allocations. | |
571 | * This behavior is a critical factor in sglist merging's success. | |
572 | * | |
573 | * -- wli | |
574 | */ | |
085cc7d5 | 575 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
576 | int low, int high, struct free_area *area) |
577 | { | |
578 | unsigned long size = 1 << high; | |
579 | ||
580 | while (high > low) { | |
581 | area--; | |
582 | high--; | |
583 | size >>= 1; | |
725d704e | 584 | VM_BUG_ON(bad_range(zone, &page[size])); |
1da177e4 LT |
585 | list_add(&page[size].lru, &area->free_list); |
586 | area->nr_free++; | |
587 | set_page_order(&page[size], high); | |
588 | } | |
1da177e4 LT |
589 | } |
590 | ||
1da177e4 LT |
591 | /* |
592 | * This page is about to be returned from the page allocator | |
593 | */ | |
17cf4406 | 594 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 595 | { |
92be2e33 NP |
596 | if (unlikely(page_mapcount(page) | |
597 | (page->mapping != NULL) | | |
598 | (page_count(page) != 0) | | |
334795ec HD |
599 | (page->flags & ( |
600 | 1 << PG_lru | | |
1da177e4 LT |
601 | 1 << PG_private | |
602 | 1 << PG_locked | | |
1da177e4 LT |
603 | 1 << PG_active | |
604 | 1 << PG_dirty | | |
605 | 1 << PG_reclaim | | |
334795ec | 606 | 1 << PG_slab | |
1da177e4 | 607 | 1 << PG_swapcache | |
b5810039 | 608 | 1 << PG_writeback | |
676165a8 NP |
609 | 1 << PG_reserved | |
610 | 1 << PG_buddy )))) | |
224abf92 | 611 | bad_page(page); |
1da177e4 | 612 | |
689bcebf HD |
613 | /* |
614 | * For now, we report if PG_reserved was found set, but do not | |
615 | * clear it, and do not allocate the page: as a safety net. | |
616 | */ | |
617 | if (PageReserved(page)) | |
618 | return 1; | |
619 | ||
1da177e4 LT |
620 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
621 | 1 << PG_referenced | 1 << PG_arch_1 | | |
5409bae0 | 622 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 623 | set_page_private(page, 0); |
7835e98b | 624 | set_page_refcounted(page); |
cc102509 NP |
625 | |
626 | arch_alloc_page(page, order); | |
1da177e4 | 627 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
628 | |
629 | if (gfp_flags & __GFP_ZERO) | |
630 | prep_zero_page(page, order, gfp_flags); | |
631 | ||
632 | if (order && (gfp_flags & __GFP_COMP)) | |
633 | prep_compound_page(page, order); | |
634 | ||
689bcebf | 635 | return 0; |
1da177e4 LT |
636 | } |
637 | ||
638 | /* | |
639 | * Do the hard work of removing an element from the buddy allocator. | |
640 | * Call me with the zone->lock already held. | |
641 | */ | |
642 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
643 | { | |
644 | struct free_area * area; | |
645 | unsigned int current_order; | |
646 | struct page *page; | |
647 | ||
648 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
649 | area = zone->free_area + current_order; | |
650 | if (list_empty(&area->free_list)) | |
651 | continue; | |
652 | ||
653 | page = list_entry(area->free_list.next, struct page, lru); | |
654 | list_del(&page->lru); | |
655 | rmv_page_order(page); | |
656 | area->nr_free--; | |
d23ad423 | 657 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); |
085cc7d5 NP |
658 | expand(zone, page, order, current_order, area); |
659 | return page; | |
1da177e4 LT |
660 | } |
661 | ||
662 | return NULL; | |
663 | } | |
664 | ||
665 | /* | |
666 | * Obtain a specified number of elements from the buddy allocator, all under | |
667 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
668 | * Returns the number of new pages which were placed at *list. | |
669 | */ | |
670 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
671 | unsigned long count, struct list_head *list) | |
672 | { | |
1da177e4 | 673 | int i; |
1da177e4 | 674 | |
c54ad30c | 675 | spin_lock(&zone->lock); |
1da177e4 | 676 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
677 | struct page *page = __rmqueue(zone, order); |
678 | if (unlikely(page == NULL)) | |
1da177e4 | 679 | break; |
1da177e4 LT |
680 | list_add_tail(&page->lru, list); |
681 | } | |
c54ad30c | 682 | spin_unlock(&zone->lock); |
085cc7d5 | 683 | return i; |
1da177e4 LT |
684 | } |
685 | ||
4ae7c039 | 686 | #ifdef CONFIG_NUMA |
8fce4d8e | 687 | /* |
4037d452 CL |
688 | * Called from the vmstat counter updater to drain pagesets of this |
689 | * currently executing processor on remote nodes after they have | |
690 | * expired. | |
691 | * | |
879336c3 CL |
692 | * Note that this function must be called with the thread pinned to |
693 | * a single processor. | |
8fce4d8e | 694 | */ |
4037d452 | 695 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 696 | { |
4ae7c039 | 697 | unsigned long flags; |
4037d452 | 698 | int to_drain; |
4ae7c039 | 699 | |
4037d452 CL |
700 | local_irq_save(flags); |
701 | if (pcp->count >= pcp->batch) | |
702 | to_drain = pcp->batch; | |
703 | else | |
704 | to_drain = pcp->count; | |
705 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
706 | pcp->count -= to_drain; | |
707 | local_irq_restore(flags); | |
4ae7c039 CL |
708 | } |
709 | #endif | |
710 | ||
1da177e4 LT |
711 | static void __drain_pages(unsigned int cpu) |
712 | { | |
c54ad30c | 713 | unsigned long flags; |
1da177e4 LT |
714 | struct zone *zone; |
715 | int i; | |
716 | ||
717 | for_each_zone(zone) { | |
718 | struct per_cpu_pageset *pset; | |
719 | ||
f2e12bb2 CL |
720 | if (!populated_zone(zone)) |
721 | continue; | |
722 | ||
e7c8d5c9 | 723 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
724 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
725 | struct per_cpu_pages *pcp; | |
726 | ||
727 | pcp = &pset->pcp[i]; | |
c54ad30c | 728 | local_irq_save(flags); |
48db57f8 NP |
729 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
730 | pcp->count = 0; | |
c54ad30c | 731 | local_irq_restore(flags); |
1da177e4 LT |
732 | } |
733 | } | |
734 | } | |
1da177e4 LT |
735 | |
736 | #ifdef CONFIG_PM | |
737 | ||
738 | void mark_free_pages(struct zone *zone) | |
739 | { | |
f623f0db RW |
740 | unsigned long pfn, max_zone_pfn; |
741 | unsigned long flags; | |
1da177e4 LT |
742 | int order; |
743 | struct list_head *curr; | |
744 | ||
745 | if (!zone->spanned_pages) | |
746 | return; | |
747 | ||
748 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
749 | |
750 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
751 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
752 | if (pfn_valid(pfn)) { | |
753 | struct page *page = pfn_to_page(pfn); | |
754 | ||
7be98234 RW |
755 | if (!swsusp_page_is_forbidden(page)) |
756 | swsusp_unset_page_free(page); | |
f623f0db | 757 | } |
1da177e4 LT |
758 | |
759 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
760 | list_for_each(curr, &zone->free_area[order].free_list) { | |
f623f0db | 761 | unsigned long i; |
1da177e4 | 762 | |
f623f0db RW |
763 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
764 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 765 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 766 | } |
1da177e4 | 767 | |
1da177e4 LT |
768 | spin_unlock_irqrestore(&zone->lock, flags); |
769 | } | |
770 | ||
771 | /* | |
772 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
773 | */ | |
774 | void drain_local_pages(void) | |
775 | { | |
776 | unsigned long flags; | |
777 | ||
778 | local_irq_save(flags); | |
779 | __drain_pages(smp_processor_id()); | |
780 | local_irq_restore(flags); | |
781 | } | |
782 | #endif /* CONFIG_PM */ | |
783 | ||
1da177e4 LT |
784 | /* |
785 | * Free a 0-order page | |
786 | */ | |
1da177e4 LT |
787 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
788 | { | |
789 | struct zone *zone = page_zone(page); | |
790 | struct per_cpu_pages *pcp; | |
791 | unsigned long flags; | |
792 | ||
1da177e4 LT |
793 | if (PageAnon(page)) |
794 | page->mapping = NULL; | |
224abf92 | 795 | if (free_pages_check(page)) |
689bcebf HD |
796 | return; |
797 | ||
9858db50 NP |
798 | if (!PageHighMem(page)) |
799 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 800 | arch_free_page(page, 0); |
689bcebf HD |
801 | kernel_map_pages(page, 1, 0); |
802 | ||
e7c8d5c9 | 803 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 804 | local_irq_save(flags); |
f8891e5e | 805 | __count_vm_event(PGFREE); |
1da177e4 LT |
806 | list_add(&page->lru, &pcp->list); |
807 | pcp->count++; | |
48db57f8 NP |
808 | if (pcp->count >= pcp->high) { |
809 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
810 | pcp->count -= pcp->batch; | |
811 | } | |
1da177e4 LT |
812 | local_irq_restore(flags); |
813 | put_cpu(); | |
814 | } | |
815 | ||
816 | void fastcall free_hot_page(struct page *page) | |
817 | { | |
818 | free_hot_cold_page(page, 0); | |
819 | } | |
820 | ||
821 | void fastcall free_cold_page(struct page *page) | |
822 | { | |
823 | free_hot_cold_page(page, 1); | |
824 | } | |
825 | ||
8dfcc9ba NP |
826 | /* |
827 | * split_page takes a non-compound higher-order page, and splits it into | |
828 | * n (1<<order) sub-pages: page[0..n] | |
829 | * Each sub-page must be freed individually. | |
830 | * | |
831 | * Note: this is probably too low level an operation for use in drivers. | |
832 | * Please consult with lkml before using this in your driver. | |
833 | */ | |
834 | void split_page(struct page *page, unsigned int order) | |
835 | { | |
836 | int i; | |
837 | ||
725d704e NP |
838 | VM_BUG_ON(PageCompound(page)); |
839 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
840 | for (i = 1; i < (1 << order); i++) |
841 | set_page_refcounted(page + i); | |
8dfcc9ba | 842 | } |
8dfcc9ba | 843 | |
1da177e4 LT |
844 | /* |
845 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
846 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
847 | * or two. | |
848 | */ | |
a74609fa NP |
849 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
850 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
851 | { |
852 | unsigned long flags; | |
689bcebf | 853 | struct page *page; |
1da177e4 | 854 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 855 | int cpu; |
1da177e4 | 856 | |
689bcebf | 857 | again: |
a74609fa | 858 | cpu = get_cpu(); |
48db57f8 | 859 | if (likely(order == 0)) { |
1da177e4 LT |
860 | struct per_cpu_pages *pcp; |
861 | ||
a74609fa | 862 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 863 | local_irq_save(flags); |
a74609fa | 864 | if (!pcp->count) { |
941c7105 | 865 | pcp->count = rmqueue_bulk(zone, 0, |
1da177e4 | 866 | pcp->batch, &pcp->list); |
a74609fa NP |
867 | if (unlikely(!pcp->count)) |
868 | goto failed; | |
1da177e4 | 869 | } |
a74609fa NP |
870 | page = list_entry(pcp->list.next, struct page, lru); |
871 | list_del(&page->lru); | |
872 | pcp->count--; | |
7fb1d9fc | 873 | } else { |
1da177e4 LT |
874 | spin_lock_irqsave(&zone->lock, flags); |
875 | page = __rmqueue(zone, order); | |
a74609fa NP |
876 | spin_unlock(&zone->lock); |
877 | if (!page) | |
878 | goto failed; | |
1da177e4 LT |
879 | } |
880 | ||
f8891e5e | 881 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 882 | zone_statistics(zonelist, zone); |
a74609fa NP |
883 | local_irq_restore(flags); |
884 | put_cpu(); | |
1da177e4 | 885 | |
725d704e | 886 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 887 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 888 | goto again; |
1da177e4 | 889 | return page; |
a74609fa NP |
890 | |
891 | failed: | |
892 | local_irq_restore(flags); | |
893 | put_cpu(); | |
894 | return NULL; | |
1da177e4 LT |
895 | } |
896 | ||
7fb1d9fc | 897 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
898 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
899 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
900 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
901 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
902 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
903 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 904 | |
933e312e AM |
905 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
906 | ||
907 | static struct fail_page_alloc_attr { | |
908 | struct fault_attr attr; | |
909 | ||
910 | u32 ignore_gfp_highmem; | |
911 | u32 ignore_gfp_wait; | |
54114994 | 912 | u32 min_order; |
933e312e AM |
913 | |
914 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
915 | ||
916 | struct dentry *ignore_gfp_highmem_file; | |
917 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 918 | struct dentry *min_order_file; |
933e312e AM |
919 | |
920 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
921 | ||
922 | } fail_page_alloc = { | |
923 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
924 | .ignore_gfp_wait = 1, |
925 | .ignore_gfp_highmem = 1, | |
54114994 | 926 | .min_order = 1, |
933e312e AM |
927 | }; |
928 | ||
929 | static int __init setup_fail_page_alloc(char *str) | |
930 | { | |
931 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
932 | } | |
933 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
934 | ||
935 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
936 | { | |
54114994 AM |
937 | if (order < fail_page_alloc.min_order) |
938 | return 0; | |
933e312e AM |
939 | if (gfp_mask & __GFP_NOFAIL) |
940 | return 0; | |
941 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
942 | return 0; | |
943 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
944 | return 0; | |
945 | ||
946 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
947 | } | |
948 | ||
949 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
950 | ||
951 | static int __init fail_page_alloc_debugfs(void) | |
952 | { | |
953 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
954 | struct dentry *dir; | |
955 | int err; | |
956 | ||
957 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
958 | "fail_page_alloc"); | |
959 | if (err) | |
960 | return err; | |
961 | dir = fail_page_alloc.attr.dentries.dir; | |
962 | ||
963 | fail_page_alloc.ignore_gfp_wait_file = | |
964 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
965 | &fail_page_alloc.ignore_gfp_wait); | |
966 | ||
967 | fail_page_alloc.ignore_gfp_highmem_file = | |
968 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
969 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
970 | fail_page_alloc.min_order_file = |
971 | debugfs_create_u32("min-order", mode, dir, | |
972 | &fail_page_alloc.min_order); | |
933e312e AM |
973 | |
974 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
975 | !fail_page_alloc.ignore_gfp_highmem_file || |
976 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
977 | err = -ENOMEM; |
978 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
979 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 980 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
981 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
982 | } | |
983 | ||
984 | return err; | |
985 | } | |
986 | ||
987 | late_initcall(fail_page_alloc_debugfs); | |
988 | ||
989 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
990 | ||
991 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
992 | ||
993 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
994 | { | |
995 | return 0; | |
996 | } | |
997 | ||
998 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
999 | ||
1da177e4 LT |
1000 | /* |
1001 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1002 | * of the allocation. | |
1003 | */ | |
1004 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1005 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1006 | { |
1007 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1008 | long min = mark; |
1009 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1010 | int o; |
1011 | ||
7fb1d9fc | 1012 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1013 | min -= min / 2; |
7fb1d9fc | 1014 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1015 | min -= min / 4; |
1016 | ||
1017 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1018 | return 0; | |
1019 | for (o = 0; o < order; o++) { | |
1020 | /* At the next order, this order's pages become unavailable */ | |
1021 | free_pages -= z->free_area[o].nr_free << o; | |
1022 | ||
1023 | /* Require fewer higher order pages to be free */ | |
1024 | min >>= 1; | |
1025 | ||
1026 | if (free_pages <= min) | |
1027 | return 0; | |
1028 | } | |
1029 | return 1; | |
1030 | } | |
1031 | ||
9276b1bc PJ |
1032 | #ifdef CONFIG_NUMA |
1033 | /* | |
1034 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1035 | * skip over zones that are not allowed by the cpuset, or that have | |
1036 | * been recently (in last second) found to be nearly full. See further | |
1037 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1038 | * that have to skip over alot of full or unallowed zones. | |
1039 | * | |
1040 | * If the zonelist cache is present in the passed in zonelist, then | |
1041 | * returns a pointer to the allowed node mask (either the current | |
1042 | * tasks mems_allowed, or node_online_map.) | |
1043 | * | |
1044 | * If the zonelist cache is not available for this zonelist, does | |
1045 | * nothing and returns NULL. | |
1046 | * | |
1047 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1048 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1049 | * | |
1050 | * We hold off even calling zlc_setup, until after we've checked the | |
1051 | * first zone in the zonelist, on the theory that most allocations will | |
1052 | * be satisfied from that first zone, so best to examine that zone as | |
1053 | * quickly as we can. | |
1054 | */ | |
1055 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1056 | { | |
1057 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1058 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1059 | ||
1060 | zlc = zonelist->zlcache_ptr; | |
1061 | if (!zlc) | |
1062 | return NULL; | |
1063 | ||
1064 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1065 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1066 | zlc->last_full_zap = jiffies; | |
1067 | } | |
1068 | ||
1069 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1070 | &cpuset_current_mems_allowed : | |
1071 | &node_online_map; | |
1072 | return allowednodes; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1077 | * if it is worth looking at further for free memory: | |
1078 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1079 | * bit set in the zonelist_cache fullzones BITMAP). | |
1080 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1081 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1082 | * Return true (non-zero) if zone is worth looking at further, or | |
1083 | * else return false (zero) if it is not. | |
1084 | * | |
1085 | * This check -ignores- the distinction between various watermarks, | |
1086 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1087 | * found to be full for any variation of these watermarks, it will | |
1088 | * be considered full for up to one second by all requests, unless | |
1089 | * we are so low on memory on all allowed nodes that we are forced | |
1090 | * into the second scan of the zonelist. | |
1091 | * | |
1092 | * In the second scan we ignore this zonelist cache and exactly | |
1093 | * apply the watermarks to all zones, even it is slower to do so. | |
1094 | * We are low on memory in the second scan, and should leave no stone | |
1095 | * unturned looking for a free page. | |
1096 | */ | |
1097 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1098 | nodemask_t *allowednodes) | |
1099 | { | |
1100 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1101 | int i; /* index of *z in zonelist zones */ | |
1102 | int n; /* node that zone *z is on */ | |
1103 | ||
1104 | zlc = zonelist->zlcache_ptr; | |
1105 | if (!zlc) | |
1106 | return 1; | |
1107 | ||
1108 | i = z - zonelist->zones; | |
1109 | n = zlc->z_to_n[i]; | |
1110 | ||
1111 | /* This zone is worth trying if it is allowed but not full */ | |
1112 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1117 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1118 | * from that zone don't waste time re-examining it. | |
1119 | */ | |
1120 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1121 | { | |
1122 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1123 | int i; /* index of *z in zonelist zones */ | |
1124 | ||
1125 | zlc = zonelist->zlcache_ptr; | |
1126 | if (!zlc) | |
1127 | return; | |
1128 | ||
1129 | i = z - zonelist->zones; | |
1130 | ||
1131 | set_bit(i, zlc->fullzones); | |
1132 | } | |
1133 | ||
1134 | #else /* CONFIG_NUMA */ | |
1135 | ||
1136 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1137 | { | |
1138 | return NULL; | |
1139 | } | |
1140 | ||
1141 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1142 | nodemask_t *allowednodes) | |
1143 | { | |
1144 | return 1; | |
1145 | } | |
1146 | ||
1147 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1148 | { | |
1149 | } | |
1150 | #endif /* CONFIG_NUMA */ | |
1151 | ||
7fb1d9fc | 1152 | /* |
0798e519 | 1153 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1154 | * a page. |
1155 | */ | |
1156 | static struct page * | |
1157 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1158 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1159 | { |
9276b1bc | 1160 | struct zone **z; |
7fb1d9fc | 1161 | struct page *page = NULL; |
9276b1bc | 1162 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1163 | struct zone *zone; |
9276b1bc PJ |
1164 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1165 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1166 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
7fb1d9fc | 1167 | |
9276b1bc | 1168 | zonelist_scan: |
7fb1d9fc | 1169 | /* |
9276b1bc | 1170 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1171 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1172 | */ | |
9276b1bc PJ |
1173 | z = zonelist->zones; |
1174 | ||
7fb1d9fc | 1175 | do { |
9276b1bc PJ |
1176 | if (NUMA_BUILD && zlc_active && |
1177 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1178 | continue; | |
1192d526 | 1179 | zone = *z; |
08e0f6a9 | 1180 | if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) && |
1192d526 | 1181 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) |
9b819d20 | 1182 | break; |
7fb1d9fc | 1183 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1184 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1185 | goto try_next_zone; |
7fb1d9fc RS |
1186 | |
1187 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1188 | unsigned long mark; |
1189 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1190 | mark = zone->pages_min; |
3148890b | 1191 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1192 | mark = zone->pages_low; |
3148890b | 1193 | else |
1192d526 | 1194 | mark = zone->pages_high; |
0798e519 PJ |
1195 | if (!zone_watermark_ok(zone, order, mark, |
1196 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1197 | if (!zone_reclaim_mode || |
1192d526 | 1198 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1199 | goto this_zone_full; |
0798e519 | 1200 | } |
7fb1d9fc RS |
1201 | } |
1202 | ||
1192d526 | 1203 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1204 | if (page) |
7fb1d9fc | 1205 | break; |
9276b1bc PJ |
1206 | this_zone_full: |
1207 | if (NUMA_BUILD) | |
1208 | zlc_mark_zone_full(zonelist, z); | |
1209 | try_next_zone: | |
1210 | if (NUMA_BUILD && !did_zlc_setup) { | |
1211 | /* we do zlc_setup after the first zone is tried */ | |
1212 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1213 | zlc_active = 1; | |
1214 | did_zlc_setup = 1; | |
1215 | } | |
7fb1d9fc | 1216 | } while (*(++z) != NULL); |
9276b1bc PJ |
1217 | |
1218 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1219 | /* Disable zlc cache for second zonelist scan */ | |
1220 | zlc_active = 0; | |
1221 | goto zonelist_scan; | |
1222 | } | |
7fb1d9fc | 1223 | return page; |
753ee728 MH |
1224 | } |
1225 | ||
1da177e4 LT |
1226 | /* |
1227 | * This is the 'heart' of the zoned buddy allocator. | |
1228 | */ | |
1229 | struct page * fastcall | |
dd0fc66f | 1230 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1231 | struct zonelist *zonelist) |
1232 | { | |
260b2367 | 1233 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1234 | struct zone **z; |
1da177e4 LT |
1235 | struct page *page; |
1236 | struct reclaim_state reclaim_state; | |
1237 | struct task_struct *p = current; | |
1da177e4 | 1238 | int do_retry; |
7fb1d9fc | 1239 | int alloc_flags; |
1da177e4 LT |
1240 | int did_some_progress; |
1241 | ||
1242 | might_sleep_if(wait); | |
1243 | ||
933e312e AM |
1244 | if (should_fail_alloc_page(gfp_mask, order)) |
1245 | return NULL; | |
1246 | ||
6b1de916 | 1247 | restart: |
7fb1d9fc | 1248 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1249 | |
7fb1d9fc | 1250 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
1251 | /* Should this ever happen?? */ |
1252 | return NULL; | |
1253 | } | |
6b1de916 | 1254 | |
7fb1d9fc | 1255 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1256 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1257 | if (page) |
1258 | goto got_pg; | |
1da177e4 | 1259 | |
952f3b51 CL |
1260 | /* |
1261 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1262 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1263 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1264 | * using a larger set of nodes after it has established that the | |
1265 | * allowed per node queues are empty and that nodes are | |
1266 | * over allocated. | |
1267 | */ | |
1268 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1269 | goto nopage; | |
1270 | ||
0798e519 | 1271 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1272 | wakeup_kswapd(*z, order); |
1da177e4 | 1273 | |
9bf2229f | 1274 | /* |
7fb1d9fc RS |
1275 | * OK, we're below the kswapd watermark and have kicked background |
1276 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1277 | * to how we want to proceed. | |
1278 | * | |
1279 | * The caller may dip into page reserves a bit more if the caller | |
1280 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1281 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1282 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1283 | */ |
3148890b | 1284 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1285 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1286 | alloc_flags |= ALLOC_HARDER; | |
1287 | if (gfp_mask & __GFP_HIGH) | |
1288 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1289 | if (wait) |
1290 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1291 | |
1292 | /* | |
1293 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1294 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1295 | * |
1296 | * This is the last chance, in general, before the goto nopage. | |
1297 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1298 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1299 | */ |
7fb1d9fc RS |
1300 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1301 | if (page) | |
1302 | goto got_pg; | |
1da177e4 LT |
1303 | |
1304 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1305 | |
b43a57bb | 1306 | rebalance: |
b84a35be NP |
1307 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1308 | && !in_interrupt()) { | |
1309 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1310 | nofail_alloc: |
b84a35be | 1311 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1312 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1313 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1314 | if (page) |
1315 | goto got_pg; | |
885036d3 | 1316 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1317 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1318 | goto nofail_alloc; |
1319 | } | |
1da177e4 LT |
1320 | } |
1321 | goto nopage; | |
1322 | } | |
1323 | ||
1324 | /* Atomic allocations - we can't balance anything */ | |
1325 | if (!wait) | |
1326 | goto nopage; | |
1327 | ||
1da177e4 LT |
1328 | cond_resched(); |
1329 | ||
1330 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1331 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1332 | p->flags |= PF_MEMALLOC; |
1333 | reclaim_state.reclaimed_slab = 0; | |
1334 | p->reclaim_state = &reclaim_state; | |
1335 | ||
7fb1d9fc | 1336 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1337 | |
1338 | p->reclaim_state = NULL; | |
1339 | p->flags &= ~PF_MEMALLOC; | |
1340 | ||
1341 | cond_resched(); | |
1342 | ||
1343 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1344 | page = get_page_from_freelist(gfp_mask, order, |
1345 | zonelist, alloc_flags); | |
1346 | if (page) | |
1347 | goto got_pg; | |
1da177e4 LT |
1348 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1349 | /* | |
1350 | * Go through the zonelist yet one more time, keep | |
1351 | * very high watermark here, this is only to catch | |
1352 | * a parallel oom killing, we must fail if we're still | |
1353 | * under heavy pressure. | |
1354 | */ | |
7fb1d9fc | 1355 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1356 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1357 | if (page) |
1358 | goto got_pg; | |
1da177e4 | 1359 | |
9b0f8b04 | 1360 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1361 | goto restart; |
1362 | } | |
1363 | ||
1364 | /* | |
1365 | * Don't let big-order allocations loop unless the caller explicitly | |
1366 | * requests that. Wait for some write requests to complete then retry. | |
1367 | * | |
1368 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1369 | * <= 3, but that may not be true in other implementations. | |
1370 | */ | |
1371 | do_retry = 0; | |
1372 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1373 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1374 | do_retry = 1; | |
1375 | if (gfp_mask & __GFP_NOFAIL) | |
1376 | do_retry = 1; | |
1377 | } | |
1378 | if (do_retry) { | |
3fcfab16 | 1379 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1380 | goto rebalance; |
1381 | } | |
1382 | ||
1383 | nopage: | |
1384 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1385 | printk(KERN_WARNING "%s: page allocation failure." | |
1386 | " order:%d, mode:0x%x\n", | |
1387 | p->comm, order, gfp_mask); | |
1388 | dump_stack(); | |
578c2fd6 | 1389 | show_mem(); |
1da177e4 | 1390 | } |
1da177e4 | 1391 | got_pg: |
1da177e4 LT |
1392 | return page; |
1393 | } | |
1394 | ||
1395 | EXPORT_SYMBOL(__alloc_pages); | |
1396 | ||
1397 | /* | |
1398 | * Common helper functions. | |
1399 | */ | |
dd0fc66f | 1400 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1401 | { |
1402 | struct page * page; | |
1403 | page = alloc_pages(gfp_mask, order); | |
1404 | if (!page) | |
1405 | return 0; | |
1406 | return (unsigned long) page_address(page); | |
1407 | } | |
1408 | ||
1409 | EXPORT_SYMBOL(__get_free_pages); | |
1410 | ||
dd0fc66f | 1411 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1412 | { |
1413 | struct page * page; | |
1414 | ||
1415 | /* | |
1416 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1417 | * a highmem page | |
1418 | */ | |
725d704e | 1419 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1420 | |
1421 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1422 | if (page) | |
1423 | return (unsigned long) page_address(page); | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | EXPORT_SYMBOL(get_zeroed_page); | |
1428 | ||
1429 | void __pagevec_free(struct pagevec *pvec) | |
1430 | { | |
1431 | int i = pagevec_count(pvec); | |
1432 | ||
1433 | while (--i >= 0) | |
1434 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1435 | } | |
1436 | ||
1437 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1438 | { | |
b5810039 | 1439 | if (put_page_testzero(page)) { |
1da177e4 LT |
1440 | if (order == 0) |
1441 | free_hot_page(page); | |
1442 | else | |
1443 | __free_pages_ok(page, order); | |
1444 | } | |
1445 | } | |
1446 | ||
1447 | EXPORT_SYMBOL(__free_pages); | |
1448 | ||
1449 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1450 | { | |
1451 | if (addr != 0) { | |
725d704e | 1452 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1453 | __free_pages(virt_to_page((void *)addr), order); |
1454 | } | |
1455 | } | |
1456 | ||
1457 | EXPORT_SYMBOL(free_pages); | |
1458 | ||
1da177e4 LT |
1459 | static unsigned int nr_free_zone_pages(int offset) |
1460 | { | |
e310fd43 MB |
1461 | /* Just pick one node, since fallback list is circular */ |
1462 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1463 | unsigned int sum = 0; |
1464 | ||
e310fd43 MB |
1465 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1466 | struct zone **zonep = zonelist->zones; | |
1467 | struct zone *zone; | |
1da177e4 | 1468 | |
e310fd43 MB |
1469 | for (zone = *zonep++; zone; zone = *zonep++) { |
1470 | unsigned long size = zone->present_pages; | |
1471 | unsigned long high = zone->pages_high; | |
1472 | if (size > high) | |
1473 | sum += size - high; | |
1da177e4 LT |
1474 | } |
1475 | ||
1476 | return sum; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1481 | */ | |
1482 | unsigned int nr_free_buffer_pages(void) | |
1483 | { | |
af4ca457 | 1484 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1485 | } |
1486 | ||
1487 | /* | |
1488 | * Amount of free RAM allocatable within all zones | |
1489 | */ | |
1490 | unsigned int nr_free_pagecache_pages(void) | |
1491 | { | |
2a1e274a | 1492 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1493 | } |
08e0f6a9 CL |
1494 | |
1495 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1496 | { |
08e0f6a9 | 1497 | if (NUMA_BUILD) |
25ba77c1 | 1498 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1499 | } |
1da177e4 | 1500 | |
1da177e4 LT |
1501 | void si_meminfo(struct sysinfo *val) |
1502 | { | |
1503 | val->totalram = totalram_pages; | |
1504 | val->sharedram = 0; | |
d23ad423 | 1505 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1506 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1507 | val->totalhigh = totalhigh_pages; |
1508 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1509 | val->mem_unit = PAGE_SIZE; |
1510 | } | |
1511 | ||
1512 | EXPORT_SYMBOL(si_meminfo); | |
1513 | ||
1514 | #ifdef CONFIG_NUMA | |
1515 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1516 | { | |
1517 | pg_data_t *pgdat = NODE_DATA(nid); | |
1518 | ||
1519 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1520 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1521 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1522 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1523 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1524 | NR_FREE_PAGES); | |
98d2b0eb CL |
1525 | #else |
1526 | val->totalhigh = 0; | |
1527 | val->freehigh = 0; | |
1528 | #endif | |
1da177e4 LT |
1529 | val->mem_unit = PAGE_SIZE; |
1530 | } | |
1531 | #endif | |
1532 | ||
1533 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1534 | ||
1535 | /* | |
1536 | * Show free area list (used inside shift_scroll-lock stuff) | |
1537 | * We also calculate the percentage fragmentation. We do this by counting the | |
1538 | * memory on each free list with the exception of the first item on the list. | |
1539 | */ | |
1540 | void show_free_areas(void) | |
1541 | { | |
c7241913 | 1542 | int cpu; |
1da177e4 LT |
1543 | struct zone *zone; |
1544 | ||
1545 | for_each_zone(zone) { | |
c7241913 | 1546 | if (!populated_zone(zone)) |
1da177e4 | 1547 | continue; |
c7241913 JS |
1548 | |
1549 | show_node(zone); | |
1550 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1551 | |
6b482c67 | 1552 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1553 | struct per_cpu_pageset *pageset; |
1554 | ||
e7c8d5c9 | 1555 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1556 | |
c7241913 JS |
1557 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " |
1558 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1559 | cpu, pageset->pcp[0].high, | |
1560 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1561 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1562 | pageset->pcp[1].count); | |
1da177e4 LT |
1563 | } |
1564 | } | |
1565 | ||
a25700a5 | 1566 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1567 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1568 | global_page_state(NR_ACTIVE), |
1569 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1570 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1571 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1572 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1573 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1574 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1575 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1576 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1577 | global_page_state(NR_PAGETABLE), |
1578 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1579 | |
1580 | for_each_zone(zone) { | |
1581 | int i; | |
1582 | ||
c7241913 JS |
1583 | if (!populated_zone(zone)) |
1584 | continue; | |
1585 | ||
1da177e4 LT |
1586 | show_node(zone); |
1587 | printk("%s" | |
1588 | " free:%lukB" | |
1589 | " min:%lukB" | |
1590 | " low:%lukB" | |
1591 | " high:%lukB" | |
1592 | " active:%lukB" | |
1593 | " inactive:%lukB" | |
1594 | " present:%lukB" | |
1595 | " pages_scanned:%lu" | |
1596 | " all_unreclaimable? %s" | |
1597 | "\n", | |
1598 | zone->name, | |
d23ad423 | 1599 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1600 | K(zone->pages_min), |
1601 | K(zone->pages_low), | |
1602 | K(zone->pages_high), | |
c8785385 CL |
1603 | K(zone_page_state(zone, NR_ACTIVE)), |
1604 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1605 | K(zone->present_pages), |
1606 | zone->pages_scanned, | |
1607 | (zone->all_unreclaimable ? "yes" : "no") | |
1608 | ); | |
1609 | printk("lowmem_reserve[]:"); | |
1610 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1611 | printk(" %lu", zone->lowmem_reserve[i]); | |
1612 | printk("\n"); | |
1613 | } | |
1614 | ||
1615 | for_each_zone(zone) { | |
8f9de51a | 1616 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1617 | |
c7241913 JS |
1618 | if (!populated_zone(zone)) |
1619 | continue; | |
1620 | ||
1da177e4 LT |
1621 | show_node(zone); |
1622 | printk("%s: ", zone->name); | |
1da177e4 LT |
1623 | |
1624 | spin_lock_irqsave(&zone->lock, flags); | |
1625 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1626 | nr[order] = zone->free_area[order].nr_free; |
1627 | total += nr[order] << order; | |
1da177e4 LT |
1628 | } |
1629 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1630 | for (order = 0; order < MAX_ORDER; order++) |
1631 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1632 | printk("= %lukB\n", K(total)); |
1633 | } | |
1634 | ||
1635 | show_swap_cache_info(); | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1640 | * |
1641 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1642 | */ |
f0c0b2b8 KH |
1643 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1644 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1645 | { |
1a93205b CL |
1646 | struct zone *zone; |
1647 | ||
98d2b0eb | 1648 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1649 | zone_type++; |
02a68a5e CL |
1650 | |
1651 | do { | |
2f6726e5 | 1652 | zone_type--; |
070f8032 | 1653 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1654 | if (populated_zone(zone)) { |
070f8032 CL |
1655 | zonelist->zones[nr_zones++] = zone; |
1656 | check_highest_zone(zone_type); | |
1da177e4 | 1657 | } |
02a68a5e | 1658 | |
2f6726e5 | 1659 | } while (zone_type); |
070f8032 | 1660 | return nr_zones; |
1da177e4 LT |
1661 | } |
1662 | ||
f0c0b2b8 KH |
1663 | |
1664 | /* | |
1665 | * zonelist_order: | |
1666 | * 0 = automatic detection of better ordering. | |
1667 | * 1 = order by ([node] distance, -zonetype) | |
1668 | * 2 = order by (-zonetype, [node] distance) | |
1669 | * | |
1670 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1671 | * the same zonelist. So only NUMA can configure this param. | |
1672 | */ | |
1673 | #define ZONELIST_ORDER_DEFAULT 0 | |
1674 | #define ZONELIST_ORDER_NODE 1 | |
1675 | #define ZONELIST_ORDER_ZONE 2 | |
1676 | ||
1677 | /* zonelist order in the kernel. | |
1678 | * set_zonelist_order() will set this to NODE or ZONE. | |
1679 | */ | |
1680 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1681 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1682 | ||
1683 | ||
1da177e4 | 1684 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1685 | /* The value user specified ....changed by config */ |
1686 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1687 | /* string for sysctl */ | |
1688 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1689 | char numa_zonelist_order[16] = "default"; | |
1690 | ||
1691 | /* | |
1692 | * interface for configure zonelist ordering. | |
1693 | * command line option "numa_zonelist_order" | |
1694 | * = "[dD]efault - default, automatic configuration. | |
1695 | * = "[nN]ode - order by node locality, then by zone within node | |
1696 | * = "[zZ]one - order by zone, then by locality within zone | |
1697 | */ | |
1698 | ||
1699 | static int __parse_numa_zonelist_order(char *s) | |
1700 | { | |
1701 | if (*s == 'd' || *s == 'D') { | |
1702 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1703 | } else if (*s == 'n' || *s == 'N') { | |
1704 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
1705 | } else if (*s == 'z' || *s == 'Z') { | |
1706 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
1707 | } else { | |
1708 | printk(KERN_WARNING | |
1709 | "Ignoring invalid numa_zonelist_order value: " | |
1710 | "%s\n", s); | |
1711 | return -EINVAL; | |
1712 | } | |
1713 | return 0; | |
1714 | } | |
1715 | ||
1716 | static __init int setup_numa_zonelist_order(char *s) | |
1717 | { | |
1718 | if (s) | |
1719 | return __parse_numa_zonelist_order(s); | |
1720 | return 0; | |
1721 | } | |
1722 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
1723 | ||
1724 | /* | |
1725 | * sysctl handler for numa_zonelist_order | |
1726 | */ | |
1727 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
1728 | struct file *file, void __user *buffer, size_t *length, | |
1729 | loff_t *ppos) | |
1730 | { | |
1731 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
1732 | int ret; | |
1733 | ||
1734 | if (write) | |
1735 | strncpy(saved_string, (char*)table->data, | |
1736 | NUMA_ZONELIST_ORDER_LEN); | |
1737 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
1738 | if (ret) | |
1739 | return ret; | |
1740 | if (write) { | |
1741 | int oldval = user_zonelist_order; | |
1742 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
1743 | /* | |
1744 | * bogus value. restore saved string | |
1745 | */ | |
1746 | strncpy((char*)table->data, saved_string, | |
1747 | NUMA_ZONELIST_ORDER_LEN); | |
1748 | user_zonelist_order = oldval; | |
1749 | } else if (oldval != user_zonelist_order) | |
1750 | build_all_zonelists(); | |
1751 | } | |
1752 | return 0; | |
1753 | } | |
1754 | ||
1755 | ||
1da177e4 | 1756 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
1757 | static int node_load[MAX_NUMNODES]; |
1758 | ||
1da177e4 | 1759 | /** |
4dc3b16b | 1760 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1761 | * @node: node whose fallback list we're appending |
1762 | * @used_node_mask: nodemask_t of already used nodes | |
1763 | * | |
1764 | * We use a number of factors to determine which is the next node that should | |
1765 | * appear on a given node's fallback list. The node should not have appeared | |
1766 | * already in @node's fallback list, and it should be the next closest node | |
1767 | * according to the distance array (which contains arbitrary distance values | |
1768 | * from each node to each node in the system), and should also prefer nodes | |
1769 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1770 | * on them otherwise. | |
1771 | * It returns -1 if no node is found. | |
1772 | */ | |
f0c0b2b8 | 1773 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1774 | { |
4cf808eb | 1775 | int n, val; |
1da177e4 LT |
1776 | int min_val = INT_MAX; |
1777 | int best_node = -1; | |
1778 | ||
4cf808eb LT |
1779 | /* Use the local node if we haven't already */ |
1780 | if (!node_isset(node, *used_node_mask)) { | |
1781 | node_set(node, *used_node_mask); | |
1782 | return node; | |
1783 | } | |
1da177e4 | 1784 | |
4cf808eb LT |
1785 | for_each_online_node(n) { |
1786 | cpumask_t tmp; | |
1da177e4 LT |
1787 | |
1788 | /* Don't want a node to appear more than once */ | |
1789 | if (node_isset(n, *used_node_mask)) | |
1790 | continue; | |
1791 | ||
1da177e4 LT |
1792 | /* Use the distance array to find the distance */ |
1793 | val = node_distance(node, n); | |
1794 | ||
4cf808eb LT |
1795 | /* Penalize nodes under us ("prefer the next node") */ |
1796 | val += (n < node); | |
1797 | ||
1da177e4 LT |
1798 | /* Give preference to headless and unused nodes */ |
1799 | tmp = node_to_cpumask(n); | |
1800 | if (!cpus_empty(tmp)) | |
1801 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1802 | ||
1803 | /* Slight preference for less loaded node */ | |
1804 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1805 | val += node_load[n]; | |
1806 | ||
1807 | if (val < min_val) { | |
1808 | min_val = val; | |
1809 | best_node = n; | |
1810 | } | |
1811 | } | |
1812 | ||
1813 | if (best_node >= 0) | |
1814 | node_set(best_node, *used_node_mask); | |
1815 | ||
1816 | return best_node; | |
1817 | } | |
1818 | ||
f0c0b2b8 KH |
1819 | |
1820 | /* | |
1821 | * Build zonelists ordered by node and zones within node. | |
1822 | * This results in maximum locality--normal zone overflows into local | |
1823 | * DMA zone, if any--but risks exhausting DMA zone. | |
1824 | */ | |
1825 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 1826 | { |
19655d34 | 1827 | enum zone_type i; |
f0c0b2b8 | 1828 | int j; |
1da177e4 | 1829 | struct zonelist *zonelist; |
f0c0b2b8 KH |
1830 | |
1831 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1832 | zonelist = pgdat->node_zonelists + i; | |
1833 | for (j = 0; zonelist->zones[j] != NULL; j++) | |
1834 | ; | |
1835 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
1836 | zonelist->zones[j] = NULL; | |
1837 | } | |
1838 | } | |
1839 | ||
1840 | /* | |
1841 | * Build zonelists ordered by zone and nodes within zones. | |
1842 | * This results in conserving DMA zone[s] until all Normal memory is | |
1843 | * exhausted, but results in overflowing to remote node while memory | |
1844 | * may still exist in local DMA zone. | |
1845 | */ | |
1846 | static int node_order[MAX_NUMNODES]; | |
1847 | ||
1848 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
1849 | { | |
1850 | enum zone_type i; | |
1851 | int pos, j, node; | |
1852 | int zone_type; /* needs to be signed */ | |
1853 | struct zone *z; | |
1854 | struct zonelist *zonelist; | |
1855 | ||
1856 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1857 | zonelist = pgdat->node_zonelists + i; | |
1858 | pos = 0; | |
1859 | for (zone_type = i; zone_type >= 0; zone_type--) { | |
1860 | for (j = 0; j < nr_nodes; j++) { | |
1861 | node = node_order[j]; | |
1862 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
1863 | if (populated_zone(z)) { | |
1864 | zonelist->zones[pos++] = z; | |
1865 | check_highest_zone(zone_type); | |
1866 | } | |
1867 | } | |
1868 | } | |
1869 | zonelist->zones[pos] = NULL; | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | static int default_zonelist_order(void) | |
1874 | { | |
1875 | int nid, zone_type; | |
1876 | unsigned long low_kmem_size,total_size; | |
1877 | struct zone *z; | |
1878 | int average_size; | |
1879 | /* | |
1880 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
1881 | * If they are really small and used heavily, the system can fall | |
1882 | * into OOM very easily. | |
1883 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
1884 | */ | |
1885 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
1886 | low_kmem_size = 0; | |
1887 | total_size = 0; | |
1888 | for_each_online_node(nid) { | |
1889 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
1890 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
1891 | if (populated_zone(z)) { | |
1892 | if (zone_type < ZONE_NORMAL) | |
1893 | low_kmem_size += z->present_pages; | |
1894 | total_size += z->present_pages; | |
1895 | } | |
1896 | } | |
1897 | } | |
1898 | if (!low_kmem_size || /* there are no DMA area. */ | |
1899 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
1900 | return ZONELIST_ORDER_NODE; | |
1901 | /* | |
1902 | * look into each node's config. | |
1903 | * If there is a node whose DMA/DMA32 memory is very big area on | |
1904 | * local memory, NODE_ORDER may be suitable. | |
1905 | */ | |
1906 | average_size = total_size / (num_online_nodes() + 1); | |
1907 | for_each_online_node(nid) { | |
1908 | low_kmem_size = 0; | |
1909 | total_size = 0; | |
1910 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
1911 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
1912 | if (populated_zone(z)) { | |
1913 | if (zone_type < ZONE_NORMAL) | |
1914 | low_kmem_size += z->present_pages; | |
1915 | total_size += z->present_pages; | |
1916 | } | |
1917 | } | |
1918 | if (low_kmem_size && | |
1919 | total_size > average_size && /* ignore small node */ | |
1920 | low_kmem_size > total_size * 70/100) | |
1921 | return ZONELIST_ORDER_NODE; | |
1922 | } | |
1923 | return ZONELIST_ORDER_ZONE; | |
1924 | } | |
1925 | ||
1926 | static void set_zonelist_order(void) | |
1927 | { | |
1928 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
1929 | current_zonelist_order = default_zonelist_order(); | |
1930 | else | |
1931 | current_zonelist_order = user_zonelist_order; | |
1932 | } | |
1933 | ||
1934 | static void build_zonelists(pg_data_t *pgdat) | |
1935 | { | |
1936 | int j, node, load; | |
1937 | enum zone_type i; | |
1da177e4 | 1938 | nodemask_t used_mask; |
f0c0b2b8 KH |
1939 | int local_node, prev_node; |
1940 | struct zonelist *zonelist; | |
1941 | int order = current_zonelist_order; | |
1da177e4 LT |
1942 | |
1943 | /* initialize zonelists */ | |
19655d34 | 1944 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1945 | zonelist = pgdat->node_zonelists + i; |
1946 | zonelist->zones[0] = NULL; | |
1947 | } | |
1948 | ||
1949 | /* NUMA-aware ordering of nodes */ | |
1950 | local_node = pgdat->node_id; | |
1951 | load = num_online_nodes(); | |
1952 | prev_node = local_node; | |
1953 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
1954 | |
1955 | memset(node_load, 0, sizeof(node_load)); | |
1956 | memset(node_order, 0, sizeof(node_order)); | |
1957 | j = 0; | |
1958 | ||
1da177e4 | 1959 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
1960 | int distance = node_distance(local_node, node); |
1961 | ||
1962 | /* | |
1963 | * If another node is sufficiently far away then it is better | |
1964 | * to reclaim pages in a zone before going off node. | |
1965 | */ | |
1966 | if (distance > RECLAIM_DISTANCE) | |
1967 | zone_reclaim_mode = 1; | |
1968 | ||
1da177e4 LT |
1969 | /* |
1970 | * We don't want to pressure a particular node. | |
1971 | * So adding penalty to the first node in same | |
1972 | * distance group to make it round-robin. | |
1973 | */ | |
9eeff239 | 1974 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
1975 | node_load[node] = load; |
1976 | ||
1da177e4 LT |
1977 | prev_node = node; |
1978 | load--; | |
f0c0b2b8 KH |
1979 | if (order == ZONELIST_ORDER_NODE) |
1980 | build_zonelists_in_node_order(pgdat, node); | |
1981 | else | |
1982 | node_order[j++] = node; /* remember order */ | |
1983 | } | |
1da177e4 | 1984 | |
f0c0b2b8 KH |
1985 | if (order == ZONELIST_ORDER_ZONE) { |
1986 | /* calculate node order -- i.e., DMA last! */ | |
1987 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 LT |
1988 | } |
1989 | } | |
1990 | ||
9276b1bc | 1991 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 1992 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
1993 | { |
1994 | int i; | |
1995 | ||
1996 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1997 | struct zonelist *zonelist; | |
1998 | struct zonelist_cache *zlc; | |
1999 | struct zone **z; | |
2000 | ||
2001 | zonelist = pgdat->node_zonelists + i; | |
2002 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2003 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2004 | for (z = zonelist->zones; *z; z++) | |
2005 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
2006 | } | |
2007 | } | |
2008 | ||
f0c0b2b8 | 2009 | |
1da177e4 LT |
2010 | #else /* CONFIG_NUMA */ |
2011 | ||
f0c0b2b8 KH |
2012 | static void set_zonelist_order(void) |
2013 | { | |
2014 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2015 | } | |
2016 | ||
2017 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2018 | { |
19655d34 CL |
2019 | int node, local_node; |
2020 | enum zone_type i,j; | |
1da177e4 LT |
2021 | |
2022 | local_node = pgdat->node_id; | |
19655d34 | 2023 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
2024 | struct zonelist *zonelist; |
2025 | ||
2026 | zonelist = pgdat->node_zonelists + i; | |
2027 | ||
19655d34 | 2028 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
2029 | /* |
2030 | * Now we build the zonelist so that it contains the zones | |
2031 | * of all the other nodes. | |
2032 | * We don't want to pressure a particular node, so when | |
2033 | * building the zones for node N, we make sure that the | |
2034 | * zones coming right after the local ones are those from | |
2035 | * node N+1 (modulo N) | |
2036 | */ | |
2037 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2038 | if (!node_online(node)) | |
2039 | continue; | |
19655d34 | 2040 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2041 | } |
2042 | for (node = 0; node < local_node; node++) { | |
2043 | if (!node_online(node)) | |
2044 | continue; | |
19655d34 | 2045 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2046 | } |
2047 | ||
2048 | zonelist->zones[j] = NULL; | |
2049 | } | |
2050 | } | |
2051 | ||
9276b1bc | 2052 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2053 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2054 | { |
2055 | int i; | |
2056 | ||
2057 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2058 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
2059 | } | |
2060 | ||
1da177e4 LT |
2061 | #endif /* CONFIG_NUMA */ |
2062 | ||
6811378e | 2063 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2064 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2065 | { |
6811378e | 2066 | int nid; |
9276b1bc PJ |
2067 | |
2068 | for_each_online_node(nid) { | |
6811378e | 2069 | build_zonelists(NODE_DATA(nid)); |
9276b1bc PJ |
2070 | build_zonelist_cache(NODE_DATA(nid)); |
2071 | } | |
6811378e YG |
2072 | return 0; |
2073 | } | |
2074 | ||
f0c0b2b8 | 2075 | void build_all_zonelists(void) |
6811378e | 2076 | { |
f0c0b2b8 KH |
2077 | set_zonelist_order(); |
2078 | ||
6811378e | 2079 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2080 | __build_all_zonelists(NULL); |
6811378e YG |
2081 | cpuset_init_current_mems_allowed(); |
2082 | } else { | |
2083 | /* we have to stop all cpus to guaranntee there is no user | |
2084 | of zonelist */ | |
2085 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2086 | /* cpuset refresh routine should be here */ | |
2087 | } | |
bd1e22b8 | 2088 | vm_total_pages = nr_free_pagecache_pages(); |
f0c0b2b8 KH |
2089 | printk("Built %i zonelists in %s order. Total pages: %ld\n", |
2090 | num_online_nodes(), | |
2091 | zonelist_order_name[current_zonelist_order], | |
2092 | vm_total_pages); | |
2093 | #ifdef CONFIG_NUMA | |
2094 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2095 | #endif | |
1da177e4 LT |
2096 | } |
2097 | ||
2098 | /* | |
2099 | * Helper functions to size the waitqueue hash table. | |
2100 | * Essentially these want to choose hash table sizes sufficiently | |
2101 | * large so that collisions trying to wait on pages are rare. | |
2102 | * But in fact, the number of active page waitqueues on typical | |
2103 | * systems is ridiculously low, less than 200. So this is even | |
2104 | * conservative, even though it seems large. | |
2105 | * | |
2106 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2107 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2108 | */ | |
2109 | #define PAGES_PER_WAITQUEUE 256 | |
2110 | ||
cca448fe | 2111 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2112 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2113 | { |
2114 | unsigned long size = 1; | |
2115 | ||
2116 | pages /= PAGES_PER_WAITQUEUE; | |
2117 | ||
2118 | while (size < pages) | |
2119 | size <<= 1; | |
2120 | ||
2121 | /* | |
2122 | * Once we have dozens or even hundreds of threads sleeping | |
2123 | * on IO we've got bigger problems than wait queue collision. | |
2124 | * Limit the size of the wait table to a reasonable size. | |
2125 | */ | |
2126 | size = min(size, 4096UL); | |
2127 | ||
2128 | return max(size, 4UL); | |
2129 | } | |
cca448fe YG |
2130 | #else |
2131 | /* | |
2132 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2133 | * a suitable size for its wait_table. So we use the maximum size now. | |
2134 | * | |
2135 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2136 | * | |
2137 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2138 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2139 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2140 | * | |
2141 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2142 | * or more by the traditional way. (See above). It equals: | |
2143 | * | |
2144 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2145 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2146 | * powerpc (64K page size) : = (32G +16M)byte. | |
2147 | */ | |
2148 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2149 | { | |
2150 | return 4096UL; | |
2151 | } | |
2152 | #endif | |
1da177e4 LT |
2153 | |
2154 | /* | |
2155 | * This is an integer logarithm so that shifts can be used later | |
2156 | * to extract the more random high bits from the multiplicative | |
2157 | * hash function before the remainder is taken. | |
2158 | */ | |
2159 | static inline unsigned long wait_table_bits(unsigned long size) | |
2160 | { | |
2161 | return ffz(~size); | |
2162 | } | |
2163 | ||
2164 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2165 | ||
1da177e4 LT |
2166 | /* |
2167 | * Initially all pages are reserved - free ones are freed | |
2168 | * up by free_all_bootmem() once the early boot process is | |
2169 | * done. Non-atomic initialization, single-pass. | |
2170 | */ | |
c09b4240 | 2171 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2172 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2173 | { |
1da177e4 | 2174 | struct page *page; |
29751f69 AW |
2175 | unsigned long end_pfn = start_pfn + size; |
2176 | unsigned long pfn; | |
1da177e4 | 2177 | |
cbe8dd4a | 2178 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2179 | /* |
2180 | * There can be holes in boot-time mem_map[]s | |
2181 | * handed to this function. They do not | |
2182 | * exist on hotplugged memory. | |
2183 | */ | |
2184 | if (context == MEMMAP_EARLY) { | |
2185 | if (!early_pfn_valid(pfn)) | |
2186 | continue; | |
2187 | if (!early_pfn_in_nid(pfn, nid)) | |
2188 | continue; | |
2189 | } | |
d41dee36 AW |
2190 | page = pfn_to_page(pfn); |
2191 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 2192 | init_page_count(page); |
1da177e4 LT |
2193 | reset_page_mapcount(page); |
2194 | SetPageReserved(page); | |
2195 | INIT_LIST_HEAD(&page->lru); | |
2196 | #ifdef WANT_PAGE_VIRTUAL | |
2197 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2198 | if (!is_highmem_idx(zone)) | |
3212c6be | 2199 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2200 | #endif |
1da177e4 LT |
2201 | } |
2202 | } | |
2203 | ||
6ea6e688 PM |
2204 | static void __meminit zone_init_free_lists(struct pglist_data *pgdat, |
2205 | struct zone *zone, unsigned long size) | |
1da177e4 LT |
2206 | { |
2207 | int order; | |
2208 | for (order = 0; order < MAX_ORDER ; order++) { | |
2209 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
2210 | zone->free_area[order].nr_free = 0; | |
2211 | } | |
2212 | } | |
2213 | ||
2214 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2215 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2216 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2217 | #endif |
2218 | ||
d09c6b80 | 2219 | static int __devinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2220 | { |
2221 | int batch; | |
2222 | ||
2223 | /* | |
2224 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2225 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2226 | * |
2227 | * OK, so we don't know how big the cache is. So guess. | |
2228 | */ | |
2229 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2230 | if (batch * PAGE_SIZE > 512 * 1024) |
2231 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2232 | batch /= 4; /* We effectively *= 4 below */ |
2233 | if (batch < 1) | |
2234 | batch = 1; | |
2235 | ||
2236 | /* | |
0ceaacc9 NP |
2237 | * Clamp the batch to a 2^n - 1 value. Having a power |
2238 | * of 2 value was found to be more likely to have | |
2239 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2240 | * |
0ceaacc9 NP |
2241 | * For example if 2 tasks are alternately allocating |
2242 | * batches of pages, one task can end up with a lot | |
2243 | * of pages of one half of the possible page colors | |
2244 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2245 | */ |
0ceaacc9 | 2246 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2247 | |
e7c8d5c9 CL |
2248 | return batch; |
2249 | } | |
2250 | ||
2caaad41 CL |
2251 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2252 | { | |
2253 | struct per_cpu_pages *pcp; | |
2254 | ||
1c6fe946 MD |
2255 | memset(p, 0, sizeof(*p)); |
2256 | ||
2caaad41 CL |
2257 | pcp = &p->pcp[0]; /* hot */ |
2258 | pcp->count = 0; | |
2caaad41 CL |
2259 | pcp->high = 6 * batch; |
2260 | pcp->batch = max(1UL, 1 * batch); | |
2261 | INIT_LIST_HEAD(&pcp->list); | |
2262 | ||
2263 | pcp = &p->pcp[1]; /* cold*/ | |
2264 | pcp->count = 0; | |
2caaad41 | 2265 | pcp->high = 2 * batch; |
e46a5e28 | 2266 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
2267 | INIT_LIST_HEAD(&pcp->list); |
2268 | } | |
2269 | ||
8ad4b1fb RS |
2270 | /* |
2271 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2272 | * to the value high for the pageset p. | |
2273 | */ | |
2274 | ||
2275 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2276 | unsigned long high) | |
2277 | { | |
2278 | struct per_cpu_pages *pcp; | |
2279 | ||
2280 | pcp = &p->pcp[0]; /* hot list */ | |
2281 | pcp->high = high; | |
2282 | pcp->batch = max(1UL, high/4); | |
2283 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2284 | pcp->batch = PAGE_SHIFT * 8; | |
2285 | } | |
2286 | ||
2287 | ||
e7c8d5c9 CL |
2288 | #ifdef CONFIG_NUMA |
2289 | /* | |
2caaad41 CL |
2290 | * Boot pageset table. One per cpu which is going to be used for all |
2291 | * zones and all nodes. The parameters will be set in such a way | |
2292 | * that an item put on a list will immediately be handed over to | |
2293 | * the buddy list. This is safe since pageset manipulation is done | |
2294 | * with interrupts disabled. | |
2295 | * | |
2296 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2297 | * |
2298 | * The boot_pagesets must be kept even after bootup is complete for | |
2299 | * unused processors and/or zones. They do play a role for bootstrapping | |
2300 | * hotplugged processors. | |
2301 | * | |
2302 | * zoneinfo_show() and maybe other functions do | |
2303 | * not check if the processor is online before following the pageset pointer. | |
2304 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2305 | */ |
88a2a4ac | 2306 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2307 | |
2308 | /* | |
2309 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2310 | * per cpu pageset array in struct zone. |
2311 | */ | |
6292d9aa | 2312 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2313 | { |
2314 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
2315 | |
2316 | for_each_zone(zone) { | |
e7c8d5c9 | 2317 | |
66a55030 CL |
2318 | if (!populated_zone(zone)) |
2319 | continue; | |
2320 | ||
23316bc8 | 2321 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 2322 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 2323 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2324 | goto bad; |
e7c8d5c9 | 2325 | |
23316bc8 | 2326 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2327 | |
2328 | if (percpu_pagelist_fraction) | |
2329 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2330 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2331 | } |
2332 | ||
2333 | return 0; | |
2334 | bad: | |
2335 | for_each_zone(dzone) { | |
2336 | if (dzone == zone) | |
2337 | break; | |
23316bc8 NP |
2338 | kfree(zone_pcp(dzone, cpu)); |
2339 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2340 | } |
2341 | return -ENOMEM; | |
2342 | } | |
2343 | ||
2344 | static inline void free_zone_pagesets(int cpu) | |
2345 | { | |
e7c8d5c9 CL |
2346 | struct zone *zone; |
2347 | ||
2348 | for_each_zone(zone) { | |
2349 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2350 | ||
f3ef9ead DR |
2351 | /* Free per_cpu_pageset if it is slab allocated */ |
2352 | if (pset != &boot_pageset[cpu]) | |
2353 | kfree(pset); | |
e7c8d5c9 | 2354 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2355 | } |
e7c8d5c9 CL |
2356 | } |
2357 | ||
9c7b216d | 2358 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2359 | unsigned long action, |
2360 | void *hcpu) | |
2361 | { | |
2362 | int cpu = (long)hcpu; | |
2363 | int ret = NOTIFY_OK; | |
2364 | ||
2365 | switch (action) { | |
ce421c79 | 2366 | case CPU_UP_PREPARE: |
8bb78442 | 2367 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2368 | if (process_zones(cpu)) |
2369 | ret = NOTIFY_BAD; | |
2370 | break; | |
2371 | case CPU_UP_CANCELED: | |
8bb78442 | 2372 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2373 | case CPU_DEAD: |
8bb78442 | 2374 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2375 | free_zone_pagesets(cpu); |
2376 | break; | |
2377 | default: | |
2378 | break; | |
e7c8d5c9 CL |
2379 | } |
2380 | return ret; | |
2381 | } | |
2382 | ||
74b85f37 | 2383 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2384 | { &pageset_cpuup_callback, NULL, 0 }; |
2385 | ||
78d9955b | 2386 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2387 | { |
2388 | int err; | |
2389 | ||
2390 | /* Initialize per_cpu_pageset for cpu 0. | |
2391 | * A cpuup callback will do this for every cpu | |
2392 | * as it comes online | |
2393 | */ | |
2394 | err = process_zones(smp_processor_id()); | |
2395 | BUG_ON(err); | |
2396 | register_cpu_notifier(&pageset_notifier); | |
2397 | } | |
2398 | ||
2399 | #endif | |
2400 | ||
577a32f6 | 2401 | static noinline __init_refok |
cca448fe | 2402 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2403 | { |
2404 | int i; | |
2405 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2406 | size_t alloc_size; |
ed8ece2e DH |
2407 | |
2408 | /* | |
2409 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2410 | * per zone. | |
2411 | */ | |
02b694de YG |
2412 | zone->wait_table_hash_nr_entries = |
2413 | wait_table_hash_nr_entries(zone_size_pages); | |
2414 | zone->wait_table_bits = | |
2415 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2416 | alloc_size = zone->wait_table_hash_nr_entries |
2417 | * sizeof(wait_queue_head_t); | |
2418 | ||
2419 | if (system_state == SYSTEM_BOOTING) { | |
2420 | zone->wait_table = (wait_queue_head_t *) | |
2421 | alloc_bootmem_node(pgdat, alloc_size); | |
2422 | } else { | |
2423 | /* | |
2424 | * This case means that a zone whose size was 0 gets new memory | |
2425 | * via memory hot-add. | |
2426 | * But it may be the case that a new node was hot-added. In | |
2427 | * this case vmalloc() will not be able to use this new node's | |
2428 | * memory - this wait_table must be initialized to use this new | |
2429 | * node itself as well. | |
2430 | * To use this new node's memory, further consideration will be | |
2431 | * necessary. | |
2432 | */ | |
2433 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
2434 | } | |
2435 | if (!zone->wait_table) | |
2436 | return -ENOMEM; | |
ed8ece2e | 2437 | |
02b694de | 2438 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2439 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2440 | |
2441 | return 0; | |
ed8ece2e DH |
2442 | } |
2443 | ||
c09b4240 | 2444 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2445 | { |
2446 | int cpu; | |
2447 | unsigned long batch = zone_batchsize(zone); | |
2448 | ||
2449 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2450 | #ifdef CONFIG_NUMA | |
2451 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2452 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2453 | setup_pageset(&boot_pageset[cpu],0); |
2454 | #else | |
2455 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2456 | #endif | |
2457 | } | |
f5335c0f AB |
2458 | if (zone->present_pages) |
2459 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2460 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2461 | } |
2462 | ||
718127cc YG |
2463 | __meminit int init_currently_empty_zone(struct zone *zone, |
2464 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2465 | unsigned long size, |
2466 | enum memmap_context context) | |
ed8ece2e DH |
2467 | { |
2468 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2469 | int ret; |
2470 | ret = zone_wait_table_init(zone, size); | |
2471 | if (ret) | |
2472 | return ret; | |
ed8ece2e DH |
2473 | pgdat->nr_zones = zone_idx(zone) + 1; |
2474 | ||
ed8ece2e DH |
2475 | zone->zone_start_pfn = zone_start_pfn; |
2476 | ||
2477 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2478 | ||
2479 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
2480 | |
2481 | return 0; | |
ed8ece2e DH |
2482 | } |
2483 | ||
c713216d MG |
2484 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2485 | /* | |
2486 | * Basic iterator support. Return the first range of PFNs for a node | |
2487 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2488 | */ | |
a3142c8e | 2489 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2490 | { |
2491 | int i; | |
2492 | ||
2493 | for (i = 0; i < nr_nodemap_entries; i++) | |
2494 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2495 | return i; | |
2496 | ||
2497 | return -1; | |
2498 | } | |
2499 | ||
2500 | /* | |
2501 | * Basic iterator support. Return the next active range of PFNs for a node | |
2502 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2503 | */ | |
a3142c8e | 2504 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2505 | { |
2506 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2507 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2508 | return index; | |
2509 | ||
2510 | return -1; | |
2511 | } | |
2512 | ||
2513 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2514 | /* | |
2515 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2516 | * Architectures may implement their own version but if add_active_range() | |
2517 | * was used and there are no special requirements, this is a convenient | |
2518 | * alternative | |
2519 | */ | |
6f076f5d | 2520 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2521 | { |
2522 | int i; | |
2523 | ||
2524 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2525 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2526 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2527 | ||
2528 | if (start_pfn <= pfn && pfn < end_pfn) | |
2529 | return early_node_map[i].nid; | |
2530 | } | |
2531 | ||
2532 | return 0; | |
2533 | } | |
2534 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2535 | ||
2536 | /* Basic iterator support to walk early_node_map[] */ | |
2537 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2538 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2539 | i = next_active_region_index_in_nid(i, nid)) | |
2540 | ||
2541 | /** | |
2542 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2543 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2544 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2545 | * |
2546 | * If an architecture guarantees that all ranges registered with | |
2547 | * add_active_ranges() contain no holes and may be freed, this | |
2548 | * this function may be used instead of calling free_bootmem() manually. | |
2549 | */ | |
2550 | void __init free_bootmem_with_active_regions(int nid, | |
2551 | unsigned long max_low_pfn) | |
2552 | { | |
2553 | int i; | |
2554 | ||
2555 | for_each_active_range_index_in_nid(i, nid) { | |
2556 | unsigned long size_pages = 0; | |
2557 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2558 | ||
2559 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2560 | continue; | |
2561 | ||
2562 | if (end_pfn > max_low_pfn) | |
2563 | end_pfn = max_low_pfn; | |
2564 | ||
2565 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2566 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2567 | PFN_PHYS(early_node_map[i].start_pfn), | |
2568 | size_pages << PAGE_SHIFT); | |
2569 | } | |
2570 | } | |
2571 | ||
2572 | /** | |
2573 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2574 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2575 | * |
2576 | * If an architecture guarantees that all ranges registered with | |
2577 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2578 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2579 | */ |
2580 | void __init sparse_memory_present_with_active_regions(int nid) | |
2581 | { | |
2582 | int i; | |
2583 | ||
2584 | for_each_active_range_index_in_nid(i, nid) | |
2585 | memory_present(early_node_map[i].nid, | |
2586 | early_node_map[i].start_pfn, | |
2587 | early_node_map[i].end_pfn); | |
2588 | } | |
2589 | ||
fb01439c MG |
2590 | /** |
2591 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2592 | * @nid: The nid of the node to push the boundary for | |
2593 | * @start_pfn: The start pfn of the node | |
2594 | * @end_pfn: The end pfn of the node | |
2595 | * | |
2596 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2597 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2598 | * be hotplugged even though no physical memory exists. This function allows | |
2599 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2600 | * be used later. | |
2601 | */ | |
2602 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2603 | void __init push_node_boundaries(unsigned int nid, | |
2604 | unsigned long start_pfn, unsigned long end_pfn) | |
2605 | { | |
2606 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2607 | nid, start_pfn, end_pfn); | |
2608 | ||
2609 | /* Initialise the boundary for this node if necessary */ | |
2610 | if (node_boundary_end_pfn[nid] == 0) | |
2611 | node_boundary_start_pfn[nid] = -1UL; | |
2612 | ||
2613 | /* Update the boundaries */ | |
2614 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2615 | node_boundary_start_pfn[nid] = start_pfn; | |
2616 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2617 | node_boundary_end_pfn[nid] = end_pfn; | |
2618 | } | |
2619 | ||
2620 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2621 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2622 | unsigned long *start_pfn, unsigned long *end_pfn) |
2623 | { | |
2624 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2625 | nid, *start_pfn, *end_pfn); | |
2626 | ||
2627 | /* Return if boundary information has not been provided */ | |
2628 | if (node_boundary_end_pfn[nid] == 0) | |
2629 | return; | |
2630 | ||
2631 | /* Check the boundaries and update if necessary */ | |
2632 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2633 | *start_pfn = node_boundary_start_pfn[nid]; | |
2634 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2635 | *end_pfn = node_boundary_end_pfn[nid]; | |
2636 | } | |
2637 | #else | |
2638 | void __init push_node_boundaries(unsigned int nid, | |
2639 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2640 | ||
98011f56 | 2641 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2642 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
2643 | #endif | |
2644 | ||
2645 | ||
c713216d MG |
2646 | /** |
2647 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2648 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2649 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2650 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2651 | * |
2652 | * It returns the start and end page frame of a node based on information | |
2653 | * provided by an arch calling add_active_range(). If called for a node | |
2654 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2655 | * PFNs will be 0. |
c713216d | 2656 | */ |
a3142c8e | 2657 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
2658 | unsigned long *start_pfn, unsigned long *end_pfn) |
2659 | { | |
2660 | int i; | |
2661 | *start_pfn = -1UL; | |
2662 | *end_pfn = 0; | |
2663 | ||
2664 | for_each_active_range_index_in_nid(i, nid) { | |
2665 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2666 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2667 | } | |
2668 | ||
2669 | if (*start_pfn == -1UL) { | |
2670 | printk(KERN_WARNING "Node %u active with no memory\n", nid); | |
2671 | *start_pfn = 0; | |
2672 | } | |
fb01439c MG |
2673 | |
2674 | /* Push the node boundaries out if requested */ | |
2675 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
2676 | } |
2677 | ||
2a1e274a MG |
2678 | /* |
2679 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
2680 | * assumption is made that zones within a node are ordered in monotonic | |
2681 | * increasing memory addresses so that the "highest" populated zone is used | |
2682 | */ | |
2683 | void __init find_usable_zone_for_movable(void) | |
2684 | { | |
2685 | int zone_index; | |
2686 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
2687 | if (zone_index == ZONE_MOVABLE) | |
2688 | continue; | |
2689 | ||
2690 | if (arch_zone_highest_possible_pfn[zone_index] > | |
2691 | arch_zone_lowest_possible_pfn[zone_index]) | |
2692 | break; | |
2693 | } | |
2694 | ||
2695 | VM_BUG_ON(zone_index == -1); | |
2696 | movable_zone = zone_index; | |
2697 | } | |
2698 | ||
2699 | /* | |
2700 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
2701 | * because it is sized independant of architecture. Unlike the other zones, | |
2702 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
2703 | * in each node depending on the size of each node and how evenly kernelcore | |
2704 | * is distributed. This helper function adjusts the zone ranges | |
2705 | * provided by the architecture for a given node by using the end of the | |
2706 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
2707 | * zones within a node are in order of monotonic increases memory addresses | |
2708 | */ | |
2709 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
2710 | unsigned long zone_type, | |
2711 | unsigned long node_start_pfn, | |
2712 | unsigned long node_end_pfn, | |
2713 | unsigned long *zone_start_pfn, | |
2714 | unsigned long *zone_end_pfn) | |
2715 | { | |
2716 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
2717 | if (zone_movable_pfn[nid]) { | |
2718 | /* Size ZONE_MOVABLE */ | |
2719 | if (zone_type == ZONE_MOVABLE) { | |
2720 | *zone_start_pfn = zone_movable_pfn[nid]; | |
2721 | *zone_end_pfn = min(node_end_pfn, | |
2722 | arch_zone_highest_possible_pfn[movable_zone]); | |
2723 | ||
2724 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
2725 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
2726 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
2727 | *zone_end_pfn = zone_movable_pfn[nid]; | |
2728 | ||
2729 | /* Check if this whole range is within ZONE_MOVABLE */ | |
2730 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
2731 | *zone_start_pfn = *zone_end_pfn; | |
2732 | } | |
2733 | } | |
2734 | ||
c713216d MG |
2735 | /* |
2736 | * Return the number of pages a zone spans in a node, including holes | |
2737 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2738 | */ | |
6ea6e688 | 2739 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
2740 | unsigned long zone_type, |
2741 | unsigned long *ignored) | |
2742 | { | |
2743 | unsigned long node_start_pfn, node_end_pfn; | |
2744 | unsigned long zone_start_pfn, zone_end_pfn; | |
2745 | ||
2746 | /* Get the start and end of the node and zone */ | |
2747 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2748 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2749 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
2750 | adjust_zone_range_for_zone_movable(nid, zone_type, |
2751 | node_start_pfn, node_end_pfn, | |
2752 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
2753 | |
2754 | /* Check that this node has pages within the zone's required range */ | |
2755 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2756 | return 0; | |
2757 | ||
2758 | /* Move the zone boundaries inside the node if necessary */ | |
2759 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2760 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2761 | ||
2762 | /* Return the spanned pages */ | |
2763 | return zone_end_pfn - zone_start_pfn; | |
2764 | } | |
2765 | ||
2766 | /* | |
2767 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 2768 | * then all holes in the requested range will be accounted for. |
c713216d | 2769 | */ |
a3142c8e | 2770 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
2771 | unsigned long range_start_pfn, |
2772 | unsigned long range_end_pfn) | |
2773 | { | |
2774 | int i = 0; | |
2775 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2776 | unsigned long start_pfn; | |
2777 | ||
2778 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2779 | i = first_active_region_index_in_nid(nid); | |
2780 | if (i == -1) | |
2781 | return 0; | |
2782 | ||
9c7cd687 MG |
2783 | /* Account for ranges before physical memory on this node */ |
2784 | if (early_node_map[i].start_pfn > range_start_pfn) | |
2785 | hole_pages = early_node_map[i].start_pfn - range_start_pfn; | |
2786 | ||
c713216d MG |
2787 | prev_end_pfn = early_node_map[i].start_pfn; |
2788 | ||
2789 | /* Find all holes for the zone within the node */ | |
2790 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2791 | ||
2792 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2793 | if (prev_end_pfn >= range_end_pfn) | |
2794 | break; | |
2795 | ||
2796 | /* Make sure the end of the zone is not within the hole */ | |
2797 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2798 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2799 | ||
2800 | /* Update the hole size cound and move on */ | |
2801 | if (start_pfn > range_start_pfn) { | |
2802 | BUG_ON(prev_end_pfn > start_pfn); | |
2803 | hole_pages += start_pfn - prev_end_pfn; | |
2804 | } | |
2805 | prev_end_pfn = early_node_map[i].end_pfn; | |
2806 | } | |
2807 | ||
9c7cd687 MG |
2808 | /* Account for ranges past physical memory on this node */ |
2809 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 2810 | hole_pages += range_end_pfn - |
9c7cd687 MG |
2811 | max(range_start_pfn, prev_end_pfn); |
2812 | ||
c713216d MG |
2813 | return hole_pages; |
2814 | } | |
2815 | ||
2816 | /** | |
2817 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2818 | * @start_pfn: The start PFN to start searching for holes | |
2819 | * @end_pfn: The end PFN to stop searching for holes | |
2820 | * | |
88ca3b94 | 2821 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
2822 | */ |
2823 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2824 | unsigned long end_pfn) | |
2825 | { | |
2826 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2827 | } | |
2828 | ||
2829 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 2830 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
2831 | unsigned long zone_type, |
2832 | unsigned long *ignored) | |
2833 | { | |
9c7cd687 MG |
2834 | unsigned long node_start_pfn, node_end_pfn; |
2835 | unsigned long zone_start_pfn, zone_end_pfn; | |
2836 | ||
2837 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2838 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
2839 | node_start_pfn); | |
2840 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
2841 | node_end_pfn); | |
2842 | ||
2a1e274a MG |
2843 | adjust_zone_range_for_zone_movable(nid, zone_type, |
2844 | node_start_pfn, node_end_pfn, | |
2845 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 2846 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 2847 | } |
0e0b864e | 2848 | |
c713216d | 2849 | #else |
6ea6e688 | 2850 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
2851 | unsigned long zone_type, |
2852 | unsigned long *zones_size) | |
2853 | { | |
2854 | return zones_size[zone_type]; | |
2855 | } | |
2856 | ||
6ea6e688 | 2857 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
2858 | unsigned long zone_type, |
2859 | unsigned long *zholes_size) | |
2860 | { | |
2861 | if (!zholes_size) | |
2862 | return 0; | |
2863 | ||
2864 | return zholes_size[zone_type]; | |
2865 | } | |
0e0b864e | 2866 | |
c713216d MG |
2867 | #endif |
2868 | ||
a3142c8e | 2869 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
2870 | unsigned long *zones_size, unsigned long *zholes_size) |
2871 | { | |
2872 | unsigned long realtotalpages, totalpages = 0; | |
2873 | enum zone_type i; | |
2874 | ||
2875 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2876 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
2877 | zones_size); | |
2878 | pgdat->node_spanned_pages = totalpages; | |
2879 | ||
2880 | realtotalpages = totalpages; | |
2881 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2882 | realtotalpages -= | |
2883 | zone_absent_pages_in_node(pgdat->node_id, i, | |
2884 | zholes_size); | |
2885 | pgdat->node_present_pages = realtotalpages; | |
2886 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
2887 | realtotalpages); | |
2888 | } | |
2889 | ||
1da177e4 LT |
2890 | /* |
2891 | * Set up the zone data structures: | |
2892 | * - mark all pages reserved | |
2893 | * - mark all memory queues empty | |
2894 | * - clear the memory bitmaps | |
2895 | */ | |
86356ab1 | 2896 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
2897 | unsigned long *zones_size, unsigned long *zholes_size) |
2898 | { | |
2f1b6248 | 2899 | enum zone_type j; |
ed8ece2e | 2900 | int nid = pgdat->node_id; |
1da177e4 | 2901 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 2902 | int ret; |
1da177e4 | 2903 | |
208d54e5 | 2904 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2905 | pgdat->nr_zones = 0; |
2906 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2907 | pgdat->kswapd_max_order = 0; | |
2908 | ||
2909 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2910 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 2911 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 2912 | |
c713216d MG |
2913 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
2914 | realsize = size - zone_absent_pages_in_node(nid, j, | |
2915 | zholes_size); | |
1da177e4 | 2916 | |
0e0b864e MG |
2917 | /* |
2918 | * Adjust realsize so that it accounts for how much memory | |
2919 | * is used by this zone for memmap. This affects the watermark | |
2920 | * and per-cpu initialisations | |
2921 | */ | |
2922 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
2923 | if (realsize >= memmap_pages) { | |
2924 | realsize -= memmap_pages; | |
2925 | printk(KERN_DEBUG | |
2926 | " %s zone: %lu pages used for memmap\n", | |
2927 | zone_names[j], memmap_pages); | |
2928 | } else | |
2929 | printk(KERN_WARNING | |
2930 | " %s zone: %lu pages exceeds realsize %lu\n", | |
2931 | zone_names[j], memmap_pages, realsize); | |
2932 | ||
6267276f CL |
2933 | /* Account for reserved pages */ |
2934 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 2935 | realsize -= dma_reserve; |
6267276f CL |
2936 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
2937 | zone_names[0], dma_reserve); | |
0e0b864e MG |
2938 | } |
2939 | ||
98d2b0eb | 2940 | if (!is_highmem_idx(j)) |
1da177e4 LT |
2941 | nr_kernel_pages += realsize; |
2942 | nr_all_pages += realsize; | |
2943 | ||
2944 | zone->spanned_pages = size; | |
2945 | zone->present_pages = realsize; | |
9614634f | 2946 | #ifdef CONFIG_NUMA |
d5f541ed | 2947 | zone->node = nid; |
8417bba4 | 2948 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 2949 | / 100; |
0ff38490 | 2950 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 2951 | #endif |
1da177e4 LT |
2952 | zone->name = zone_names[j]; |
2953 | spin_lock_init(&zone->lock); | |
2954 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2955 | zone_seqlock_init(zone); |
1da177e4 | 2956 | zone->zone_pgdat = pgdat; |
1da177e4 | 2957 | |
3bb1a852 | 2958 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 2959 | |
ed8ece2e | 2960 | zone_pcp_init(zone); |
1da177e4 LT |
2961 | INIT_LIST_HEAD(&zone->active_list); |
2962 | INIT_LIST_HEAD(&zone->inactive_list); | |
2963 | zone->nr_scan_active = 0; | |
2964 | zone->nr_scan_inactive = 0; | |
2244b95a | 2965 | zap_zone_vm_stats(zone); |
53e9a615 | 2966 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2967 | if (!size) |
2968 | continue; | |
2969 | ||
a2f3aa02 DH |
2970 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
2971 | size, MEMMAP_EARLY); | |
718127cc | 2972 | BUG_ON(ret); |
1da177e4 | 2973 | zone_start_pfn += size; |
1da177e4 LT |
2974 | } |
2975 | } | |
2976 | ||
577a32f6 | 2977 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 2978 | { |
1da177e4 LT |
2979 | /* Skip empty nodes */ |
2980 | if (!pgdat->node_spanned_pages) | |
2981 | return; | |
2982 | ||
d41dee36 | 2983 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2984 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2985 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2986 | unsigned long size, start, end; |
d41dee36 AW |
2987 | struct page *map; |
2988 | ||
e984bb43 BP |
2989 | /* |
2990 | * The zone's endpoints aren't required to be MAX_ORDER | |
2991 | * aligned but the node_mem_map endpoints must be in order | |
2992 | * for the buddy allocator to function correctly. | |
2993 | */ | |
2994 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2995 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2996 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2997 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2998 | map = alloc_remap(pgdat->node_id, size); |
2999 | if (!map) | |
3000 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3001 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3002 | } |
12d810c1 | 3003 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3004 | /* |
3005 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3006 | */ | |
c713216d | 3007 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3008 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3009 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3010 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
3011 | mem_map -= pgdat->node_start_pfn; | |
3012 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
3013 | } | |
1da177e4 | 3014 | #endif |
d41dee36 | 3015 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3016 | } |
3017 | ||
86356ab1 | 3018 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
3019 | unsigned long *zones_size, unsigned long node_start_pfn, |
3020 | unsigned long *zholes_size) | |
3021 | { | |
3022 | pgdat->node_id = nid; | |
3023 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3024 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3025 | |
3026 | alloc_node_mem_map(pgdat); | |
3027 | ||
3028 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3029 | } | |
3030 | ||
c713216d | 3031 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3032 | |
3033 | #if MAX_NUMNODES > 1 | |
3034 | /* | |
3035 | * Figure out the number of possible node ids. | |
3036 | */ | |
3037 | static void __init setup_nr_node_ids(void) | |
3038 | { | |
3039 | unsigned int node; | |
3040 | unsigned int highest = 0; | |
3041 | ||
3042 | for_each_node_mask(node, node_possible_map) | |
3043 | highest = node; | |
3044 | nr_node_ids = highest + 1; | |
3045 | } | |
3046 | #else | |
3047 | static inline void setup_nr_node_ids(void) | |
3048 | { | |
3049 | } | |
3050 | #endif | |
3051 | ||
c713216d MG |
3052 | /** |
3053 | * add_active_range - Register a range of PFNs backed by physical memory | |
3054 | * @nid: The node ID the range resides on | |
3055 | * @start_pfn: The start PFN of the available physical memory | |
3056 | * @end_pfn: The end PFN of the available physical memory | |
3057 | * | |
3058 | * These ranges are stored in an early_node_map[] and later used by | |
3059 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3060 | * range spans a memory hole, it is up to the architecture to ensure | |
3061 | * the memory is not freed by the bootmem allocator. If possible | |
3062 | * the range being registered will be merged with existing ranges. | |
3063 | */ | |
3064 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3065 | unsigned long end_pfn) | |
3066 | { | |
3067 | int i; | |
3068 | ||
3069 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
3070 | "%d entries of %d used\n", | |
3071 | nid, start_pfn, end_pfn, | |
3072 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3073 | ||
3074 | /* Merge with existing active regions if possible */ | |
3075 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3076 | if (early_node_map[i].nid != nid) | |
3077 | continue; | |
3078 | ||
3079 | /* Skip if an existing region covers this new one */ | |
3080 | if (start_pfn >= early_node_map[i].start_pfn && | |
3081 | end_pfn <= early_node_map[i].end_pfn) | |
3082 | return; | |
3083 | ||
3084 | /* Merge forward if suitable */ | |
3085 | if (start_pfn <= early_node_map[i].end_pfn && | |
3086 | end_pfn > early_node_map[i].end_pfn) { | |
3087 | early_node_map[i].end_pfn = end_pfn; | |
3088 | return; | |
3089 | } | |
3090 | ||
3091 | /* Merge backward if suitable */ | |
3092 | if (start_pfn < early_node_map[i].end_pfn && | |
3093 | end_pfn >= early_node_map[i].start_pfn) { | |
3094 | early_node_map[i].start_pfn = start_pfn; | |
3095 | return; | |
3096 | } | |
3097 | } | |
3098 | ||
3099 | /* Check that early_node_map is large enough */ | |
3100 | if (i >= MAX_ACTIVE_REGIONS) { | |
3101 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3102 | MAX_ACTIVE_REGIONS); | |
3103 | return; | |
3104 | } | |
3105 | ||
3106 | early_node_map[i].nid = nid; | |
3107 | early_node_map[i].start_pfn = start_pfn; | |
3108 | early_node_map[i].end_pfn = end_pfn; | |
3109 | nr_nodemap_entries = i + 1; | |
3110 | } | |
3111 | ||
3112 | /** | |
3113 | * shrink_active_range - Shrink an existing registered range of PFNs | |
3114 | * @nid: The node id the range is on that should be shrunk | |
3115 | * @old_end_pfn: The old end PFN of the range | |
3116 | * @new_end_pfn: The new PFN of the range | |
3117 | * | |
3118 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
3119 | * The map is kept at the end physical page range that has already been | |
3120 | * registered with add_active_range(). This function allows an arch to shrink | |
3121 | * an existing registered range. | |
3122 | */ | |
3123 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
3124 | unsigned long new_end_pfn) | |
3125 | { | |
3126 | int i; | |
3127 | ||
3128 | /* Find the old active region end and shrink */ | |
3129 | for_each_active_range_index_in_nid(i, nid) | |
3130 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
3131 | early_node_map[i].end_pfn = new_end_pfn; | |
3132 | break; | |
3133 | } | |
3134 | } | |
3135 | ||
3136 | /** | |
3137 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3138 | * |
c713216d MG |
3139 | * During discovery, it may be found that a table like SRAT is invalid |
3140 | * and an alternative discovery method must be used. This function removes | |
3141 | * all currently registered regions. | |
3142 | */ | |
88ca3b94 | 3143 | void __init remove_all_active_ranges(void) |
c713216d MG |
3144 | { |
3145 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3146 | nr_nodemap_entries = 0; | |
fb01439c MG |
3147 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3148 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3149 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3150 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3151 | } |
3152 | ||
3153 | /* Compare two active node_active_regions */ | |
3154 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3155 | { | |
3156 | struct node_active_region *arange = (struct node_active_region *)a; | |
3157 | struct node_active_region *brange = (struct node_active_region *)b; | |
3158 | ||
3159 | /* Done this way to avoid overflows */ | |
3160 | if (arange->start_pfn > brange->start_pfn) | |
3161 | return 1; | |
3162 | if (arange->start_pfn < brange->start_pfn) | |
3163 | return -1; | |
3164 | ||
3165 | return 0; | |
3166 | } | |
3167 | ||
3168 | /* sort the node_map by start_pfn */ | |
3169 | static void __init sort_node_map(void) | |
3170 | { | |
3171 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3172 | sizeof(struct node_active_region), | |
3173 | cmp_node_active_region, NULL); | |
3174 | } | |
3175 | ||
a6af2bc3 | 3176 | /* Find the lowest pfn for a node */ |
c713216d MG |
3177 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
3178 | { | |
3179 | int i; | |
a6af2bc3 | 3180 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3181 | |
c713216d MG |
3182 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3183 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3184 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3185 | |
a6af2bc3 MG |
3186 | if (min_pfn == ULONG_MAX) { |
3187 | printk(KERN_WARNING | |
3188 | "Could not find start_pfn for node %lu\n", nid); | |
3189 | return 0; | |
3190 | } | |
3191 | ||
3192 | return min_pfn; | |
c713216d MG |
3193 | } |
3194 | ||
3195 | /** | |
3196 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3197 | * | |
3198 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3199 | * add_active_range(). |
c713216d MG |
3200 | */ |
3201 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3202 | { | |
3203 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3204 | } | |
3205 | ||
3206 | /** | |
3207 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3208 | * | |
3209 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3210 | * add_active_range(). |
c713216d MG |
3211 | */ |
3212 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3213 | { | |
3214 | int i; | |
3215 | unsigned long max_pfn = 0; | |
3216 | ||
3217 | for (i = 0; i < nr_nodemap_entries; i++) | |
3218 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3219 | ||
3220 | return max_pfn; | |
3221 | } | |
3222 | ||
7e63efef MG |
3223 | unsigned long __init early_calculate_totalpages(void) |
3224 | { | |
3225 | int i; | |
3226 | unsigned long totalpages = 0; | |
3227 | ||
3228 | for (i = 0; i < nr_nodemap_entries; i++) | |
3229 | totalpages += early_node_map[i].end_pfn - | |
3230 | early_node_map[i].start_pfn; | |
3231 | ||
3232 | return totalpages; | |
3233 | } | |
3234 | ||
2a1e274a MG |
3235 | /* |
3236 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3237 | * is spread evenly between nodes as long as the nodes have enough | |
3238 | * memory. When they don't, some nodes will have more kernelcore than | |
3239 | * others | |
3240 | */ | |
3241 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3242 | { | |
3243 | int i, nid; | |
3244 | unsigned long usable_startpfn; | |
3245 | unsigned long kernelcore_node, kernelcore_remaining; | |
3246 | int usable_nodes = num_online_nodes(); | |
3247 | ||
7e63efef MG |
3248 | /* |
3249 | * If movablecore was specified, calculate what size of | |
3250 | * kernelcore that corresponds so that memory usable for | |
3251 | * any allocation type is evenly spread. If both kernelcore | |
3252 | * and movablecore are specified, then the value of kernelcore | |
3253 | * will be used for required_kernelcore if it's greater than | |
3254 | * what movablecore would have allowed. | |
3255 | */ | |
3256 | if (required_movablecore) { | |
3257 | unsigned long totalpages = early_calculate_totalpages(); | |
3258 | unsigned long corepages; | |
3259 | ||
3260 | /* | |
3261 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3262 | * was requested by the user | |
3263 | */ | |
3264 | required_movablecore = | |
3265 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3266 | corepages = totalpages - required_movablecore; | |
3267 | ||
3268 | required_kernelcore = max(required_kernelcore, corepages); | |
3269 | } | |
3270 | ||
2a1e274a MG |
3271 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3272 | if (!required_kernelcore) | |
3273 | return; | |
3274 | ||
3275 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3276 | find_usable_zone_for_movable(); | |
3277 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3278 | ||
3279 | restart: | |
3280 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3281 | kernelcore_node = required_kernelcore / usable_nodes; | |
3282 | for_each_online_node(nid) { | |
3283 | /* | |
3284 | * Recalculate kernelcore_node if the division per node | |
3285 | * now exceeds what is necessary to satisfy the requested | |
3286 | * amount of memory for the kernel | |
3287 | */ | |
3288 | if (required_kernelcore < kernelcore_node) | |
3289 | kernelcore_node = required_kernelcore / usable_nodes; | |
3290 | ||
3291 | /* | |
3292 | * As the map is walked, we track how much memory is usable | |
3293 | * by the kernel using kernelcore_remaining. When it is | |
3294 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3295 | */ | |
3296 | kernelcore_remaining = kernelcore_node; | |
3297 | ||
3298 | /* Go through each range of PFNs within this node */ | |
3299 | for_each_active_range_index_in_nid(i, nid) { | |
3300 | unsigned long start_pfn, end_pfn; | |
3301 | unsigned long size_pages; | |
3302 | ||
3303 | start_pfn = max(early_node_map[i].start_pfn, | |
3304 | zone_movable_pfn[nid]); | |
3305 | end_pfn = early_node_map[i].end_pfn; | |
3306 | if (start_pfn >= end_pfn) | |
3307 | continue; | |
3308 | ||
3309 | /* Account for what is only usable for kernelcore */ | |
3310 | if (start_pfn < usable_startpfn) { | |
3311 | unsigned long kernel_pages; | |
3312 | kernel_pages = min(end_pfn, usable_startpfn) | |
3313 | - start_pfn; | |
3314 | ||
3315 | kernelcore_remaining -= min(kernel_pages, | |
3316 | kernelcore_remaining); | |
3317 | required_kernelcore -= min(kernel_pages, | |
3318 | required_kernelcore); | |
3319 | ||
3320 | /* Continue if range is now fully accounted */ | |
3321 | if (end_pfn <= usable_startpfn) { | |
3322 | ||
3323 | /* | |
3324 | * Push zone_movable_pfn to the end so | |
3325 | * that if we have to rebalance | |
3326 | * kernelcore across nodes, we will | |
3327 | * not double account here | |
3328 | */ | |
3329 | zone_movable_pfn[nid] = end_pfn; | |
3330 | continue; | |
3331 | } | |
3332 | start_pfn = usable_startpfn; | |
3333 | } | |
3334 | ||
3335 | /* | |
3336 | * The usable PFN range for ZONE_MOVABLE is from | |
3337 | * start_pfn->end_pfn. Calculate size_pages as the | |
3338 | * number of pages used as kernelcore | |
3339 | */ | |
3340 | size_pages = end_pfn - start_pfn; | |
3341 | if (size_pages > kernelcore_remaining) | |
3342 | size_pages = kernelcore_remaining; | |
3343 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3344 | ||
3345 | /* | |
3346 | * Some kernelcore has been met, update counts and | |
3347 | * break if the kernelcore for this node has been | |
3348 | * satisified | |
3349 | */ | |
3350 | required_kernelcore -= min(required_kernelcore, | |
3351 | size_pages); | |
3352 | kernelcore_remaining -= size_pages; | |
3353 | if (!kernelcore_remaining) | |
3354 | break; | |
3355 | } | |
3356 | } | |
3357 | ||
3358 | /* | |
3359 | * If there is still required_kernelcore, we do another pass with one | |
3360 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3361 | * along on the nodes that still have memory until kernelcore is | |
3362 | * satisified | |
3363 | */ | |
3364 | usable_nodes--; | |
3365 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3366 | goto restart; | |
3367 | ||
3368 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3369 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3370 | zone_movable_pfn[nid] = | |
3371 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3372 | } | |
3373 | ||
c713216d MG |
3374 | /** |
3375 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3376 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3377 | * |
3378 | * This will call free_area_init_node() for each active node in the system. | |
3379 | * Using the page ranges provided by add_active_range(), the size of each | |
3380 | * zone in each node and their holes is calculated. If the maximum PFN | |
3381 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3382 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3383 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3384 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3385 | * at arch_max_dma_pfn. | |
3386 | */ | |
3387 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3388 | { | |
3389 | unsigned long nid; | |
3390 | enum zone_type i; | |
3391 | ||
a6af2bc3 MG |
3392 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3393 | sort_node_map(); | |
3394 | ||
c713216d MG |
3395 | /* Record where the zone boundaries are */ |
3396 | memset(arch_zone_lowest_possible_pfn, 0, | |
3397 | sizeof(arch_zone_lowest_possible_pfn)); | |
3398 | memset(arch_zone_highest_possible_pfn, 0, | |
3399 | sizeof(arch_zone_highest_possible_pfn)); | |
3400 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3401 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3402 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3403 | if (i == ZONE_MOVABLE) |
3404 | continue; | |
c713216d MG |
3405 | arch_zone_lowest_possible_pfn[i] = |
3406 | arch_zone_highest_possible_pfn[i-1]; | |
3407 | arch_zone_highest_possible_pfn[i] = | |
3408 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3409 | } | |
2a1e274a MG |
3410 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3411 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3412 | ||
3413 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3414 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3415 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3416 | |
c713216d MG |
3417 | /* Print out the zone ranges */ |
3418 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3419 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3420 | if (i == ZONE_MOVABLE) | |
3421 | continue; | |
c713216d MG |
3422 | printk(" %-8s %8lu -> %8lu\n", |
3423 | zone_names[i], | |
3424 | arch_zone_lowest_possible_pfn[i], | |
3425 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3426 | } |
3427 | ||
3428 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3429 | printk("Movable zone start PFN for each node\n"); | |
3430 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3431 | if (zone_movable_pfn[i]) | |
3432 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3433 | } | |
c713216d MG |
3434 | |
3435 | /* Print out the early_node_map[] */ | |
3436 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3437 | for (i = 0; i < nr_nodemap_entries; i++) | |
3438 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
3439 | early_node_map[i].start_pfn, | |
3440 | early_node_map[i].end_pfn); | |
3441 | ||
3442 | /* Initialise every node */ | |
8ef82866 | 3443 | setup_nr_node_ids(); |
c713216d MG |
3444 | for_each_online_node(nid) { |
3445 | pg_data_t *pgdat = NODE_DATA(nid); | |
3446 | free_area_init_node(nid, pgdat, NULL, | |
3447 | find_min_pfn_for_node(nid), NULL); | |
3448 | } | |
3449 | } | |
2a1e274a | 3450 | |
7e63efef | 3451 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3452 | { |
3453 | unsigned long long coremem; | |
3454 | if (!p) | |
3455 | return -EINVAL; | |
3456 | ||
3457 | coremem = memparse(p, &p); | |
7e63efef | 3458 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3459 | |
7e63efef | 3460 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3461 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3462 | ||
3463 | return 0; | |
3464 | } | |
ed7ed365 | 3465 | |
7e63efef MG |
3466 | /* |
3467 | * kernelcore=size sets the amount of memory for use for allocations that | |
3468 | * cannot be reclaimed or migrated. | |
3469 | */ | |
3470 | static int __init cmdline_parse_kernelcore(char *p) | |
3471 | { | |
3472 | return cmdline_parse_core(p, &required_kernelcore); | |
3473 | } | |
3474 | ||
3475 | /* | |
3476 | * movablecore=size sets the amount of memory for use for allocations that | |
3477 | * can be reclaimed or migrated. | |
3478 | */ | |
3479 | static int __init cmdline_parse_movablecore(char *p) | |
3480 | { | |
3481 | return cmdline_parse_core(p, &required_movablecore); | |
3482 | } | |
3483 | ||
ed7ed365 | 3484 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 3485 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 3486 | |
c713216d MG |
3487 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3488 | ||
0e0b864e | 3489 | /** |
88ca3b94 RD |
3490 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
3491 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
3492 | * |
3493 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
3494 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
3495 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
3496 | * function may optionally be used to account for unfreeable pages in the |
3497 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
3498 | * smaller per-cpu batchsize. | |
0e0b864e MG |
3499 | */ |
3500 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
3501 | { | |
3502 | dma_reserve = new_dma_reserve; | |
3503 | } | |
3504 | ||
93b7504e | 3505 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3506 | static bootmem_data_t contig_bootmem_data; |
3507 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
3508 | ||
3509 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 3510 | #endif |
1da177e4 LT |
3511 | |
3512 | void __init free_area_init(unsigned long *zones_size) | |
3513 | { | |
93b7504e | 3514 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3515 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3516 | } | |
1da177e4 | 3517 | |
1da177e4 LT |
3518 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3519 | unsigned long action, void *hcpu) | |
3520 | { | |
3521 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 3522 | |
8bb78442 | 3523 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
1da177e4 LT |
3524 | local_irq_disable(); |
3525 | __drain_pages(cpu); | |
f8891e5e | 3526 | vm_events_fold_cpu(cpu); |
1da177e4 | 3527 | local_irq_enable(); |
2244b95a | 3528 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3529 | } |
3530 | return NOTIFY_OK; | |
3531 | } | |
1da177e4 LT |
3532 | |
3533 | void __init page_alloc_init(void) | |
3534 | { | |
3535 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3536 | } | |
3537 | ||
cb45b0e9 HA |
3538 | /* |
3539 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3540 | * or min_free_kbytes changes. | |
3541 | */ | |
3542 | static void calculate_totalreserve_pages(void) | |
3543 | { | |
3544 | struct pglist_data *pgdat; | |
3545 | unsigned long reserve_pages = 0; | |
2f6726e5 | 3546 | enum zone_type i, j; |
cb45b0e9 HA |
3547 | |
3548 | for_each_online_pgdat(pgdat) { | |
3549 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3550 | struct zone *zone = pgdat->node_zones + i; | |
3551 | unsigned long max = 0; | |
3552 | ||
3553 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3554 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3555 | if (zone->lowmem_reserve[j] > max) | |
3556 | max = zone->lowmem_reserve[j]; | |
3557 | } | |
3558 | ||
3559 | /* we treat pages_high as reserved pages. */ | |
3560 | max += zone->pages_high; | |
3561 | ||
3562 | if (max > zone->present_pages) | |
3563 | max = zone->present_pages; | |
3564 | reserve_pages += max; | |
3565 | } | |
3566 | } | |
3567 | totalreserve_pages = reserve_pages; | |
3568 | } | |
3569 | ||
1da177e4 LT |
3570 | /* |
3571 | * setup_per_zone_lowmem_reserve - called whenever | |
3572 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3573 | * has a correct pages reserved value, so an adequate number of | |
3574 | * pages are left in the zone after a successful __alloc_pages(). | |
3575 | */ | |
3576 | static void setup_per_zone_lowmem_reserve(void) | |
3577 | { | |
3578 | struct pglist_data *pgdat; | |
2f6726e5 | 3579 | enum zone_type j, idx; |
1da177e4 | 3580 | |
ec936fc5 | 3581 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
3582 | for (j = 0; j < MAX_NR_ZONES; j++) { |
3583 | struct zone *zone = pgdat->node_zones + j; | |
3584 | unsigned long present_pages = zone->present_pages; | |
3585 | ||
3586 | zone->lowmem_reserve[j] = 0; | |
3587 | ||
2f6726e5 CL |
3588 | idx = j; |
3589 | while (idx) { | |
1da177e4 LT |
3590 | struct zone *lower_zone; |
3591 | ||
2f6726e5 CL |
3592 | idx--; |
3593 | ||
1da177e4 LT |
3594 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
3595 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
3596 | ||
3597 | lower_zone = pgdat->node_zones + idx; | |
3598 | lower_zone->lowmem_reserve[j] = present_pages / | |
3599 | sysctl_lowmem_reserve_ratio[idx]; | |
3600 | present_pages += lower_zone->present_pages; | |
3601 | } | |
3602 | } | |
3603 | } | |
cb45b0e9 HA |
3604 | |
3605 | /* update totalreserve_pages */ | |
3606 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3607 | } |
3608 | ||
88ca3b94 RD |
3609 | /** |
3610 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
3611 | * | |
3612 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
3613 | * with respect to min_free_kbytes. | |
1da177e4 | 3614 | */ |
3947be19 | 3615 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
3616 | { |
3617 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
3618 | unsigned long lowmem_pages = 0; | |
3619 | struct zone *zone; | |
3620 | unsigned long flags; | |
3621 | ||
3622 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
3623 | for_each_zone(zone) { | |
3624 | if (!is_highmem(zone)) | |
3625 | lowmem_pages += zone->present_pages; | |
3626 | } | |
3627 | ||
3628 | for_each_zone(zone) { | |
ac924c60 AM |
3629 | u64 tmp; |
3630 | ||
1da177e4 | 3631 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
3632 | tmp = (u64)pages_min * zone->present_pages; |
3633 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
3634 | if (is_highmem(zone)) { |
3635 | /* | |
669ed175 NP |
3636 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
3637 | * need highmem pages, so cap pages_min to a small | |
3638 | * value here. | |
3639 | * | |
3640 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
3641 | * deltas controls asynch page reclaim, and so should | |
3642 | * not be capped for highmem. | |
1da177e4 LT |
3643 | */ |
3644 | int min_pages; | |
3645 | ||
3646 | min_pages = zone->present_pages / 1024; | |
3647 | if (min_pages < SWAP_CLUSTER_MAX) | |
3648 | min_pages = SWAP_CLUSTER_MAX; | |
3649 | if (min_pages > 128) | |
3650 | min_pages = 128; | |
3651 | zone->pages_min = min_pages; | |
3652 | } else { | |
669ed175 NP |
3653 | /* |
3654 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
3655 | * proportionate to the zone's size. |
3656 | */ | |
669ed175 | 3657 | zone->pages_min = tmp; |
1da177e4 LT |
3658 | } |
3659 | ||
ac924c60 AM |
3660 | zone->pages_low = zone->pages_min + (tmp >> 2); |
3661 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
3662 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
3663 | } | |
cb45b0e9 HA |
3664 | |
3665 | /* update totalreserve_pages */ | |
3666 | calculate_totalreserve_pages(); | |
1da177e4 LT |
3667 | } |
3668 | ||
3669 | /* | |
3670 | * Initialise min_free_kbytes. | |
3671 | * | |
3672 | * For small machines we want it small (128k min). For large machines | |
3673 | * we want it large (64MB max). But it is not linear, because network | |
3674 | * bandwidth does not increase linearly with machine size. We use | |
3675 | * | |
3676 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
3677 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
3678 | * | |
3679 | * which yields | |
3680 | * | |
3681 | * 16MB: 512k | |
3682 | * 32MB: 724k | |
3683 | * 64MB: 1024k | |
3684 | * 128MB: 1448k | |
3685 | * 256MB: 2048k | |
3686 | * 512MB: 2896k | |
3687 | * 1024MB: 4096k | |
3688 | * 2048MB: 5792k | |
3689 | * 4096MB: 8192k | |
3690 | * 8192MB: 11584k | |
3691 | * 16384MB: 16384k | |
3692 | */ | |
3693 | static int __init init_per_zone_pages_min(void) | |
3694 | { | |
3695 | unsigned long lowmem_kbytes; | |
3696 | ||
3697 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
3698 | ||
3699 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
3700 | if (min_free_kbytes < 128) | |
3701 | min_free_kbytes = 128; | |
3702 | if (min_free_kbytes > 65536) | |
3703 | min_free_kbytes = 65536; | |
3704 | setup_per_zone_pages_min(); | |
3705 | setup_per_zone_lowmem_reserve(); | |
3706 | return 0; | |
3707 | } | |
3708 | module_init(init_per_zone_pages_min) | |
3709 | ||
3710 | /* | |
3711 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
3712 | * that we can call two helper functions whenever min_free_kbytes | |
3713 | * changes. | |
3714 | */ | |
3715 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
3716 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3717 | { | |
3718 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
3719 | if (write) |
3720 | setup_per_zone_pages_min(); | |
1da177e4 LT |
3721 | return 0; |
3722 | } | |
3723 | ||
9614634f CL |
3724 | #ifdef CONFIG_NUMA |
3725 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
3726 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3727 | { | |
3728 | struct zone *zone; | |
3729 | int rc; | |
3730 | ||
3731 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3732 | if (rc) | |
3733 | return rc; | |
3734 | ||
3735 | for_each_zone(zone) | |
8417bba4 | 3736 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
3737 | sysctl_min_unmapped_ratio) / 100; |
3738 | return 0; | |
3739 | } | |
0ff38490 CL |
3740 | |
3741 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
3742 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3743 | { | |
3744 | struct zone *zone; | |
3745 | int rc; | |
3746 | ||
3747 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3748 | if (rc) | |
3749 | return rc; | |
3750 | ||
3751 | for_each_zone(zone) | |
3752 | zone->min_slab_pages = (zone->present_pages * | |
3753 | sysctl_min_slab_ratio) / 100; | |
3754 | return 0; | |
3755 | } | |
9614634f CL |
3756 | #endif |
3757 | ||
1da177e4 LT |
3758 | /* |
3759 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
3760 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
3761 | * whenever sysctl_lowmem_reserve_ratio changes. | |
3762 | * | |
3763 | * The reserve ratio obviously has absolutely no relation with the | |
3764 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
3765 | * if in function of the boot time zone sizes. | |
3766 | */ | |
3767 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
3768 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3769 | { | |
3770 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3771 | setup_per_zone_lowmem_reserve(); | |
3772 | return 0; | |
3773 | } | |
3774 | ||
8ad4b1fb RS |
3775 | /* |
3776 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
3777 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
3778 | * can have before it gets flushed back to buddy allocator. | |
3779 | */ | |
3780 | ||
3781 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
3782 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3783 | { | |
3784 | struct zone *zone; | |
3785 | unsigned int cpu; | |
3786 | int ret; | |
3787 | ||
3788 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3789 | if (!write || (ret == -EINVAL)) | |
3790 | return ret; | |
3791 | for_each_zone(zone) { | |
3792 | for_each_online_cpu(cpu) { | |
3793 | unsigned long high; | |
3794 | high = zone->present_pages / percpu_pagelist_fraction; | |
3795 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
3796 | } | |
3797 | } | |
3798 | return 0; | |
3799 | } | |
3800 | ||
f034b5d4 | 3801 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
3802 | |
3803 | #ifdef CONFIG_NUMA | |
3804 | static int __init set_hashdist(char *str) | |
3805 | { | |
3806 | if (!str) | |
3807 | return 0; | |
3808 | hashdist = simple_strtoul(str, &str, 0); | |
3809 | return 1; | |
3810 | } | |
3811 | __setup("hashdist=", set_hashdist); | |
3812 | #endif | |
3813 | ||
3814 | /* | |
3815 | * allocate a large system hash table from bootmem | |
3816 | * - it is assumed that the hash table must contain an exact power-of-2 | |
3817 | * quantity of entries | |
3818 | * - limit is the number of hash buckets, not the total allocation size | |
3819 | */ | |
3820 | void *__init alloc_large_system_hash(const char *tablename, | |
3821 | unsigned long bucketsize, | |
3822 | unsigned long numentries, | |
3823 | int scale, | |
3824 | int flags, | |
3825 | unsigned int *_hash_shift, | |
3826 | unsigned int *_hash_mask, | |
3827 | unsigned long limit) | |
3828 | { | |
3829 | unsigned long long max = limit; | |
3830 | unsigned long log2qty, size; | |
3831 | void *table = NULL; | |
3832 | ||
3833 | /* allow the kernel cmdline to have a say */ | |
3834 | if (!numentries) { | |
3835 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 3836 | numentries = nr_kernel_pages; |
1da177e4 LT |
3837 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
3838 | numentries >>= 20 - PAGE_SHIFT; | |
3839 | numentries <<= 20 - PAGE_SHIFT; | |
3840 | ||
3841 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
3842 | if (scale > PAGE_SHIFT) | |
3843 | numentries >>= (scale - PAGE_SHIFT); | |
3844 | else | |
3845 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
3846 | |
3847 | /* Make sure we've got at least a 0-order allocation.. */ | |
3848 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
3849 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 3850 | } |
6e692ed3 | 3851 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
3852 | |
3853 | /* limit allocation size to 1/16 total memory by default */ | |
3854 | if (max == 0) { | |
3855 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
3856 | do_div(max, bucketsize); | |
3857 | } | |
3858 | ||
3859 | if (numentries > max) | |
3860 | numentries = max; | |
3861 | ||
f0d1b0b3 | 3862 | log2qty = ilog2(numentries); |
1da177e4 LT |
3863 | |
3864 | do { | |
3865 | size = bucketsize << log2qty; | |
3866 | if (flags & HASH_EARLY) | |
3867 | table = alloc_bootmem(size); | |
3868 | else if (hashdist) | |
3869 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
3870 | else { | |
3871 | unsigned long order; | |
3872 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
3873 | ; | |
3874 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
1037b83b ED |
3875 | /* |
3876 | * If bucketsize is not a power-of-two, we may free | |
3877 | * some pages at the end of hash table. | |
3878 | */ | |
3879 | if (table) { | |
3880 | unsigned long alloc_end = (unsigned long)table + | |
3881 | (PAGE_SIZE << order); | |
3882 | unsigned long used = (unsigned long)table + | |
3883 | PAGE_ALIGN(size); | |
3884 | split_page(virt_to_page(table), order); | |
3885 | while (used < alloc_end) { | |
3886 | free_page(used); | |
3887 | used += PAGE_SIZE; | |
3888 | } | |
3889 | } | |
1da177e4 LT |
3890 | } |
3891 | } while (!table && size > PAGE_SIZE && --log2qty); | |
3892 | ||
3893 | if (!table) | |
3894 | panic("Failed to allocate %s hash table\n", tablename); | |
3895 | ||
b49ad484 | 3896 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
3897 | tablename, |
3898 | (1U << log2qty), | |
f0d1b0b3 | 3899 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
3900 | size); |
3901 | ||
3902 | if (_hash_shift) | |
3903 | *_hash_shift = log2qty; | |
3904 | if (_hash_mask) | |
3905 | *_hash_mask = (1 << log2qty) - 1; | |
3906 | ||
3907 | return table; | |
3908 | } | |
a117e66e KH |
3909 | |
3910 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
3911 | struct page *pfn_to_page(unsigned long pfn) |
3912 | { | |
67de6482 | 3913 | return __pfn_to_page(pfn); |
a117e66e KH |
3914 | } |
3915 | unsigned long page_to_pfn(struct page *page) | |
3916 | { | |
67de6482 | 3917 | return __page_to_pfn(page); |
a117e66e | 3918 | } |
a117e66e KH |
3919 | EXPORT_SYMBOL(pfn_to_page); |
3920 | EXPORT_SYMBOL(page_to_pfn); | |
3921 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 3922 | |
6220ec78 | 3923 |