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> | |
1da177e4 LT |
42 | |
43 | #include <asm/tlbflush.h> | |
ac924c60 | 44 | #include <asm/div64.h> |
1da177e4 LT |
45 | #include "internal.h" |
46 | ||
47 | /* | |
48 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
49 | * initializer cleaner | |
50 | */ | |
c3d8c141 | 51 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 52 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 53 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 54 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b | 55 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 56 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 57 | long nr_swap_pages; |
8ad4b1fb | 58 | int percpu_pagelist_fraction; |
1da177e4 | 59 | |
d98c7a09 | 60 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 61 | |
1da177e4 LT |
62 | /* |
63 | * results with 256, 32 in the lowmem_reserve sysctl: | |
64 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
65 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
66 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
67 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
68 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
69 | * |
70 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
71 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 72 | */ |
2f1b6248 CL |
73 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
74 | 256, | |
fb0e7942 | 75 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 76 | 256, |
fb0e7942 | 77 | #endif |
e53ef38d | 78 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 79 | 32 |
e53ef38d | 80 | #endif |
2f1b6248 | 81 | }; |
1da177e4 LT |
82 | |
83 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
84 | |
85 | /* | |
86 | * Used by page_zone() to look up the address of the struct zone whose | |
87 | * id is encoded in the upper bits of page->flags | |
88 | */ | |
c3d8c141 | 89 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
90 | EXPORT_SYMBOL(zone_table); |
91 | ||
2f1b6248 CL |
92 | static char *zone_names[MAX_NR_ZONES] = { |
93 | "DMA", | |
fb0e7942 | 94 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 95 | "DMA32", |
fb0e7942 | 96 | #endif |
2f1b6248 | 97 | "Normal", |
e53ef38d | 98 | #ifdef CONFIG_HIGHMEM |
2f1b6248 | 99 | "HighMem" |
e53ef38d | 100 | #endif |
2f1b6248 CL |
101 | }; |
102 | ||
1da177e4 LT |
103 | int min_free_kbytes = 1024; |
104 | ||
86356ab1 YG |
105 | unsigned long __meminitdata nr_kernel_pages; |
106 | unsigned long __meminitdata nr_all_pages; | |
1da177e4 | 107 | |
c713216d MG |
108 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
109 | /* | |
110 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
111 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
112 | * Ranges passed to add_active_range() will be merged if possible | |
113 | * so the number of times add_active_range() can be called is | |
114 | * related to the number of nodes and the number of holes | |
115 | */ | |
116 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
117 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
118 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
119 | #else | |
120 | #if MAX_NUMNODES >= 32 | |
121 | /* If there can be many nodes, allow up to 50 holes per node */ | |
122 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
123 | #else | |
124 | /* By default, allow up to 256 distinct regions */ | |
125 | #define MAX_ACTIVE_REGIONS 256 | |
126 | #endif | |
127 | #endif | |
128 | ||
129 | struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS]; | |
130 | int __initdata nr_nodemap_entries; | |
131 | unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
132 | unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
133 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
134 | ||
13e7444b | 135 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 136 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 137 | { |
bdc8cb98 DH |
138 | int ret = 0; |
139 | unsigned seq; | |
140 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 141 | |
bdc8cb98 DH |
142 | do { |
143 | seq = zone_span_seqbegin(zone); | |
144 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
145 | ret = 1; | |
146 | else if (pfn < zone->zone_start_pfn) | |
147 | ret = 1; | |
148 | } while (zone_span_seqretry(zone, seq)); | |
149 | ||
150 | return ret; | |
c6a57e19 DH |
151 | } |
152 | ||
153 | static int page_is_consistent(struct zone *zone, struct page *page) | |
154 | { | |
1da177e4 LT |
155 | #ifdef CONFIG_HOLES_IN_ZONE |
156 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 157 | return 0; |
1da177e4 LT |
158 | #endif |
159 | if (zone != page_zone(page)) | |
c6a57e19 DH |
160 | return 0; |
161 | ||
162 | return 1; | |
163 | } | |
164 | /* | |
165 | * Temporary debugging check for pages not lying within a given zone. | |
166 | */ | |
167 | static int bad_range(struct zone *zone, struct page *page) | |
168 | { | |
169 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 170 | return 1; |
c6a57e19 DH |
171 | if (!page_is_consistent(zone, page)) |
172 | return 1; | |
173 | ||
1da177e4 LT |
174 | return 0; |
175 | } | |
13e7444b NP |
176 | #else |
177 | static inline int bad_range(struct zone *zone, struct page *page) | |
178 | { | |
179 | return 0; | |
180 | } | |
181 | #endif | |
182 | ||
224abf92 | 183 | static void bad_page(struct page *page) |
1da177e4 | 184 | { |
224abf92 | 185 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
186 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
187 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
188 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
189 | current->comm, page, (int)(2*sizeof(unsigned long)), |
190 | (unsigned long)page->flags, page->mapping, | |
191 | page_mapcount(page), page_count(page)); | |
1da177e4 | 192 | dump_stack(); |
334795ec HD |
193 | page->flags &= ~(1 << PG_lru | |
194 | 1 << PG_private | | |
1da177e4 | 195 | 1 << PG_locked | |
1da177e4 LT |
196 | 1 << PG_active | |
197 | 1 << PG_dirty | | |
334795ec HD |
198 | 1 << PG_reclaim | |
199 | 1 << PG_slab | | |
1da177e4 | 200 | 1 << PG_swapcache | |
676165a8 NP |
201 | 1 << PG_writeback | |
202 | 1 << PG_buddy ); | |
1da177e4 LT |
203 | set_page_count(page, 0); |
204 | reset_page_mapcount(page); | |
205 | page->mapping = NULL; | |
9f158333 | 206 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
207 | } |
208 | ||
1da177e4 LT |
209 | /* |
210 | * Higher-order pages are called "compound pages". They are structured thusly: | |
211 | * | |
212 | * The first PAGE_SIZE page is called the "head page". | |
213 | * | |
214 | * The remaining PAGE_SIZE pages are called "tail pages". | |
215 | * | |
216 | * All pages have PG_compound set. All pages have their ->private pointing at | |
217 | * the head page (even the head page has this). | |
218 | * | |
41d78ba5 HD |
219 | * The first tail page's ->lru.next holds the address of the compound page's |
220 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
221 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 222 | */ |
d98c7a09 HD |
223 | |
224 | static void free_compound_page(struct page *page) | |
225 | { | |
226 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
227 | } | |
228 | ||
1da177e4 LT |
229 | static void prep_compound_page(struct page *page, unsigned long order) |
230 | { | |
231 | int i; | |
232 | int nr_pages = 1 << order; | |
233 | ||
d98c7a09 | 234 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 235 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
236 | for (i = 0; i < nr_pages; i++) { |
237 | struct page *p = page + i; | |
238 | ||
5e9dace8 | 239 | __SetPageCompound(p); |
4c21e2f2 | 240 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
241 | } |
242 | } | |
243 | ||
244 | static void destroy_compound_page(struct page *page, unsigned long order) | |
245 | { | |
246 | int i; | |
247 | int nr_pages = 1 << order; | |
248 | ||
41d78ba5 | 249 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 250 | bad_page(page); |
1da177e4 LT |
251 | |
252 | for (i = 0; i < nr_pages; i++) { | |
253 | struct page *p = page + i; | |
254 | ||
224abf92 NP |
255 | if (unlikely(!PageCompound(p) | |
256 | (page_private(p) != (unsigned long)page))) | |
257 | bad_page(page); | |
5e9dace8 | 258 | __ClearPageCompound(p); |
1da177e4 LT |
259 | } |
260 | } | |
1da177e4 | 261 | |
17cf4406 NP |
262 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
263 | { | |
264 | int i; | |
265 | ||
725d704e | 266 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); |
6626c5d5 AM |
267 | /* |
268 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
269 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
270 | */ | |
725d704e | 271 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
272 | for (i = 0; i < (1 << order); i++) |
273 | clear_highpage(page + i); | |
274 | } | |
275 | ||
1da177e4 LT |
276 | /* |
277 | * function for dealing with page's order in buddy system. | |
278 | * zone->lock is already acquired when we use these. | |
279 | * So, we don't need atomic page->flags operations here. | |
280 | */ | |
6aa3001b AM |
281 | static inline unsigned long page_order(struct page *page) |
282 | { | |
4c21e2f2 | 283 | return page_private(page); |
1da177e4 LT |
284 | } |
285 | ||
6aa3001b AM |
286 | static inline void set_page_order(struct page *page, int order) |
287 | { | |
4c21e2f2 | 288 | set_page_private(page, order); |
676165a8 | 289 | __SetPageBuddy(page); |
1da177e4 LT |
290 | } |
291 | ||
292 | static inline void rmv_page_order(struct page *page) | |
293 | { | |
676165a8 | 294 | __ClearPageBuddy(page); |
4c21e2f2 | 295 | set_page_private(page, 0); |
1da177e4 LT |
296 | } |
297 | ||
298 | /* | |
299 | * Locate the struct page for both the matching buddy in our | |
300 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
301 | * | |
302 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
303 | * the following equation: | |
304 | * B2 = B1 ^ (1 << O) | |
305 | * For example, if the starting buddy (buddy2) is #8 its order | |
306 | * 1 buddy is #10: | |
307 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
308 | * | |
309 | * 2) Any buddy B will have an order O+1 parent P which | |
310 | * satisfies the following equation: | |
311 | * P = B & ~(1 << O) | |
312 | * | |
d6e05edc | 313 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
314 | */ |
315 | static inline struct page * | |
316 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
317 | { | |
318 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
319 | ||
320 | return page + (buddy_idx - page_idx); | |
321 | } | |
322 | ||
323 | static inline unsigned long | |
324 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
325 | { | |
326 | return (page_idx & ~(1 << order)); | |
327 | } | |
328 | ||
329 | /* | |
330 | * This function checks whether a page is free && is the buddy | |
331 | * we can do coalesce a page and its buddy if | |
13e7444b | 332 | * (a) the buddy is not in a hole && |
676165a8 | 333 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
334 | * (c) a page and its buddy have the same order && |
335 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
336 | * |
337 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
338 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 339 | * |
676165a8 | 340 | * For recording page's order, we use page_private(page). |
1da177e4 | 341 | */ |
cb2b95e1 AW |
342 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
343 | int order) | |
1da177e4 | 344 | { |
13e7444b | 345 | #ifdef CONFIG_HOLES_IN_ZONE |
cb2b95e1 | 346 | if (!pfn_valid(page_to_pfn(buddy))) |
13e7444b NP |
347 | return 0; |
348 | #endif | |
349 | ||
cb2b95e1 AW |
350 | if (page_zone_id(page) != page_zone_id(buddy)) |
351 | return 0; | |
352 | ||
353 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
354 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 355 | return 1; |
676165a8 | 356 | } |
6aa3001b | 357 | return 0; |
1da177e4 LT |
358 | } |
359 | ||
360 | /* | |
361 | * Freeing function for a buddy system allocator. | |
362 | * | |
363 | * The concept of a buddy system is to maintain direct-mapped table | |
364 | * (containing bit values) for memory blocks of various "orders". | |
365 | * The bottom level table contains the map for the smallest allocatable | |
366 | * units of memory (here, pages), and each level above it describes | |
367 | * pairs of units from the levels below, hence, "buddies". | |
368 | * At a high level, all that happens here is marking the table entry | |
369 | * at the bottom level available, and propagating the changes upward | |
370 | * as necessary, plus some accounting needed to play nicely with other | |
371 | * parts of the VM system. | |
372 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 373 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 374 | * order is recorded in page_private(page) field. |
1da177e4 LT |
375 | * So when we are allocating or freeing one, we can derive the state of the |
376 | * other. That is, if we allocate a small block, and both were | |
377 | * free, the remainder of the region must be split into blocks. | |
378 | * If a block is freed, and its buddy is also free, then this | |
379 | * triggers coalescing into a block of larger size. | |
380 | * | |
381 | * -- wli | |
382 | */ | |
383 | ||
48db57f8 | 384 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
385 | struct zone *zone, unsigned int order) |
386 | { | |
387 | unsigned long page_idx; | |
388 | int order_size = 1 << order; | |
389 | ||
224abf92 | 390 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
391 | destroy_compound_page(page, order); |
392 | ||
393 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
394 | ||
725d704e NP |
395 | VM_BUG_ON(page_idx & (order_size - 1)); |
396 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 LT |
397 | |
398 | zone->free_pages += order_size; | |
399 | while (order < MAX_ORDER-1) { | |
400 | unsigned long combined_idx; | |
401 | struct free_area *area; | |
402 | struct page *buddy; | |
403 | ||
1da177e4 | 404 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 405 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 406 | break; /* Move the buddy up one level. */ |
13e7444b | 407 | |
1da177e4 LT |
408 | list_del(&buddy->lru); |
409 | area = zone->free_area + order; | |
410 | area->nr_free--; | |
411 | rmv_page_order(buddy); | |
13e7444b | 412 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
413 | page = page + (combined_idx - page_idx); |
414 | page_idx = combined_idx; | |
415 | order++; | |
416 | } | |
417 | set_page_order(page, order); | |
418 | list_add(&page->lru, &zone->free_area[order].free_list); | |
419 | zone->free_area[order].nr_free++; | |
420 | } | |
421 | ||
224abf92 | 422 | static inline int free_pages_check(struct page *page) |
1da177e4 | 423 | { |
92be2e33 NP |
424 | if (unlikely(page_mapcount(page) | |
425 | (page->mapping != NULL) | | |
426 | (page_count(page) != 0) | | |
1da177e4 LT |
427 | (page->flags & ( |
428 | 1 << PG_lru | | |
429 | 1 << PG_private | | |
430 | 1 << PG_locked | | |
431 | 1 << PG_active | | |
432 | 1 << PG_reclaim | | |
433 | 1 << PG_slab | | |
434 | 1 << PG_swapcache | | |
b5810039 | 435 | 1 << PG_writeback | |
676165a8 NP |
436 | 1 << PG_reserved | |
437 | 1 << PG_buddy )))) | |
224abf92 | 438 | bad_page(page); |
1da177e4 | 439 | if (PageDirty(page)) |
242e5468 | 440 | __ClearPageDirty(page); |
689bcebf HD |
441 | /* |
442 | * For now, we report if PG_reserved was found set, but do not | |
443 | * clear it, and do not free the page. But we shall soon need | |
444 | * to do more, for when the ZERO_PAGE count wraps negative. | |
445 | */ | |
446 | return PageReserved(page); | |
1da177e4 LT |
447 | } |
448 | ||
449 | /* | |
450 | * Frees a list of pages. | |
451 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 452 | * count is the number of pages to free. |
1da177e4 LT |
453 | * |
454 | * If the zone was previously in an "all pages pinned" state then look to | |
455 | * see if this freeing clears that state. | |
456 | * | |
457 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
458 | * pinned" detection logic. | |
459 | */ | |
48db57f8 NP |
460 | static void free_pages_bulk(struct zone *zone, int count, |
461 | struct list_head *list, int order) | |
1da177e4 | 462 | { |
c54ad30c | 463 | spin_lock(&zone->lock); |
1da177e4 LT |
464 | zone->all_unreclaimable = 0; |
465 | zone->pages_scanned = 0; | |
48db57f8 NP |
466 | while (count--) { |
467 | struct page *page; | |
468 | ||
725d704e | 469 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 470 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 471 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 472 | list_del(&page->lru); |
48db57f8 | 473 | __free_one_page(page, zone, order); |
1da177e4 | 474 | } |
c54ad30c | 475 | spin_unlock(&zone->lock); |
1da177e4 LT |
476 | } |
477 | ||
48db57f8 | 478 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 479 | { |
006d22d9 CL |
480 | spin_lock(&zone->lock); |
481 | zone->all_unreclaimable = 0; | |
482 | zone->pages_scanned = 0; | |
483 | __free_one_page(page, zone ,order); | |
484 | spin_unlock(&zone->lock); | |
48db57f8 NP |
485 | } |
486 | ||
487 | static void __free_pages_ok(struct page *page, unsigned int order) | |
488 | { | |
489 | unsigned long flags; | |
1da177e4 | 490 | int i; |
689bcebf | 491 | int reserved = 0; |
1da177e4 LT |
492 | |
493 | arch_free_page(page, order); | |
de5097c2 | 494 | if (!PageHighMem(page)) |
f9b8404c IM |
495 | debug_check_no_locks_freed(page_address(page), |
496 | PAGE_SIZE<<order); | |
1da177e4 | 497 | |
1da177e4 | 498 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 499 | reserved += free_pages_check(page + i); |
689bcebf HD |
500 | if (reserved) |
501 | return; | |
502 | ||
48db57f8 | 503 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 504 | local_irq_save(flags); |
f8891e5e | 505 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 506 | free_one_page(page_zone(page), page, order); |
c54ad30c | 507 | local_irq_restore(flags); |
1da177e4 LT |
508 | } |
509 | ||
a226f6c8 DH |
510 | /* |
511 | * permit the bootmem allocator to evade page validation on high-order frees | |
512 | */ | |
513 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
514 | { | |
515 | if (order == 0) { | |
516 | __ClearPageReserved(page); | |
517 | set_page_count(page, 0); | |
7835e98b | 518 | set_page_refcounted(page); |
545b1ea9 | 519 | __free_page(page); |
a226f6c8 | 520 | } else { |
a226f6c8 DH |
521 | int loop; |
522 | ||
545b1ea9 | 523 | prefetchw(page); |
a226f6c8 DH |
524 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
525 | struct page *p = &page[loop]; | |
526 | ||
545b1ea9 NP |
527 | if (loop + 1 < BITS_PER_LONG) |
528 | prefetchw(p + 1); | |
a226f6c8 DH |
529 | __ClearPageReserved(p); |
530 | set_page_count(p, 0); | |
531 | } | |
532 | ||
7835e98b | 533 | set_page_refcounted(page); |
545b1ea9 | 534 | __free_pages(page, order); |
a226f6c8 DH |
535 | } |
536 | } | |
537 | ||
1da177e4 LT |
538 | |
539 | /* | |
540 | * The order of subdivision here is critical for the IO subsystem. | |
541 | * Please do not alter this order without good reasons and regression | |
542 | * testing. Specifically, as large blocks of memory are subdivided, | |
543 | * the order in which smaller blocks are delivered depends on the order | |
544 | * they're subdivided in this function. This is the primary factor | |
545 | * influencing the order in which pages are delivered to the IO | |
546 | * subsystem according to empirical testing, and this is also justified | |
547 | * by considering the behavior of a buddy system containing a single | |
548 | * large block of memory acted on by a series of small allocations. | |
549 | * This behavior is a critical factor in sglist merging's success. | |
550 | * | |
551 | * -- wli | |
552 | */ | |
085cc7d5 | 553 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
554 | int low, int high, struct free_area *area) |
555 | { | |
556 | unsigned long size = 1 << high; | |
557 | ||
558 | while (high > low) { | |
559 | area--; | |
560 | high--; | |
561 | size >>= 1; | |
725d704e | 562 | VM_BUG_ON(bad_range(zone, &page[size])); |
1da177e4 LT |
563 | list_add(&page[size].lru, &area->free_list); |
564 | area->nr_free++; | |
565 | set_page_order(&page[size], high); | |
566 | } | |
1da177e4 LT |
567 | } |
568 | ||
1da177e4 LT |
569 | /* |
570 | * This page is about to be returned from the page allocator | |
571 | */ | |
17cf4406 | 572 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 573 | { |
92be2e33 NP |
574 | if (unlikely(page_mapcount(page) | |
575 | (page->mapping != NULL) | | |
576 | (page_count(page) != 0) | | |
334795ec HD |
577 | (page->flags & ( |
578 | 1 << PG_lru | | |
1da177e4 LT |
579 | 1 << PG_private | |
580 | 1 << PG_locked | | |
1da177e4 LT |
581 | 1 << PG_active | |
582 | 1 << PG_dirty | | |
583 | 1 << PG_reclaim | | |
334795ec | 584 | 1 << PG_slab | |
1da177e4 | 585 | 1 << PG_swapcache | |
b5810039 | 586 | 1 << PG_writeback | |
676165a8 NP |
587 | 1 << PG_reserved | |
588 | 1 << PG_buddy )))) | |
224abf92 | 589 | bad_page(page); |
1da177e4 | 590 | |
689bcebf HD |
591 | /* |
592 | * For now, we report if PG_reserved was found set, but do not | |
593 | * clear it, and do not allocate the page: as a safety net. | |
594 | */ | |
595 | if (PageReserved(page)) | |
596 | return 1; | |
597 | ||
1da177e4 LT |
598 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
599 | 1 << PG_referenced | 1 << PG_arch_1 | | |
600 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 601 | set_page_private(page, 0); |
7835e98b | 602 | set_page_refcounted(page); |
1da177e4 | 603 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
604 | |
605 | if (gfp_flags & __GFP_ZERO) | |
606 | prep_zero_page(page, order, gfp_flags); | |
607 | ||
608 | if (order && (gfp_flags & __GFP_COMP)) | |
609 | prep_compound_page(page, order); | |
610 | ||
689bcebf | 611 | return 0; |
1da177e4 LT |
612 | } |
613 | ||
614 | /* | |
615 | * Do the hard work of removing an element from the buddy allocator. | |
616 | * Call me with the zone->lock already held. | |
617 | */ | |
618 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
619 | { | |
620 | struct free_area * area; | |
621 | unsigned int current_order; | |
622 | struct page *page; | |
623 | ||
624 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
625 | area = zone->free_area + current_order; | |
626 | if (list_empty(&area->free_list)) | |
627 | continue; | |
628 | ||
629 | page = list_entry(area->free_list.next, struct page, lru); | |
630 | list_del(&page->lru); | |
631 | rmv_page_order(page); | |
632 | area->nr_free--; | |
633 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
634 | expand(zone, page, order, current_order, area); |
635 | return page; | |
1da177e4 LT |
636 | } |
637 | ||
638 | return NULL; | |
639 | } | |
640 | ||
641 | /* | |
642 | * Obtain a specified number of elements from the buddy allocator, all under | |
643 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
644 | * Returns the number of new pages which were placed at *list. | |
645 | */ | |
646 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
647 | unsigned long count, struct list_head *list) | |
648 | { | |
1da177e4 | 649 | int i; |
1da177e4 | 650 | |
c54ad30c | 651 | spin_lock(&zone->lock); |
1da177e4 | 652 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
653 | struct page *page = __rmqueue(zone, order); |
654 | if (unlikely(page == NULL)) | |
1da177e4 | 655 | break; |
1da177e4 LT |
656 | list_add_tail(&page->lru, list); |
657 | } | |
c54ad30c | 658 | spin_unlock(&zone->lock); |
085cc7d5 | 659 | return i; |
1da177e4 LT |
660 | } |
661 | ||
4ae7c039 | 662 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
663 | /* |
664 | * Called from the slab reaper to drain pagesets on a particular node that | |
39bbcb8f | 665 | * belongs to the currently executing processor. |
879336c3 CL |
666 | * Note that this function must be called with the thread pinned to |
667 | * a single processor. | |
8fce4d8e CL |
668 | */ |
669 | void drain_node_pages(int nodeid) | |
4ae7c039 | 670 | { |
2f6726e5 CL |
671 | int i; |
672 | enum zone_type z; | |
4ae7c039 CL |
673 | unsigned long flags; |
674 | ||
8fce4d8e CL |
675 | for (z = 0; z < MAX_NR_ZONES; z++) { |
676 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
677 | struct per_cpu_pageset *pset; |
678 | ||
39bbcb8f CL |
679 | if (!populated_zone(zone)) |
680 | continue; | |
681 | ||
23316bc8 | 682 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
683 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
684 | struct per_cpu_pages *pcp; | |
685 | ||
686 | pcp = &pset->pcp[i]; | |
879336c3 CL |
687 | if (pcp->count) { |
688 | local_irq_save(flags); | |
689 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
690 | pcp->count = 0; | |
691 | local_irq_restore(flags); | |
692 | } | |
4ae7c039 CL |
693 | } |
694 | } | |
4ae7c039 CL |
695 | } |
696 | #endif | |
697 | ||
1da177e4 LT |
698 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
699 | static void __drain_pages(unsigned int cpu) | |
700 | { | |
c54ad30c | 701 | unsigned long flags; |
1da177e4 LT |
702 | struct zone *zone; |
703 | int i; | |
704 | ||
705 | for_each_zone(zone) { | |
706 | struct per_cpu_pageset *pset; | |
707 | ||
e7c8d5c9 | 708 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
709 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
710 | struct per_cpu_pages *pcp; | |
711 | ||
712 | pcp = &pset->pcp[i]; | |
c54ad30c | 713 | local_irq_save(flags); |
48db57f8 NP |
714 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
715 | pcp->count = 0; | |
c54ad30c | 716 | local_irq_restore(flags); |
1da177e4 LT |
717 | } |
718 | } | |
719 | } | |
720 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
721 | ||
722 | #ifdef CONFIG_PM | |
723 | ||
724 | void mark_free_pages(struct zone *zone) | |
725 | { | |
f623f0db RW |
726 | unsigned long pfn, max_zone_pfn; |
727 | unsigned long flags; | |
1da177e4 LT |
728 | int order; |
729 | struct list_head *curr; | |
730 | ||
731 | if (!zone->spanned_pages) | |
732 | return; | |
733 | ||
734 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
735 | |
736 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
737 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
738 | if (pfn_valid(pfn)) { | |
739 | struct page *page = pfn_to_page(pfn); | |
740 | ||
741 | if (!PageNosave(page)) | |
742 | ClearPageNosaveFree(page); | |
743 | } | |
1da177e4 LT |
744 | |
745 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
746 | list_for_each(curr, &zone->free_area[order].free_list) { | |
f623f0db | 747 | unsigned long i; |
1da177e4 | 748 | |
f623f0db RW |
749 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
750 | for (i = 0; i < (1UL << order); i++) | |
751 | SetPageNosaveFree(pfn_to_page(pfn + i)); | |
752 | } | |
1da177e4 | 753 | |
1da177e4 LT |
754 | spin_unlock_irqrestore(&zone->lock, flags); |
755 | } | |
756 | ||
757 | /* | |
758 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
759 | */ | |
760 | void drain_local_pages(void) | |
761 | { | |
762 | unsigned long flags; | |
763 | ||
764 | local_irq_save(flags); | |
765 | __drain_pages(smp_processor_id()); | |
766 | local_irq_restore(flags); | |
767 | } | |
768 | #endif /* CONFIG_PM */ | |
769 | ||
1da177e4 LT |
770 | /* |
771 | * Free a 0-order page | |
772 | */ | |
1da177e4 LT |
773 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
774 | { | |
775 | struct zone *zone = page_zone(page); | |
776 | struct per_cpu_pages *pcp; | |
777 | unsigned long flags; | |
778 | ||
779 | arch_free_page(page, 0); | |
780 | ||
1da177e4 LT |
781 | if (PageAnon(page)) |
782 | page->mapping = NULL; | |
224abf92 | 783 | if (free_pages_check(page)) |
689bcebf HD |
784 | return; |
785 | ||
689bcebf HD |
786 | kernel_map_pages(page, 1, 0); |
787 | ||
e7c8d5c9 | 788 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 789 | local_irq_save(flags); |
f8891e5e | 790 | __count_vm_event(PGFREE); |
1da177e4 LT |
791 | list_add(&page->lru, &pcp->list); |
792 | pcp->count++; | |
48db57f8 NP |
793 | if (pcp->count >= pcp->high) { |
794 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
795 | pcp->count -= pcp->batch; | |
796 | } | |
1da177e4 LT |
797 | local_irq_restore(flags); |
798 | put_cpu(); | |
799 | } | |
800 | ||
801 | void fastcall free_hot_page(struct page *page) | |
802 | { | |
803 | free_hot_cold_page(page, 0); | |
804 | } | |
805 | ||
806 | void fastcall free_cold_page(struct page *page) | |
807 | { | |
808 | free_hot_cold_page(page, 1); | |
809 | } | |
810 | ||
8dfcc9ba NP |
811 | /* |
812 | * split_page takes a non-compound higher-order page, and splits it into | |
813 | * n (1<<order) sub-pages: page[0..n] | |
814 | * Each sub-page must be freed individually. | |
815 | * | |
816 | * Note: this is probably too low level an operation for use in drivers. | |
817 | * Please consult with lkml before using this in your driver. | |
818 | */ | |
819 | void split_page(struct page *page, unsigned int order) | |
820 | { | |
821 | int i; | |
822 | ||
725d704e NP |
823 | VM_BUG_ON(PageCompound(page)); |
824 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
825 | for (i = 1; i < (1 << order); i++) |
826 | set_page_refcounted(page + i); | |
8dfcc9ba | 827 | } |
8dfcc9ba | 828 | |
1da177e4 LT |
829 | /* |
830 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
831 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
832 | * or two. | |
833 | */ | |
a74609fa NP |
834 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
835 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
836 | { |
837 | unsigned long flags; | |
689bcebf | 838 | struct page *page; |
1da177e4 | 839 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 840 | int cpu; |
1da177e4 | 841 | |
689bcebf | 842 | again: |
a74609fa | 843 | cpu = get_cpu(); |
48db57f8 | 844 | if (likely(order == 0)) { |
1da177e4 LT |
845 | struct per_cpu_pages *pcp; |
846 | ||
a74609fa | 847 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 848 | local_irq_save(flags); |
a74609fa | 849 | if (!pcp->count) { |
1da177e4 LT |
850 | pcp->count += rmqueue_bulk(zone, 0, |
851 | pcp->batch, &pcp->list); | |
a74609fa NP |
852 | if (unlikely(!pcp->count)) |
853 | goto failed; | |
1da177e4 | 854 | } |
a74609fa NP |
855 | page = list_entry(pcp->list.next, struct page, lru); |
856 | list_del(&page->lru); | |
857 | pcp->count--; | |
7fb1d9fc | 858 | } else { |
1da177e4 LT |
859 | spin_lock_irqsave(&zone->lock, flags); |
860 | page = __rmqueue(zone, order); | |
a74609fa NP |
861 | spin_unlock(&zone->lock); |
862 | if (!page) | |
863 | goto failed; | |
1da177e4 LT |
864 | } |
865 | ||
f8891e5e | 866 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 867 | zone_statistics(zonelist, zone); |
a74609fa NP |
868 | local_irq_restore(flags); |
869 | put_cpu(); | |
1da177e4 | 870 | |
725d704e | 871 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 872 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 873 | goto again; |
1da177e4 | 874 | return page; |
a74609fa NP |
875 | |
876 | failed: | |
877 | local_irq_restore(flags); | |
878 | put_cpu(); | |
879 | return NULL; | |
1da177e4 LT |
880 | } |
881 | ||
7fb1d9fc | 882 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
883 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
884 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
885 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
886 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
887 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
888 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 889 | |
1da177e4 LT |
890 | /* |
891 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
892 | * of the allocation. | |
893 | */ | |
894 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 895 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
896 | { |
897 | /* free_pages my go negative - that's OK */ | |
898 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
899 | int o; | |
900 | ||
7fb1d9fc | 901 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 902 | min -= min / 2; |
7fb1d9fc | 903 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
904 | min -= min / 4; |
905 | ||
906 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
907 | return 0; | |
908 | for (o = 0; o < order; o++) { | |
909 | /* At the next order, this order's pages become unavailable */ | |
910 | free_pages -= z->free_area[o].nr_free << o; | |
911 | ||
912 | /* Require fewer higher order pages to be free */ | |
913 | min >>= 1; | |
914 | ||
915 | if (free_pages <= min) | |
916 | return 0; | |
917 | } | |
918 | return 1; | |
919 | } | |
920 | ||
7fb1d9fc RS |
921 | /* |
922 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
923 | * a page. | |
924 | */ | |
925 | static struct page * | |
926 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
927 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 928 | { |
7fb1d9fc RS |
929 | struct zone **z = zonelist->zones; |
930 | struct page *page = NULL; | |
931 | int classzone_idx = zone_idx(*z); | |
1192d526 | 932 | struct zone *zone; |
7fb1d9fc RS |
933 | |
934 | /* | |
935 | * Go through the zonelist once, looking for a zone with enough free. | |
936 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
937 | */ | |
938 | do { | |
1192d526 | 939 | zone = *z; |
9b819d20 | 940 | if (unlikely((gfp_mask & __GFP_THISNODE) && |
1192d526 | 941 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) |
9b819d20 | 942 | break; |
7fb1d9fc | 943 | if ((alloc_flags & ALLOC_CPUSET) && |
1192d526 | 944 | !cpuset_zone_allowed(zone, gfp_mask)) |
7fb1d9fc RS |
945 | continue; |
946 | ||
947 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
948 | unsigned long mark; |
949 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 950 | mark = zone->pages_min; |
3148890b | 951 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 952 | mark = zone->pages_low; |
3148890b | 953 | else |
1192d526 CL |
954 | mark = zone->pages_high; |
955 | if (!zone_watermark_ok(zone , order, mark, | |
7fb1d9fc | 956 | classzone_idx, alloc_flags)) |
9eeff239 | 957 | if (!zone_reclaim_mode || |
1192d526 | 958 | !zone_reclaim(zone, gfp_mask, order)) |
9eeff239 | 959 | continue; |
7fb1d9fc RS |
960 | } |
961 | ||
1192d526 | 962 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
7fb1d9fc | 963 | if (page) { |
7fb1d9fc RS |
964 | break; |
965 | } | |
966 | } while (*(++z) != NULL); | |
967 | return page; | |
753ee728 MH |
968 | } |
969 | ||
1da177e4 LT |
970 | /* |
971 | * This is the 'heart' of the zoned buddy allocator. | |
972 | */ | |
973 | struct page * fastcall | |
dd0fc66f | 974 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
975 | struct zonelist *zonelist) |
976 | { | |
260b2367 | 977 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 978 | struct zone **z; |
1da177e4 LT |
979 | struct page *page; |
980 | struct reclaim_state reclaim_state; | |
981 | struct task_struct *p = current; | |
1da177e4 | 982 | int do_retry; |
7fb1d9fc | 983 | int alloc_flags; |
1da177e4 LT |
984 | int did_some_progress; |
985 | ||
986 | might_sleep_if(wait); | |
987 | ||
6b1de916 | 988 | restart: |
7fb1d9fc | 989 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 990 | |
7fb1d9fc | 991 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
992 | /* Should this ever happen?? */ |
993 | return NULL; | |
994 | } | |
6b1de916 | 995 | |
7fb1d9fc | 996 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 997 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
998 | if (page) |
999 | goto got_pg; | |
1da177e4 | 1000 | |
6b1de916 | 1001 | do { |
43b0bc00 | 1002 | wakeup_kswapd(*z, order); |
6b1de916 | 1003 | } while (*(++z)); |
1da177e4 | 1004 | |
9bf2229f | 1005 | /* |
7fb1d9fc RS |
1006 | * OK, we're below the kswapd watermark and have kicked background |
1007 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1008 | * to how we want to proceed. | |
1009 | * | |
1010 | * The caller may dip into page reserves a bit more if the caller | |
1011 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1012 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1013 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1014 | */ |
3148890b | 1015 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1016 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1017 | alloc_flags |= ALLOC_HARDER; | |
1018 | if (gfp_mask & __GFP_HIGH) | |
1019 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1020 | if (wait) |
1021 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1022 | |
1023 | /* | |
1024 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1025 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1026 | * |
1027 | * This is the last chance, in general, before the goto nopage. | |
1028 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1029 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1030 | */ |
7fb1d9fc RS |
1031 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1032 | if (page) | |
1033 | goto got_pg; | |
1da177e4 LT |
1034 | |
1035 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
1036 | |
1037 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
1038 | && !in_interrupt()) { | |
1039 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1040 | nofail_alloc: |
b84a35be | 1041 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1042 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1043 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1044 | if (page) |
1045 | goto got_pg; | |
885036d3 KK |
1046 | if (gfp_mask & __GFP_NOFAIL) { |
1047 | blk_congestion_wait(WRITE, HZ/50); | |
1048 | goto nofail_alloc; | |
1049 | } | |
1da177e4 LT |
1050 | } |
1051 | goto nopage; | |
1052 | } | |
1053 | ||
1054 | /* Atomic allocations - we can't balance anything */ | |
1055 | if (!wait) | |
1056 | goto nopage; | |
1057 | ||
1058 | rebalance: | |
1059 | cond_resched(); | |
1060 | ||
1061 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1062 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1063 | p->flags |= PF_MEMALLOC; |
1064 | reclaim_state.reclaimed_slab = 0; | |
1065 | p->reclaim_state = &reclaim_state; | |
1066 | ||
7fb1d9fc | 1067 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1068 | |
1069 | p->reclaim_state = NULL; | |
1070 | p->flags &= ~PF_MEMALLOC; | |
1071 | ||
1072 | cond_resched(); | |
1073 | ||
1074 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1075 | page = get_page_from_freelist(gfp_mask, order, |
1076 | zonelist, alloc_flags); | |
1077 | if (page) | |
1078 | goto got_pg; | |
1da177e4 LT |
1079 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1080 | /* | |
1081 | * Go through the zonelist yet one more time, keep | |
1082 | * very high watermark here, this is only to catch | |
1083 | * a parallel oom killing, we must fail if we're still | |
1084 | * under heavy pressure. | |
1085 | */ | |
7fb1d9fc | 1086 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1087 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1088 | if (page) |
1089 | goto got_pg; | |
1da177e4 | 1090 | |
9b0f8b04 | 1091 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1092 | goto restart; |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Don't let big-order allocations loop unless the caller explicitly | |
1097 | * requests that. Wait for some write requests to complete then retry. | |
1098 | * | |
1099 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1100 | * <= 3, but that may not be true in other implementations. | |
1101 | */ | |
1102 | do_retry = 0; | |
1103 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1104 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1105 | do_retry = 1; | |
1106 | if (gfp_mask & __GFP_NOFAIL) | |
1107 | do_retry = 1; | |
1108 | } | |
1109 | if (do_retry) { | |
1110 | blk_congestion_wait(WRITE, HZ/50); | |
1111 | goto rebalance; | |
1112 | } | |
1113 | ||
1114 | nopage: | |
1115 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1116 | printk(KERN_WARNING "%s: page allocation failure." | |
1117 | " order:%d, mode:0x%x\n", | |
1118 | p->comm, order, gfp_mask); | |
1119 | dump_stack(); | |
578c2fd6 | 1120 | show_mem(); |
1da177e4 | 1121 | } |
1da177e4 | 1122 | got_pg: |
1da177e4 LT |
1123 | return page; |
1124 | } | |
1125 | ||
1126 | EXPORT_SYMBOL(__alloc_pages); | |
1127 | ||
1128 | /* | |
1129 | * Common helper functions. | |
1130 | */ | |
dd0fc66f | 1131 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1132 | { |
1133 | struct page * page; | |
1134 | page = alloc_pages(gfp_mask, order); | |
1135 | if (!page) | |
1136 | return 0; | |
1137 | return (unsigned long) page_address(page); | |
1138 | } | |
1139 | ||
1140 | EXPORT_SYMBOL(__get_free_pages); | |
1141 | ||
dd0fc66f | 1142 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1143 | { |
1144 | struct page * page; | |
1145 | ||
1146 | /* | |
1147 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1148 | * a highmem page | |
1149 | */ | |
725d704e | 1150 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1151 | |
1152 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1153 | if (page) | |
1154 | return (unsigned long) page_address(page); | |
1155 | return 0; | |
1156 | } | |
1157 | ||
1158 | EXPORT_SYMBOL(get_zeroed_page); | |
1159 | ||
1160 | void __pagevec_free(struct pagevec *pvec) | |
1161 | { | |
1162 | int i = pagevec_count(pvec); | |
1163 | ||
1164 | while (--i >= 0) | |
1165 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1166 | } | |
1167 | ||
1168 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1169 | { | |
b5810039 | 1170 | if (put_page_testzero(page)) { |
1da177e4 LT |
1171 | if (order == 0) |
1172 | free_hot_page(page); | |
1173 | else | |
1174 | __free_pages_ok(page, order); | |
1175 | } | |
1176 | } | |
1177 | ||
1178 | EXPORT_SYMBOL(__free_pages); | |
1179 | ||
1180 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1181 | { | |
1182 | if (addr != 0) { | |
725d704e | 1183 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1184 | __free_pages(virt_to_page((void *)addr), order); |
1185 | } | |
1186 | } | |
1187 | ||
1188 | EXPORT_SYMBOL(free_pages); | |
1189 | ||
1190 | /* | |
1191 | * Total amount of free (allocatable) RAM: | |
1192 | */ | |
1193 | unsigned int nr_free_pages(void) | |
1194 | { | |
1195 | unsigned int sum = 0; | |
1196 | struct zone *zone; | |
1197 | ||
1198 | for_each_zone(zone) | |
1199 | sum += zone->free_pages; | |
1200 | ||
1201 | return sum; | |
1202 | } | |
1203 | ||
1204 | EXPORT_SYMBOL(nr_free_pages); | |
1205 | ||
1206 | #ifdef CONFIG_NUMA | |
1207 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1208 | { | |
2f6726e5 CL |
1209 | unsigned int sum = 0; |
1210 | enum zone_type i; | |
1da177e4 LT |
1211 | |
1212 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1213 | sum += pgdat->node_zones[i].free_pages; | |
1214 | ||
1215 | return sum; | |
1216 | } | |
1217 | #endif | |
1218 | ||
1219 | static unsigned int nr_free_zone_pages(int offset) | |
1220 | { | |
e310fd43 MB |
1221 | /* Just pick one node, since fallback list is circular */ |
1222 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1223 | unsigned int sum = 0; |
1224 | ||
e310fd43 MB |
1225 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1226 | struct zone **zonep = zonelist->zones; | |
1227 | struct zone *zone; | |
1da177e4 | 1228 | |
e310fd43 MB |
1229 | for (zone = *zonep++; zone; zone = *zonep++) { |
1230 | unsigned long size = zone->present_pages; | |
1231 | unsigned long high = zone->pages_high; | |
1232 | if (size > high) | |
1233 | sum += size - high; | |
1da177e4 LT |
1234 | } |
1235 | ||
1236 | return sum; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1241 | */ | |
1242 | unsigned int nr_free_buffer_pages(void) | |
1243 | { | |
af4ca457 | 1244 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1245 | } |
1246 | ||
1247 | /* | |
1248 | * Amount of free RAM allocatable within all zones | |
1249 | */ | |
1250 | unsigned int nr_free_pagecache_pages(void) | |
1251 | { | |
af4ca457 | 1252 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 | 1253 | } |
1da177e4 LT |
1254 | #ifdef CONFIG_NUMA |
1255 | static void show_node(struct zone *zone) | |
1256 | { | |
89fa3024 | 1257 | printk("Node %ld ", zone_to_nid(zone)); |
1da177e4 LT |
1258 | } |
1259 | #else | |
1260 | #define show_node(zone) do { } while (0) | |
1261 | #endif | |
1262 | ||
1da177e4 LT |
1263 | void si_meminfo(struct sysinfo *val) |
1264 | { | |
1265 | val->totalram = totalram_pages; | |
1266 | val->sharedram = 0; | |
1267 | val->freeram = nr_free_pages(); | |
1268 | val->bufferram = nr_blockdev_pages(); | |
1da177e4 LT |
1269 | val->totalhigh = totalhigh_pages; |
1270 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1271 | val->mem_unit = PAGE_SIZE; |
1272 | } | |
1273 | ||
1274 | EXPORT_SYMBOL(si_meminfo); | |
1275 | ||
1276 | #ifdef CONFIG_NUMA | |
1277 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1278 | { | |
1279 | pg_data_t *pgdat = NODE_DATA(nid); | |
1280 | ||
1281 | val->totalram = pgdat->node_present_pages; | |
1282 | val->freeram = nr_free_pages_pgdat(pgdat); | |
98d2b0eb | 1283 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
1284 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
1285 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
98d2b0eb CL |
1286 | #else |
1287 | val->totalhigh = 0; | |
1288 | val->freehigh = 0; | |
1289 | #endif | |
1da177e4 LT |
1290 | val->mem_unit = PAGE_SIZE; |
1291 | } | |
1292 | #endif | |
1293 | ||
1294 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1295 | ||
1296 | /* | |
1297 | * Show free area list (used inside shift_scroll-lock stuff) | |
1298 | * We also calculate the percentage fragmentation. We do this by counting the | |
1299 | * memory on each free list with the exception of the first item on the list. | |
1300 | */ | |
1301 | void show_free_areas(void) | |
1302 | { | |
1da177e4 LT |
1303 | int cpu, temperature; |
1304 | unsigned long active; | |
1305 | unsigned long inactive; | |
1306 | unsigned long free; | |
1307 | struct zone *zone; | |
1308 | ||
1309 | for_each_zone(zone) { | |
1310 | show_node(zone); | |
1311 | printk("%s per-cpu:", zone->name); | |
1312 | ||
f3fe6512 | 1313 | if (!populated_zone(zone)) { |
1da177e4 LT |
1314 | printk(" empty\n"); |
1315 | continue; | |
1316 | } else | |
1317 | printk("\n"); | |
1318 | ||
6b482c67 | 1319 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1320 | struct per_cpu_pageset *pageset; |
1321 | ||
e7c8d5c9 | 1322 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1323 | |
1324 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1325 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1326 | cpu, |
1327 | temperature ? "cold" : "hot", | |
1da177e4 | 1328 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1329 | pageset->pcp[temperature].batch, |
1330 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1331 | } |
1332 | } | |
1333 | ||
1da177e4 LT |
1334 | get_zone_counts(&active, &inactive, &free); |
1335 | ||
1da177e4 LT |
1336 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " |
1337 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1338 | active, | |
1339 | inactive, | |
b1e7a8fd | 1340 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1341 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1342 | global_page_state(NR_UNSTABLE_NFS), |
1da177e4 | 1343 | nr_free_pages(), |
972d1a7b CL |
1344 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1345 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1346 | global_page_state(NR_FILE_MAPPED), |
df849a15 | 1347 | global_page_state(NR_PAGETABLE)); |
1da177e4 LT |
1348 | |
1349 | for_each_zone(zone) { | |
1350 | int i; | |
1351 | ||
1352 | show_node(zone); | |
1353 | printk("%s" | |
1354 | " free:%lukB" | |
1355 | " min:%lukB" | |
1356 | " low:%lukB" | |
1357 | " high:%lukB" | |
1358 | " active:%lukB" | |
1359 | " inactive:%lukB" | |
1360 | " present:%lukB" | |
1361 | " pages_scanned:%lu" | |
1362 | " all_unreclaimable? %s" | |
1363 | "\n", | |
1364 | zone->name, | |
1365 | K(zone->free_pages), | |
1366 | K(zone->pages_min), | |
1367 | K(zone->pages_low), | |
1368 | K(zone->pages_high), | |
1369 | K(zone->nr_active), | |
1370 | K(zone->nr_inactive), | |
1371 | K(zone->present_pages), | |
1372 | zone->pages_scanned, | |
1373 | (zone->all_unreclaimable ? "yes" : "no") | |
1374 | ); | |
1375 | printk("lowmem_reserve[]:"); | |
1376 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1377 | printk(" %lu", zone->lowmem_reserve[i]); | |
1378 | printk("\n"); | |
1379 | } | |
1380 | ||
1381 | for_each_zone(zone) { | |
8f9de51a | 1382 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
1383 | |
1384 | show_node(zone); | |
1385 | printk("%s: ", zone->name); | |
f3fe6512 | 1386 | if (!populated_zone(zone)) { |
1da177e4 LT |
1387 | printk("empty\n"); |
1388 | continue; | |
1389 | } | |
1390 | ||
1391 | spin_lock_irqsave(&zone->lock, flags); | |
1392 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1393 | nr[order] = zone->free_area[order].nr_free; |
1394 | total += nr[order] << order; | |
1da177e4 LT |
1395 | } |
1396 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1397 | for (order = 0; order < MAX_ORDER; order++) |
1398 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1399 | printk("= %lukB\n", K(total)); |
1400 | } | |
1401 | ||
1402 | show_swap_cache_info(); | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1407 | * |
1408 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1409 | */ |
86356ab1 | 1410 | static int __meminit build_zonelists_node(pg_data_t *pgdat, |
2f6726e5 | 1411 | struct zonelist *zonelist, int nr_zones, enum zone_type zone_type) |
1da177e4 | 1412 | { |
1a93205b CL |
1413 | struct zone *zone; |
1414 | ||
98d2b0eb | 1415 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1416 | zone_type++; |
02a68a5e CL |
1417 | |
1418 | do { | |
2f6726e5 | 1419 | zone_type--; |
070f8032 | 1420 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1421 | if (populated_zone(zone)) { |
070f8032 CL |
1422 | zonelist->zones[nr_zones++] = zone; |
1423 | check_highest_zone(zone_type); | |
1da177e4 | 1424 | } |
02a68a5e | 1425 | |
2f6726e5 | 1426 | } while (zone_type); |
070f8032 | 1427 | return nr_zones; |
1da177e4 LT |
1428 | } |
1429 | ||
1430 | #ifdef CONFIG_NUMA | |
1431 | #define MAX_NODE_LOAD (num_online_nodes()) | |
86356ab1 | 1432 | static int __meminitdata node_load[MAX_NUMNODES]; |
1da177e4 | 1433 | /** |
4dc3b16b | 1434 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1435 | * @node: node whose fallback list we're appending |
1436 | * @used_node_mask: nodemask_t of already used nodes | |
1437 | * | |
1438 | * We use a number of factors to determine which is the next node that should | |
1439 | * appear on a given node's fallback list. The node should not have appeared | |
1440 | * already in @node's fallback list, and it should be the next closest node | |
1441 | * according to the distance array (which contains arbitrary distance values | |
1442 | * from each node to each node in the system), and should also prefer nodes | |
1443 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1444 | * on them otherwise. | |
1445 | * It returns -1 if no node is found. | |
1446 | */ | |
86356ab1 | 1447 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 1448 | { |
4cf808eb | 1449 | int n, val; |
1da177e4 LT |
1450 | int min_val = INT_MAX; |
1451 | int best_node = -1; | |
1452 | ||
4cf808eb LT |
1453 | /* Use the local node if we haven't already */ |
1454 | if (!node_isset(node, *used_node_mask)) { | |
1455 | node_set(node, *used_node_mask); | |
1456 | return node; | |
1457 | } | |
1da177e4 | 1458 | |
4cf808eb LT |
1459 | for_each_online_node(n) { |
1460 | cpumask_t tmp; | |
1da177e4 LT |
1461 | |
1462 | /* Don't want a node to appear more than once */ | |
1463 | if (node_isset(n, *used_node_mask)) | |
1464 | continue; | |
1465 | ||
1da177e4 LT |
1466 | /* Use the distance array to find the distance */ |
1467 | val = node_distance(node, n); | |
1468 | ||
4cf808eb LT |
1469 | /* Penalize nodes under us ("prefer the next node") */ |
1470 | val += (n < node); | |
1471 | ||
1da177e4 LT |
1472 | /* Give preference to headless and unused nodes */ |
1473 | tmp = node_to_cpumask(n); | |
1474 | if (!cpus_empty(tmp)) | |
1475 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1476 | ||
1477 | /* Slight preference for less loaded node */ | |
1478 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1479 | val += node_load[n]; | |
1480 | ||
1481 | if (val < min_val) { | |
1482 | min_val = val; | |
1483 | best_node = n; | |
1484 | } | |
1485 | } | |
1486 | ||
1487 | if (best_node >= 0) | |
1488 | node_set(best_node, *used_node_mask); | |
1489 | ||
1490 | return best_node; | |
1491 | } | |
1492 | ||
86356ab1 | 1493 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1494 | { |
19655d34 CL |
1495 | int j, node, local_node; |
1496 | enum zone_type i; | |
1da177e4 LT |
1497 | int prev_node, load; |
1498 | struct zonelist *zonelist; | |
1499 | nodemask_t used_mask; | |
1500 | ||
1501 | /* initialize zonelists */ | |
19655d34 | 1502 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1503 | zonelist = pgdat->node_zonelists + i; |
1504 | zonelist->zones[0] = NULL; | |
1505 | } | |
1506 | ||
1507 | /* NUMA-aware ordering of nodes */ | |
1508 | local_node = pgdat->node_id; | |
1509 | load = num_online_nodes(); | |
1510 | prev_node = local_node; | |
1511 | nodes_clear(used_mask); | |
1512 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1513 | int distance = node_distance(local_node, node); |
1514 | ||
1515 | /* | |
1516 | * If another node is sufficiently far away then it is better | |
1517 | * to reclaim pages in a zone before going off node. | |
1518 | */ | |
1519 | if (distance > RECLAIM_DISTANCE) | |
1520 | zone_reclaim_mode = 1; | |
1521 | ||
1da177e4 LT |
1522 | /* |
1523 | * We don't want to pressure a particular node. | |
1524 | * So adding penalty to the first node in same | |
1525 | * distance group to make it round-robin. | |
1526 | */ | |
9eeff239 CL |
1527 | |
1528 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1529 | node_load[node] += load; |
1530 | prev_node = node; | |
1531 | load--; | |
19655d34 | 1532 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1533 | zonelist = pgdat->node_zonelists + i; |
1534 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1535 | ||
19655d34 | 1536 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1537 | zonelist->zones[j] = NULL; |
1538 | } | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | #else /* CONFIG_NUMA */ | |
1543 | ||
86356ab1 | 1544 | static void __meminit build_zonelists(pg_data_t *pgdat) |
1da177e4 | 1545 | { |
19655d34 CL |
1546 | int node, local_node; |
1547 | enum zone_type i,j; | |
1da177e4 LT |
1548 | |
1549 | local_node = pgdat->node_id; | |
19655d34 | 1550 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
1551 | struct zonelist *zonelist; |
1552 | ||
1553 | zonelist = pgdat->node_zonelists + i; | |
1554 | ||
19655d34 | 1555 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
1556 | /* |
1557 | * Now we build the zonelist so that it contains the zones | |
1558 | * of all the other nodes. | |
1559 | * We don't want to pressure a particular node, so when | |
1560 | * building the zones for node N, we make sure that the | |
1561 | * zones coming right after the local ones are those from | |
1562 | * node N+1 (modulo N) | |
1563 | */ | |
1564 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1565 | if (!node_online(node)) | |
1566 | continue; | |
19655d34 | 1567 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1568 | } |
1569 | for (node = 0; node < local_node; node++) { | |
1570 | if (!node_online(node)) | |
1571 | continue; | |
19655d34 | 1572 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
1573 | } |
1574 | ||
1575 | zonelist->zones[j] = NULL; | |
1576 | } | |
1577 | } | |
1578 | ||
1579 | #endif /* CONFIG_NUMA */ | |
1580 | ||
6811378e YG |
1581 | /* return values int ....just for stop_machine_run() */ |
1582 | static int __meminit __build_all_zonelists(void *dummy) | |
1da177e4 | 1583 | { |
6811378e YG |
1584 | int nid; |
1585 | for_each_online_node(nid) | |
1586 | build_zonelists(NODE_DATA(nid)); | |
1587 | return 0; | |
1588 | } | |
1589 | ||
1590 | void __meminit build_all_zonelists(void) | |
1591 | { | |
1592 | if (system_state == SYSTEM_BOOTING) { | |
1593 | __build_all_zonelists(0); | |
1594 | cpuset_init_current_mems_allowed(); | |
1595 | } else { | |
1596 | /* we have to stop all cpus to guaranntee there is no user | |
1597 | of zonelist */ | |
1598 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1599 | /* cpuset refresh routine should be here */ | |
1600 | } | |
bd1e22b8 AM |
1601 | vm_total_pages = nr_free_pagecache_pages(); |
1602 | printk("Built %i zonelists. Total pages: %ld\n", | |
1603 | num_online_nodes(), vm_total_pages); | |
1da177e4 LT |
1604 | } |
1605 | ||
1606 | /* | |
1607 | * Helper functions to size the waitqueue hash table. | |
1608 | * Essentially these want to choose hash table sizes sufficiently | |
1609 | * large so that collisions trying to wait on pages are rare. | |
1610 | * But in fact, the number of active page waitqueues on typical | |
1611 | * systems is ridiculously low, less than 200. So this is even | |
1612 | * conservative, even though it seems large. | |
1613 | * | |
1614 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1615 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1616 | */ | |
1617 | #define PAGES_PER_WAITQUEUE 256 | |
1618 | ||
cca448fe | 1619 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 1620 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
1621 | { |
1622 | unsigned long size = 1; | |
1623 | ||
1624 | pages /= PAGES_PER_WAITQUEUE; | |
1625 | ||
1626 | while (size < pages) | |
1627 | size <<= 1; | |
1628 | ||
1629 | /* | |
1630 | * Once we have dozens or even hundreds of threads sleeping | |
1631 | * on IO we've got bigger problems than wait queue collision. | |
1632 | * Limit the size of the wait table to a reasonable size. | |
1633 | */ | |
1634 | size = min(size, 4096UL); | |
1635 | ||
1636 | return max(size, 4UL); | |
1637 | } | |
cca448fe YG |
1638 | #else |
1639 | /* | |
1640 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1641 | * a suitable size for its wait_table. So we use the maximum size now. | |
1642 | * | |
1643 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1644 | * | |
1645 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1646 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1647 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1648 | * | |
1649 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1650 | * or more by the traditional way. (See above). It equals: | |
1651 | * | |
1652 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1653 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1654 | * powerpc (64K page size) : = (32G +16M)byte. | |
1655 | */ | |
1656 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1657 | { | |
1658 | return 4096UL; | |
1659 | } | |
1660 | #endif | |
1da177e4 LT |
1661 | |
1662 | /* | |
1663 | * This is an integer logarithm so that shifts can be used later | |
1664 | * to extract the more random high bits from the multiplicative | |
1665 | * hash function before the remainder is taken. | |
1666 | */ | |
1667 | static inline unsigned long wait_table_bits(unsigned long size) | |
1668 | { | |
1669 | return ffz(~size); | |
1670 | } | |
1671 | ||
1672 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1673 | ||
1da177e4 LT |
1674 | /* |
1675 | * Initially all pages are reserved - free ones are freed | |
1676 | * up by free_all_bootmem() once the early boot process is | |
1677 | * done. Non-atomic initialization, single-pass. | |
1678 | */ | |
c09b4240 | 1679 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1680 | unsigned long start_pfn) |
1681 | { | |
1da177e4 | 1682 | struct page *page; |
29751f69 AW |
1683 | unsigned long end_pfn = start_pfn + size; |
1684 | unsigned long pfn; | |
1da177e4 | 1685 | |
cbe8dd4a | 1686 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1687 | if (!early_pfn_valid(pfn)) |
1688 | continue; | |
1689 | page = pfn_to_page(pfn); | |
1690 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1691 | init_page_count(page); |
1da177e4 LT |
1692 | reset_page_mapcount(page); |
1693 | SetPageReserved(page); | |
1694 | INIT_LIST_HEAD(&page->lru); | |
1695 | #ifdef WANT_PAGE_VIRTUAL | |
1696 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1697 | if (!is_highmem_idx(zone)) | |
3212c6be | 1698 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1699 | #endif |
1da177e4 LT |
1700 | } |
1701 | } | |
1702 | ||
1703 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1704 | unsigned long size) | |
1705 | { | |
1706 | int order; | |
1707 | for (order = 0; order < MAX_ORDER ; order++) { | |
1708 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1709 | zone->free_area[order].nr_free = 0; | |
1710 | } | |
1711 | } | |
1712 | ||
d41dee36 | 1713 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
2f1b6248 CL |
1714 | void zonetable_add(struct zone *zone, int nid, enum zone_type zid, |
1715 | unsigned long pfn, unsigned long size) | |
d41dee36 AW |
1716 | { |
1717 | unsigned long snum = pfn_to_section_nr(pfn); | |
1718 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1719 | ||
1720 | if (FLAGS_HAS_NODE) | |
1721 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1722 | else | |
1723 | for (; snum <= end; snum++) | |
1724 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1725 | } | |
1726 | ||
1da177e4 LT |
1727 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1728 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1729 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1730 | #endif | |
1731 | ||
6292d9aa | 1732 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1733 | { |
1734 | int batch; | |
1735 | ||
1736 | /* | |
1737 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1738 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1739 | * |
1740 | * OK, so we don't know how big the cache is. So guess. | |
1741 | */ | |
1742 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1743 | if (batch * PAGE_SIZE > 512 * 1024) |
1744 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1745 | batch /= 4; /* We effectively *= 4 below */ |
1746 | if (batch < 1) | |
1747 | batch = 1; | |
1748 | ||
1749 | /* | |
0ceaacc9 NP |
1750 | * Clamp the batch to a 2^n - 1 value. Having a power |
1751 | * of 2 value was found to be more likely to have | |
1752 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1753 | * |
0ceaacc9 NP |
1754 | * For example if 2 tasks are alternately allocating |
1755 | * batches of pages, one task can end up with a lot | |
1756 | * of pages of one half of the possible page colors | |
1757 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1758 | */ |
0ceaacc9 | 1759 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1760 | |
e7c8d5c9 CL |
1761 | return batch; |
1762 | } | |
1763 | ||
2caaad41 CL |
1764 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1765 | { | |
1766 | struct per_cpu_pages *pcp; | |
1767 | ||
1c6fe946 MD |
1768 | memset(p, 0, sizeof(*p)); |
1769 | ||
2caaad41 CL |
1770 | pcp = &p->pcp[0]; /* hot */ |
1771 | pcp->count = 0; | |
2caaad41 CL |
1772 | pcp->high = 6 * batch; |
1773 | pcp->batch = max(1UL, 1 * batch); | |
1774 | INIT_LIST_HEAD(&pcp->list); | |
1775 | ||
1776 | pcp = &p->pcp[1]; /* cold*/ | |
1777 | pcp->count = 0; | |
2caaad41 | 1778 | pcp->high = 2 * batch; |
e46a5e28 | 1779 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1780 | INIT_LIST_HEAD(&pcp->list); |
1781 | } | |
1782 | ||
8ad4b1fb RS |
1783 | /* |
1784 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1785 | * to the value high for the pageset p. | |
1786 | */ | |
1787 | ||
1788 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1789 | unsigned long high) | |
1790 | { | |
1791 | struct per_cpu_pages *pcp; | |
1792 | ||
1793 | pcp = &p->pcp[0]; /* hot list */ | |
1794 | pcp->high = high; | |
1795 | pcp->batch = max(1UL, high/4); | |
1796 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1797 | pcp->batch = PAGE_SHIFT * 8; | |
1798 | } | |
1799 | ||
1800 | ||
e7c8d5c9 CL |
1801 | #ifdef CONFIG_NUMA |
1802 | /* | |
2caaad41 CL |
1803 | * Boot pageset table. One per cpu which is going to be used for all |
1804 | * zones and all nodes. The parameters will be set in such a way | |
1805 | * that an item put on a list will immediately be handed over to | |
1806 | * the buddy list. This is safe since pageset manipulation is done | |
1807 | * with interrupts disabled. | |
1808 | * | |
1809 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1810 | * |
1811 | * The boot_pagesets must be kept even after bootup is complete for | |
1812 | * unused processors and/or zones. They do play a role for bootstrapping | |
1813 | * hotplugged processors. | |
1814 | * | |
1815 | * zoneinfo_show() and maybe other functions do | |
1816 | * not check if the processor is online before following the pageset pointer. | |
1817 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1818 | */ |
88a2a4ac | 1819 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1820 | |
1821 | /* | |
1822 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1823 | * per cpu pageset array in struct zone. |
1824 | */ | |
6292d9aa | 1825 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1826 | { |
1827 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1828 | |
1829 | for_each_zone(zone) { | |
e7c8d5c9 | 1830 | |
23316bc8 | 1831 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1832 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1833 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1834 | goto bad; |
e7c8d5c9 | 1835 | |
23316bc8 | 1836 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1837 | |
1838 | if (percpu_pagelist_fraction) | |
1839 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1840 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1841 | } |
1842 | ||
1843 | return 0; | |
1844 | bad: | |
1845 | for_each_zone(dzone) { | |
1846 | if (dzone == zone) | |
1847 | break; | |
23316bc8 NP |
1848 | kfree(zone_pcp(dzone, cpu)); |
1849 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1850 | } |
1851 | return -ENOMEM; | |
1852 | } | |
1853 | ||
1854 | static inline void free_zone_pagesets(int cpu) | |
1855 | { | |
e7c8d5c9 CL |
1856 | struct zone *zone; |
1857 | ||
1858 | for_each_zone(zone) { | |
1859 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1860 | ||
f3ef9ead DR |
1861 | /* Free per_cpu_pageset if it is slab allocated */ |
1862 | if (pset != &boot_pageset[cpu]) | |
1863 | kfree(pset); | |
e7c8d5c9 | 1864 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 1865 | } |
e7c8d5c9 CL |
1866 | } |
1867 | ||
9c7b216d | 1868 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1869 | unsigned long action, |
1870 | void *hcpu) | |
1871 | { | |
1872 | int cpu = (long)hcpu; | |
1873 | int ret = NOTIFY_OK; | |
1874 | ||
1875 | switch (action) { | |
1876 | case CPU_UP_PREPARE: | |
1877 | if (process_zones(cpu)) | |
1878 | ret = NOTIFY_BAD; | |
1879 | break; | |
b0d41693 | 1880 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1881 | case CPU_DEAD: |
1882 | free_zone_pagesets(cpu); | |
1883 | break; | |
e7c8d5c9 CL |
1884 | default: |
1885 | break; | |
1886 | } | |
1887 | return ret; | |
1888 | } | |
1889 | ||
74b85f37 | 1890 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
1891 | { &pageset_cpuup_callback, NULL, 0 }; |
1892 | ||
78d9955b | 1893 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1894 | { |
1895 | int err; | |
1896 | ||
1897 | /* Initialize per_cpu_pageset for cpu 0. | |
1898 | * A cpuup callback will do this for every cpu | |
1899 | * as it comes online | |
1900 | */ | |
1901 | err = process_zones(smp_processor_id()); | |
1902 | BUG_ON(err); | |
1903 | register_cpu_notifier(&pageset_notifier); | |
1904 | } | |
1905 | ||
1906 | #endif | |
1907 | ||
c09b4240 | 1908 | static __meminit |
cca448fe | 1909 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
1910 | { |
1911 | int i; | |
1912 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 1913 | size_t alloc_size; |
ed8ece2e DH |
1914 | |
1915 | /* | |
1916 | * The per-page waitqueue mechanism uses hashed waitqueues | |
1917 | * per zone. | |
1918 | */ | |
02b694de YG |
1919 | zone->wait_table_hash_nr_entries = |
1920 | wait_table_hash_nr_entries(zone_size_pages); | |
1921 | zone->wait_table_bits = | |
1922 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
1923 | alloc_size = zone->wait_table_hash_nr_entries |
1924 | * sizeof(wait_queue_head_t); | |
1925 | ||
1926 | if (system_state == SYSTEM_BOOTING) { | |
1927 | zone->wait_table = (wait_queue_head_t *) | |
1928 | alloc_bootmem_node(pgdat, alloc_size); | |
1929 | } else { | |
1930 | /* | |
1931 | * This case means that a zone whose size was 0 gets new memory | |
1932 | * via memory hot-add. | |
1933 | * But it may be the case that a new node was hot-added. In | |
1934 | * this case vmalloc() will not be able to use this new node's | |
1935 | * memory - this wait_table must be initialized to use this new | |
1936 | * node itself as well. | |
1937 | * To use this new node's memory, further consideration will be | |
1938 | * necessary. | |
1939 | */ | |
1940 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
1941 | } | |
1942 | if (!zone->wait_table) | |
1943 | return -ENOMEM; | |
ed8ece2e | 1944 | |
02b694de | 1945 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 1946 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
1947 | |
1948 | return 0; | |
ed8ece2e DH |
1949 | } |
1950 | ||
c09b4240 | 1951 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
1952 | { |
1953 | int cpu; | |
1954 | unsigned long batch = zone_batchsize(zone); | |
1955 | ||
1956 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
1957 | #ifdef CONFIG_NUMA | |
1958 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 1959 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
1960 | setup_pageset(&boot_pageset[cpu],0); |
1961 | #else | |
1962 | setup_pageset(zone_pcp(zone,cpu), batch); | |
1963 | #endif | |
1964 | } | |
f5335c0f AB |
1965 | if (zone->present_pages) |
1966 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
1967 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
1968 | } |
1969 | ||
718127cc YG |
1970 | __meminit int init_currently_empty_zone(struct zone *zone, |
1971 | unsigned long zone_start_pfn, | |
1972 | unsigned long size) | |
ed8ece2e DH |
1973 | { |
1974 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
1975 | int ret; |
1976 | ret = zone_wait_table_init(zone, size); | |
1977 | if (ret) | |
1978 | return ret; | |
ed8ece2e DH |
1979 | pgdat->nr_zones = zone_idx(zone) + 1; |
1980 | ||
ed8ece2e DH |
1981 | zone->zone_start_pfn = zone_start_pfn; |
1982 | ||
1983 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
1984 | ||
1985 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
718127cc YG |
1986 | |
1987 | return 0; | |
ed8ece2e DH |
1988 | } |
1989 | ||
c713216d MG |
1990 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
1991 | /* | |
1992 | * Basic iterator support. Return the first range of PFNs for a node | |
1993 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
1994 | */ | |
1995 | static int __init first_active_region_index_in_nid(int nid) | |
1996 | { | |
1997 | int i; | |
1998 | ||
1999 | for (i = 0; i < nr_nodemap_entries; i++) | |
2000 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2001 | return i; | |
2002 | ||
2003 | return -1; | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * Basic iterator support. Return the next active range of PFNs for a node | |
2008 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2009 | */ | |
2010 | static int __init next_active_region_index_in_nid(int index, int nid) | |
2011 | { | |
2012 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2013 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2014 | return index; | |
2015 | ||
2016 | return -1; | |
2017 | } | |
2018 | ||
2019 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2020 | /* | |
2021 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2022 | * Architectures may implement their own version but if add_active_range() | |
2023 | * was used and there are no special requirements, this is a convenient | |
2024 | * alternative | |
2025 | */ | |
2026 | int __init early_pfn_to_nid(unsigned long pfn) | |
2027 | { | |
2028 | int i; | |
2029 | ||
2030 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2031 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2032 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2033 | ||
2034 | if (start_pfn <= pfn && pfn < end_pfn) | |
2035 | return early_node_map[i].nid; | |
2036 | } | |
2037 | ||
2038 | return 0; | |
2039 | } | |
2040 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2041 | ||
2042 | /* Basic iterator support to walk early_node_map[] */ | |
2043 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2044 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2045 | i = next_active_region_index_in_nid(i, nid)) | |
2046 | ||
2047 | /** | |
2048 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
2049 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed | |
2050 | * @max_low_pfn: The highest PFN that till be passed to free_bootmem_node | |
2051 | * | |
2052 | * If an architecture guarantees that all ranges registered with | |
2053 | * add_active_ranges() contain no holes and may be freed, this | |
2054 | * this function may be used instead of calling free_bootmem() manually. | |
2055 | */ | |
2056 | void __init free_bootmem_with_active_regions(int nid, | |
2057 | unsigned long max_low_pfn) | |
2058 | { | |
2059 | int i; | |
2060 | ||
2061 | for_each_active_range_index_in_nid(i, nid) { | |
2062 | unsigned long size_pages = 0; | |
2063 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2064 | ||
2065 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2066 | continue; | |
2067 | ||
2068 | if (end_pfn > max_low_pfn) | |
2069 | end_pfn = max_low_pfn; | |
2070 | ||
2071 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2072 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2073 | PFN_PHYS(early_node_map[i].start_pfn), | |
2074 | size_pages << PAGE_SHIFT); | |
2075 | } | |
2076 | } | |
2077 | ||
2078 | /** | |
2079 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
2080 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used | |
2081 | * | |
2082 | * If an architecture guarantees that all ranges registered with | |
2083 | * add_active_ranges() contain no holes and may be freed, this | |
2084 | * this function may be used instead of calling memory_present() manually. | |
2085 | */ | |
2086 | void __init sparse_memory_present_with_active_regions(int nid) | |
2087 | { | |
2088 | int i; | |
2089 | ||
2090 | for_each_active_range_index_in_nid(i, nid) | |
2091 | memory_present(early_node_map[i].nid, | |
2092 | early_node_map[i].start_pfn, | |
2093 | early_node_map[i].end_pfn); | |
2094 | } | |
2095 | ||
2096 | /** | |
2097 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
2098 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned | |
2099 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn | |
2100 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn | |
2101 | * | |
2102 | * It returns the start and end page frame of a node based on information | |
2103 | * provided by an arch calling add_active_range(). If called for a node | |
2104 | * with no available memory, a warning is printed and the start and end | |
2105 | * PFNs will be 0 | |
2106 | */ | |
2107 | void __init get_pfn_range_for_nid(unsigned int nid, | |
2108 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2109 | { | |
2110 | int i; | |
2111 | *start_pfn = -1UL; | |
2112 | *end_pfn = 0; | |
2113 | ||
2114 | for_each_active_range_index_in_nid(i, nid) { | |
2115 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2116 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2117 | } | |
2118 | ||
2119 | if (*start_pfn == -1UL) { | |
2120 | printk(KERN_WARNING "Node %u active with no memory\n", nid); | |
2121 | *start_pfn = 0; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | /* | |
2126 | * Return the number of pages a zone spans in a node, including holes | |
2127 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2128 | */ | |
2129 | unsigned long __init zone_spanned_pages_in_node(int nid, | |
2130 | unsigned long zone_type, | |
2131 | unsigned long *ignored) | |
2132 | { | |
2133 | unsigned long node_start_pfn, node_end_pfn; | |
2134 | unsigned long zone_start_pfn, zone_end_pfn; | |
2135 | ||
2136 | /* Get the start and end of the node and zone */ | |
2137 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2138 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2139 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2140 | ||
2141 | /* Check that this node has pages within the zone's required range */ | |
2142 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2143 | return 0; | |
2144 | ||
2145 | /* Move the zone boundaries inside the node if necessary */ | |
2146 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2147 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2148 | ||
2149 | /* Return the spanned pages */ | |
2150 | return zone_end_pfn - zone_start_pfn; | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
2155 | * then all holes in the requested range will be accounted for | |
2156 | */ | |
2157 | unsigned long __init __absent_pages_in_range(int nid, | |
2158 | unsigned long range_start_pfn, | |
2159 | unsigned long range_end_pfn) | |
2160 | { | |
2161 | int i = 0; | |
2162 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2163 | unsigned long start_pfn; | |
2164 | ||
2165 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2166 | i = first_active_region_index_in_nid(nid); | |
2167 | if (i == -1) | |
2168 | return 0; | |
2169 | ||
2170 | prev_end_pfn = early_node_map[i].start_pfn; | |
2171 | ||
2172 | /* Find all holes for the zone within the node */ | |
2173 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2174 | ||
2175 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2176 | if (prev_end_pfn >= range_end_pfn) | |
2177 | break; | |
2178 | ||
2179 | /* Make sure the end of the zone is not within the hole */ | |
2180 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2181 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2182 | ||
2183 | /* Update the hole size cound and move on */ | |
2184 | if (start_pfn > range_start_pfn) { | |
2185 | BUG_ON(prev_end_pfn > start_pfn); | |
2186 | hole_pages += start_pfn - prev_end_pfn; | |
2187 | } | |
2188 | prev_end_pfn = early_node_map[i].end_pfn; | |
2189 | } | |
2190 | ||
2191 | return hole_pages; | |
2192 | } | |
2193 | ||
2194 | /** | |
2195 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2196 | * @start_pfn: The start PFN to start searching for holes | |
2197 | * @end_pfn: The end PFN to stop searching for holes | |
2198 | * | |
2199 | * It returns the number of pages frames in memory holes within a range | |
2200 | */ | |
2201 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2202 | unsigned long end_pfn) | |
2203 | { | |
2204 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2205 | } | |
2206 | ||
2207 | /* Return the number of page frames in holes in a zone on a node */ | |
2208 | unsigned long __init zone_absent_pages_in_node(int nid, | |
2209 | unsigned long zone_type, | |
2210 | unsigned long *ignored) | |
2211 | { | |
2212 | return __absent_pages_in_range(nid, | |
2213 | arch_zone_lowest_possible_pfn[zone_type], | |
2214 | arch_zone_highest_possible_pfn[zone_type]); | |
2215 | } | |
2216 | #else | |
2217 | static inline unsigned long zone_spanned_pages_in_node(int nid, | |
2218 | unsigned long zone_type, | |
2219 | unsigned long *zones_size) | |
2220 | { | |
2221 | return zones_size[zone_type]; | |
2222 | } | |
2223 | ||
2224 | static inline unsigned long zone_absent_pages_in_node(int nid, | |
2225 | unsigned long zone_type, | |
2226 | unsigned long *zholes_size) | |
2227 | { | |
2228 | if (!zholes_size) | |
2229 | return 0; | |
2230 | ||
2231 | return zholes_size[zone_type]; | |
2232 | } | |
2233 | #endif | |
2234 | ||
2235 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, | |
2236 | unsigned long *zones_size, unsigned long *zholes_size) | |
2237 | { | |
2238 | unsigned long realtotalpages, totalpages = 0; | |
2239 | enum zone_type i; | |
2240 | ||
2241 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2242 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
2243 | zones_size); | |
2244 | pgdat->node_spanned_pages = totalpages; | |
2245 | ||
2246 | realtotalpages = totalpages; | |
2247 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2248 | realtotalpages -= | |
2249 | zone_absent_pages_in_node(pgdat->node_id, i, | |
2250 | zholes_size); | |
2251 | pgdat->node_present_pages = realtotalpages; | |
2252 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
2253 | realtotalpages); | |
2254 | } | |
2255 | ||
1da177e4 LT |
2256 | /* |
2257 | * Set up the zone data structures: | |
2258 | * - mark all pages reserved | |
2259 | * - mark all memory queues empty | |
2260 | * - clear the memory bitmaps | |
2261 | */ | |
86356ab1 | 2262 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
2263 | unsigned long *zones_size, unsigned long *zholes_size) |
2264 | { | |
2f1b6248 | 2265 | enum zone_type j; |
ed8ece2e | 2266 | int nid = pgdat->node_id; |
1da177e4 | 2267 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 2268 | int ret; |
1da177e4 | 2269 | |
208d54e5 | 2270 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2271 | pgdat->nr_zones = 0; |
2272 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2273 | pgdat->kswapd_max_order = 0; | |
2274 | ||
2275 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2276 | struct zone *zone = pgdat->node_zones + j; | |
2277 | unsigned long size, realsize; | |
1da177e4 | 2278 | |
c713216d MG |
2279 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
2280 | realsize = size - zone_absent_pages_in_node(nid, j, | |
2281 | zholes_size); | |
1da177e4 | 2282 | |
98d2b0eb | 2283 | if (!is_highmem_idx(j)) |
1da177e4 LT |
2284 | nr_kernel_pages += realsize; |
2285 | nr_all_pages += realsize; | |
2286 | ||
2287 | zone->spanned_pages = size; | |
2288 | zone->present_pages = realsize; | |
9614634f | 2289 | #ifdef CONFIG_NUMA |
8417bba4 | 2290 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 2291 | / 100; |
0ff38490 | 2292 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 2293 | #endif |
1da177e4 LT |
2294 | zone->name = zone_names[j]; |
2295 | spin_lock_init(&zone->lock); | |
2296 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2297 | zone_seqlock_init(zone); |
1da177e4 LT |
2298 | zone->zone_pgdat = pgdat; |
2299 | zone->free_pages = 0; | |
2300 | ||
2301 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2302 | ||
ed8ece2e | 2303 | zone_pcp_init(zone); |
1da177e4 LT |
2304 | INIT_LIST_HEAD(&zone->active_list); |
2305 | INIT_LIST_HEAD(&zone->inactive_list); | |
2306 | zone->nr_scan_active = 0; | |
2307 | zone->nr_scan_inactive = 0; | |
2308 | zone->nr_active = 0; | |
2309 | zone->nr_inactive = 0; | |
2244b95a | 2310 | zap_zone_vm_stats(zone); |
53e9a615 | 2311 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2312 | if (!size) |
2313 | continue; | |
2314 | ||
d41dee36 | 2315 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
718127cc YG |
2316 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
2317 | BUG_ON(ret); | |
1da177e4 | 2318 | zone_start_pfn += size; |
1da177e4 LT |
2319 | } |
2320 | } | |
2321 | ||
2322 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2323 | { | |
1da177e4 LT |
2324 | /* Skip empty nodes */ |
2325 | if (!pgdat->node_spanned_pages) | |
2326 | return; | |
2327 | ||
d41dee36 | 2328 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2329 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2330 | if (!pgdat->node_mem_map) { | |
e984bb43 | 2331 | unsigned long size, start, end; |
d41dee36 AW |
2332 | struct page *map; |
2333 | ||
e984bb43 BP |
2334 | /* |
2335 | * The zone's endpoints aren't required to be MAX_ORDER | |
2336 | * aligned but the node_mem_map endpoints must be in order | |
2337 | * for the buddy allocator to function correctly. | |
2338 | */ | |
2339 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2340 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2341 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2342 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
2343 | map = alloc_remap(pgdat->node_id, size); |
2344 | if (!map) | |
2345 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 2346 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 2347 | } |
d41dee36 | 2348 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2349 | /* |
2350 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2351 | */ | |
c713216d | 2352 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 2353 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
2354 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2355 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
2356 | mem_map -= pgdat->node_start_pfn; | |
2357 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2358 | } | |
1da177e4 | 2359 | #endif |
d41dee36 | 2360 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2361 | } |
2362 | ||
86356ab1 | 2363 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
2364 | unsigned long *zones_size, unsigned long node_start_pfn, |
2365 | unsigned long *zholes_size) | |
2366 | { | |
2367 | pgdat->node_id = nid; | |
2368 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 2369 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
2370 | |
2371 | alloc_node_mem_map(pgdat); | |
2372 | ||
2373 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2374 | } | |
2375 | ||
c713216d MG |
2376 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2377 | /** | |
2378 | * add_active_range - Register a range of PFNs backed by physical memory | |
2379 | * @nid: The node ID the range resides on | |
2380 | * @start_pfn: The start PFN of the available physical memory | |
2381 | * @end_pfn: The end PFN of the available physical memory | |
2382 | * | |
2383 | * These ranges are stored in an early_node_map[] and later used by | |
2384 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
2385 | * range spans a memory hole, it is up to the architecture to ensure | |
2386 | * the memory is not freed by the bootmem allocator. If possible | |
2387 | * the range being registered will be merged with existing ranges. | |
2388 | */ | |
2389 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
2390 | unsigned long end_pfn) | |
2391 | { | |
2392 | int i; | |
2393 | ||
2394 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
2395 | "%d entries of %d used\n", | |
2396 | nid, start_pfn, end_pfn, | |
2397 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
2398 | ||
2399 | /* Merge with existing active regions if possible */ | |
2400 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2401 | if (early_node_map[i].nid != nid) | |
2402 | continue; | |
2403 | ||
2404 | /* Skip if an existing region covers this new one */ | |
2405 | if (start_pfn >= early_node_map[i].start_pfn && | |
2406 | end_pfn <= early_node_map[i].end_pfn) | |
2407 | return; | |
2408 | ||
2409 | /* Merge forward if suitable */ | |
2410 | if (start_pfn <= early_node_map[i].end_pfn && | |
2411 | end_pfn > early_node_map[i].end_pfn) { | |
2412 | early_node_map[i].end_pfn = end_pfn; | |
2413 | return; | |
2414 | } | |
2415 | ||
2416 | /* Merge backward if suitable */ | |
2417 | if (start_pfn < early_node_map[i].end_pfn && | |
2418 | end_pfn >= early_node_map[i].start_pfn) { | |
2419 | early_node_map[i].start_pfn = start_pfn; | |
2420 | return; | |
2421 | } | |
2422 | } | |
2423 | ||
2424 | /* Check that early_node_map is large enough */ | |
2425 | if (i >= MAX_ACTIVE_REGIONS) { | |
2426 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
2427 | MAX_ACTIVE_REGIONS); | |
2428 | return; | |
2429 | } | |
2430 | ||
2431 | early_node_map[i].nid = nid; | |
2432 | early_node_map[i].start_pfn = start_pfn; | |
2433 | early_node_map[i].end_pfn = end_pfn; | |
2434 | nr_nodemap_entries = i + 1; | |
2435 | } | |
2436 | ||
2437 | /** | |
2438 | * shrink_active_range - Shrink an existing registered range of PFNs | |
2439 | * @nid: The node id the range is on that should be shrunk | |
2440 | * @old_end_pfn: The old end PFN of the range | |
2441 | * @new_end_pfn: The new PFN of the range | |
2442 | * | |
2443 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
2444 | * The map is kept at the end physical page range that has already been | |
2445 | * registered with add_active_range(). This function allows an arch to shrink | |
2446 | * an existing registered range. | |
2447 | */ | |
2448 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
2449 | unsigned long new_end_pfn) | |
2450 | { | |
2451 | int i; | |
2452 | ||
2453 | /* Find the old active region end and shrink */ | |
2454 | for_each_active_range_index_in_nid(i, nid) | |
2455 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
2456 | early_node_map[i].end_pfn = new_end_pfn; | |
2457 | break; | |
2458 | } | |
2459 | } | |
2460 | ||
2461 | /** | |
2462 | * remove_all_active_ranges - Remove all currently registered regions | |
2463 | * During discovery, it may be found that a table like SRAT is invalid | |
2464 | * and an alternative discovery method must be used. This function removes | |
2465 | * all currently registered regions. | |
2466 | */ | |
2467 | void __init remove_all_active_ranges() | |
2468 | { | |
2469 | memset(early_node_map, 0, sizeof(early_node_map)); | |
2470 | nr_nodemap_entries = 0; | |
2471 | } | |
2472 | ||
2473 | /* Compare two active node_active_regions */ | |
2474 | static int __init cmp_node_active_region(const void *a, const void *b) | |
2475 | { | |
2476 | struct node_active_region *arange = (struct node_active_region *)a; | |
2477 | struct node_active_region *brange = (struct node_active_region *)b; | |
2478 | ||
2479 | /* Done this way to avoid overflows */ | |
2480 | if (arange->start_pfn > brange->start_pfn) | |
2481 | return 1; | |
2482 | if (arange->start_pfn < brange->start_pfn) | |
2483 | return -1; | |
2484 | ||
2485 | return 0; | |
2486 | } | |
2487 | ||
2488 | /* sort the node_map by start_pfn */ | |
2489 | static void __init sort_node_map(void) | |
2490 | { | |
2491 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
2492 | sizeof(struct node_active_region), | |
2493 | cmp_node_active_region, NULL); | |
2494 | } | |
2495 | ||
2496 | /* Find the lowest pfn for a node. This depends on a sorted early_node_map */ | |
2497 | unsigned long __init find_min_pfn_for_node(unsigned long nid) | |
2498 | { | |
2499 | int i; | |
2500 | ||
2501 | /* Assuming a sorted map, the first range found has the starting pfn */ | |
2502 | for_each_active_range_index_in_nid(i, nid) | |
2503 | return early_node_map[i].start_pfn; | |
2504 | ||
2505 | printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid); | |
2506 | return 0; | |
2507 | } | |
2508 | ||
2509 | /** | |
2510 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
2511 | * | |
2512 | * It returns the minimum PFN based on information provided via | |
2513 | * add_active_range() | |
2514 | */ | |
2515 | unsigned long __init find_min_pfn_with_active_regions(void) | |
2516 | { | |
2517 | return find_min_pfn_for_node(MAX_NUMNODES); | |
2518 | } | |
2519 | ||
2520 | /** | |
2521 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
2522 | * | |
2523 | * It returns the maximum PFN based on information provided via | |
2524 | * add_active_range() | |
2525 | */ | |
2526 | unsigned long __init find_max_pfn_with_active_regions(void) | |
2527 | { | |
2528 | int i; | |
2529 | unsigned long max_pfn = 0; | |
2530 | ||
2531 | for (i = 0; i < nr_nodemap_entries; i++) | |
2532 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
2533 | ||
2534 | return max_pfn; | |
2535 | } | |
2536 | ||
2537 | /** | |
2538 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
2539 | * @arch_max_dma_pfn: The maximum PFN usable for ZONE_DMA | |
2540 | * @arch_max_dma32_pfn: The maximum PFN usable for ZONE_DMA32 | |
2541 | * @arch_max_low_pfn: The maximum PFN usable for ZONE_NORMAL | |
2542 | * @arch_max_high_pfn: The maximum PFN usable for ZONE_HIGHMEM | |
2543 | * | |
2544 | * This will call free_area_init_node() for each active node in the system. | |
2545 | * Using the page ranges provided by add_active_range(), the size of each | |
2546 | * zone in each node and their holes is calculated. If the maximum PFN | |
2547 | * between two adjacent zones match, it is assumed that the zone is empty. | |
2548 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
2549 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
2550 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
2551 | * at arch_max_dma_pfn. | |
2552 | */ | |
2553 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
2554 | { | |
2555 | unsigned long nid; | |
2556 | enum zone_type i; | |
2557 | ||
2558 | /* Record where the zone boundaries are */ | |
2559 | memset(arch_zone_lowest_possible_pfn, 0, | |
2560 | sizeof(arch_zone_lowest_possible_pfn)); | |
2561 | memset(arch_zone_highest_possible_pfn, 0, | |
2562 | sizeof(arch_zone_highest_possible_pfn)); | |
2563 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
2564 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
2565 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2566 | arch_zone_lowest_possible_pfn[i] = | |
2567 | arch_zone_highest_possible_pfn[i-1]; | |
2568 | arch_zone_highest_possible_pfn[i] = | |
2569 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
2570 | } | |
2571 | ||
2572 | /* Regions in the early_node_map can be in any order */ | |
2573 | sort_node_map(); | |
2574 | ||
2575 | /* Print out the zone ranges */ | |
2576 | printk("Zone PFN ranges:\n"); | |
2577 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2578 | printk(" %-8s %8lu -> %8lu\n", | |
2579 | zone_names[i], | |
2580 | arch_zone_lowest_possible_pfn[i], | |
2581 | arch_zone_highest_possible_pfn[i]); | |
2582 | ||
2583 | /* Print out the early_node_map[] */ | |
2584 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
2585 | for (i = 0; i < nr_nodemap_entries; i++) | |
2586 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
2587 | early_node_map[i].start_pfn, | |
2588 | early_node_map[i].end_pfn); | |
2589 | ||
2590 | /* Initialise every node */ | |
2591 | for_each_online_node(nid) { | |
2592 | pg_data_t *pgdat = NODE_DATA(nid); | |
2593 | free_area_init_node(nid, pgdat, NULL, | |
2594 | find_min_pfn_for_node(nid), NULL); | |
2595 | } | |
2596 | } | |
2597 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2598 | ||
93b7504e | 2599 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2600 | static bootmem_data_t contig_bootmem_data; |
2601 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2602 | ||
2603 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2604 | #endif |
1da177e4 LT |
2605 | |
2606 | void __init free_area_init(unsigned long *zones_size) | |
2607 | { | |
93b7504e | 2608 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2609 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2610 | } | |
1da177e4 | 2611 | |
1da177e4 LT |
2612 | #ifdef CONFIG_HOTPLUG_CPU |
2613 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2614 | unsigned long action, void *hcpu) | |
2615 | { | |
2616 | int cpu = (unsigned long)hcpu; | |
1da177e4 LT |
2617 | |
2618 | if (action == CPU_DEAD) { | |
1da177e4 LT |
2619 | local_irq_disable(); |
2620 | __drain_pages(cpu); | |
f8891e5e | 2621 | vm_events_fold_cpu(cpu); |
1da177e4 | 2622 | local_irq_enable(); |
2244b95a | 2623 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
2624 | } |
2625 | return NOTIFY_OK; | |
2626 | } | |
2627 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2628 | ||
2629 | void __init page_alloc_init(void) | |
2630 | { | |
2631 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2632 | } | |
2633 | ||
cb45b0e9 HA |
2634 | /* |
2635 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
2636 | * or min_free_kbytes changes. | |
2637 | */ | |
2638 | static void calculate_totalreserve_pages(void) | |
2639 | { | |
2640 | struct pglist_data *pgdat; | |
2641 | unsigned long reserve_pages = 0; | |
2f6726e5 | 2642 | enum zone_type i, j; |
cb45b0e9 HA |
2643 | |
2644 | for_each_online_pgdat(pgdat) { | |
2645 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2646 | struct zone *zone = pgdat->node_zones + i; | |
2647 | unsigned long max = 0; | |
2648 | ||
2649 | /* Find valid and maximum lowmem_reserve in the zone */ | |
2650 | for (j = i; j < MAX_NR_ZONES; j++) { | |
2651 | if (zone->lowmem_reserve[j] > max) | |
2652 | max = zone->lowmem_reserve[j]; | |
2653 | } | |
2654 | ||
2655 | /* we treat pages_high as reserved pages. */ | |
2656 | max += zone->pages_high; | |
2657 | ||
2658 | if (max > zone->present_pages) | |
2659 | max = zone->present_pages; | |
2660 | reserve_pages += max; | |
2661 | } | |
2662 | } | |
2663 | totalreserve_pages = reserve_pages; | |
2664 | } | |
2665 | ||
1da177e4 LT |
2666 | /* |
2667 | * setup_per_zone_lowmem_reserve - called whenever | |
2668 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2669 | * has a correct pages reserved value, so an adequate number of | |
2670 | * pages are left in the zone after a successful __alloc_pages(). | |
2671 | */ | |
2672 | static void setup_per_zone_lowmem_reserve(void) | |
2673 | { | |
2674 | struct pglist_data *pgdat; | |
2f6726e5 | 2675 | enum zone_type j, idx; |
1da177e4 | 2676 | |
ec936fc5 | 2677 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2678 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2679 | struct zone *zone = pgdat->node_zones + j; | |
2680 | unsigned long present_pages = zone->present_pages; | |
2681 | ||
2682 | zone->lowmem_reserve[j] = 0; | |
2683 | ||
2f6726e5 CL |
2684 | idx = j; |
2685 | while (idx) { | |
1da177e4 LT |
2686 | struct zone *lower_zone; |
2687 | ||
2f6726e5 CL |
2688 | idx--; |
2689 | ||
1da177e4 LT |
2690 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
2691 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2692 | ||
2693 | lower_zone = pgdat->node_zones + idx; | |
2694 | lower_zone->lowmem_reserve[j] = present_pages / | |
2695 | sysctl_lowmem_reserve_ratio[idx]; | |
2696 | present_pages += lower_zone->present_pages; | |
2697 | } | |
2698 | } | |
2699 | } | |
cb45b0e9 HA |
2700 | |
2701 | /* update totalreserve_pages */ | |
2702 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2703 | } |
2704 | ||
2705 | /* | |
2706 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2707 | * that the pages_{min,low,high} values for each zone are set correctly | |
2708 | * with respect to min_free_kbytes. | |
2709 | */ | |
3947be19 | 2710 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2711 | { |
2712 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2713 | unsigned long lowmem_pages = 0; | |
2714 | struct zone *zone; | |
2715 | unsigned long flags; | |
2716 | ||
2717 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2718 | for_each_zone(zone) { | |
2719 | if (!is_highmem(zone)) | |
2720 | lowmem_pages += zone->present_pages; | |
2721 | } | |
2722 | ||
2723 | for_each_zone(zone) { | |
ac924c60 AM |
2724 | u64 tmp; |
2725 | ||
1da177e4 | 2726 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
2727 | tmp = (u64)pages_min * zone->present_pages; |
2728 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
2729 | if (is_highmem(zone)) { |
2730 | /* | |
669ed175 NP |
2731 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2732 | * need highmem pages, so cap pages_min to a small | |
2733 | * value here. | |
2734 | * | |
2735 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2736 | * deltas controls asynch page reclaim, and so should | |
2737 | * not be capped for highmem. | |
1da177e4 LT |
2738 | */ |
2739 | int min_pages; | |
2740 | ||
2741 | min_pages = zone->present_pages / 1024; | |
2742 | if (min_pages < SWAP_CLUSTER_MAX) | |
2743 | min_pages = SWAP_CLUSTER_MAX; | |
2744 | if (min_pages > 128) | |
2745 | min_pages = 128; | |
2746 | zone->pages_min = min_pages; | |
2747 | } else { | |
669ed175 NP |
2748 | /* |
2749 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2750 | * proportionate to the zone's size. |
2751 | */ | |
669ed175 | 2752 | zone->pages_min = tmp; |
1da177e4 LT |
2753 | } |
2754 | ||
ac924c60 AM |
2755 | zone->pages_low = zone->pages_min + (tmp >> 2); |
2756 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
1da177e4 LT |
2757 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2758 | } | |
cb45b0e9 HA |
2759 | |
2760 | /* update totalreserve_pages */ | |
2761 | calculate_totalreserve_pages(); | |
1da177e4 LT |
2762 | } |
2763 | ||
2764 | /* | |
2765 | * Initialise min_free_kbytes. | |
2766 | * | |
2767 | * For small machines we want it small (128k min). For large machines | |
2768 | * we want it large (64MB max). But it is not linear, because network | |
2769 | * bandwidth does not increase linearly with machine size. We use | |
2770 | * | |
2771 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2772 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2773 | * | |
2774 | * which yields | |
2775 | * | |
2776 | * 16MB: 512k | |
2777 | * 32MB: 724k | |
2778 | * 64MB: 1024k | |
2779 | * 128MB: 1448k | |
2780 | * 256MB: 2048k | |
2781 | * 512MB: 2896k | |
2782 | * 1024MB: 4096k | |
2783 | * 2048MB: 5792k | |
2784 | * 4096MB: 8192k | |
2785 | * 8192MB: 11584k | |
2786 | * 16384MB: 16384k | |
2787 | */ | |
2788 | static int __init init_per_zone_pages_min(void) | |
2789 | { | |
2790 | unsigned long lowmem_kbytes; | |
2791 | ||
2792 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2793 | ||
2794 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2795 | if (min_free_kbytes < 128) | |
2796 | min_free_kbytes = 128; | |
2797 | if (min_free_kbytes > 65536) | |
2798 | min_free_kbytes = 65536; | |
2799 | setup_per_zone_pages_min(); | |
2800 | setup_per_zone_lowmem_reserve(); | |
2801 | return 0; | |
2802 | } | |
2803 | module_init(init_per_zone_pages_min) | |
2804 | ||
2805 | /* | |
2806 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2807 | * that we can call two helper functions whenever min_free_kbytes | |
2808 | * changes. | |
2809 | */ | |
2810 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2811 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2812 | { | |
2813 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2814 | setup_per_zone_pages_min(); | |
2815 | return 0; | |
2816 | } | |
2817 | ||
9614634f CL |
2818 | #ifdef CONFIG_NUMA |
2819 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
2820 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2821 | { | |
2822 | struct zone *zone; | |
2823 | int rc; | |
2824 | ||
2825 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2826 | if (rc) | |
2827 | return rc; | |
2828 | ||
2829 | for_each_zone(zone) | |
8417bba4 | 2830 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
2831 | sysctl_min_unmapped_ratio) / 100; |
2832 | return 0; | |
2833 | } | |
0ff38490 CL |
2834 | |
2835 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
2836 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2837 | { | |
2838 | struct zone *zone; | |
2839 | int rc; | |
2840 | ||
2841 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2842 | if (rc) | |
2843 | return rc; | |
2844 | ||
2845 | for_each_zone(zone) | |
2846 | zone->min_slab_pages = (zone->present_pages * | |
2847 | sysctl_min_slab_ratio) / 100; | |
2848 | return 0; | |
2849 | } | |
9614634f CL |
2850 | #endif |
2851 | ||
1da177e4 LT |
2852 | /* |
2853 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2854 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2855 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2856 | * | |
2857 | * The reserve ratio obviously has absolutely no relation with the | |
2858 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2859 | * if in function of the boot time zone sizes. | |
2860 | */ | |
2861 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2862 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2863 | { | |
2864 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2865 | setup_per_zone_lowmem_reserve(); | |
2866 | return 0; | |
2867 | } | |
2868 | ||
8ad4b1fb RS |
2869 | /* |
2870 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2871 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2872 | * can have before it gets flushed back to buddy allocator. | |
2873 | */ | |
2874 | ||
2875 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2876 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2877 | { | |
2878 | struct zone *zone; | |
2879 | unsigned int cpu; | |
2880 | int ret; | |
2881 | ||
2882 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2883 | if (!write || (ret == -EINVAL)) | |
2884 | return ret; | |
2885 | for_each_zone(zone) { | |
2886 | for_each_online_cpu(cpu) { | |
2887 | unsigned long high; | |
2888 | high = zone->present_pages / percpu_pagelist_fraction; | |
2889 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2890 | } | |
2891 | } | |
2892 | return 0; | |
2893 | } | |
2894 | ||
f034b5d4 | 2895 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
2896 | |
2897 | #ifdef CONFIG_NUMA | |
2898 | static int __init set_hashdist(char *str) | |
2899 | { | |
2900 | if (!str) | |
2901 | return 0; | |
2902 | hashdist = simple_strtoul(str, &str, 0); | |
2903 | return 1; | |
2904 | } | |
2905 | __setup("hashdist=", set_hashdist); | |
2906 | #endif | |
2907 | ||
2908 | /* | |
2909 | * allocate a large system hash table from bootmem | |
2910 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2911 | * quantity of entries | |
2912 | * - limit is the number of hash buckets, not the total allocation size | |
2913 | */ | |
2914 | void *__init alloc_large_system_hash(const char *tablename, | |
2915 | unsigned long bucketsize, | |
2916 | unsigned long numentries, | |
2917 | int scale, | |
2918 | int flags, | |
2919 | unsigned int *_hash_shift, | |
2920 | unsigned int *_hash_mask, | |
2921 | unsigned long limit) | |
2922 | { | |
2923 | unsigned long long max = limit; | |
2924 | unsigned long log2qty, size; | |
2925 | void *table = NULL; | |
2926 | ||
2927 | /* allow the kernel cmdline to have a say */ | |
2928 | if (!numentries) { | |
2929 | /* round applicable memory size up to nearest megabyte */ | |
2930 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2931 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2932 | numentries >>= 20 - PAGE_SHIFT; | |
2933 | numentries <<= 20 - PAGE_SHIFT; | |
2934 | ||
2935 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2936 | if (scale > PAGE_SHIFT) | |
2937 | numentries >>= (scale - PAGE_SHIFT); | |
2938 | else | |
2939 | numentries <<= (PAGE_SHIFT - scale); | |
2940 | } | |
6e692ed3 | 2941 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
2942 | |
2943 | /* limit allocation size to 1/16 total memory by default */ | |
2944 | if (max == 0) { | |
2945 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2946 | do_div(max, bucketsize); | |
2947 | } | |
2948 | ||
2949 | if (numentries > max) | |
2950 | numentries = max; | |
2951 | ||
2952 | log2qty = long_log2(numentries); | |
2953 | ||
2954 | do { | |
2955 | size = bucketsize << log2qty; | |
2956 | if (flags & HASH_EARLY) | |
2957 | table = alloc_bootmem(size); | |
2958 | else if (hashdist) | |
2959 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2960 | else { | |
2961 | unsigned long order; | |
2962 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2963 | ; | |
2964 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2965 | } | |
2966 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2967 | ||
2968 | if (!table) | |
2969 | panic("Failed to allocate %s hash table\n", tablename); | |
2970 | ||
2971 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2972 | tablename, | |
2973 | (1U << log2qty), | |
2974 | long_log2(size) - PAGE_SHIFT, | |
2975 | size); | |
2976 | ||
2977 | if (_hash_shift) | |
2978 | *_hash_shift = log2qty; | |
2979 | if (_hash_mask) | |
2980 | *_hash_mask = (1 << log2qty) - 1; | |
2981 | ||
2982 | return table; | |
2983 | } | |
a117e66e KH |
2984 | |
2985 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
2986 | struct page *pfn_to_page(unsigned long pfn) |
2987 | { | |
67de6482 | 2988 | return __pfn_to_page(pfn); |
a117e66e KH |
2989 | } |
2990 | unsigned long page_to_pfn(struct page *page) | |
2991 | { | |
67de6482 | 2992 | return __page_to_pfn(page); |
a117e66e | 2993 | } |
a117e66e KH |
2994 | EXPORT_SYMBOL(pfn_to_page); |
2995 | EXPORT_SYMBOL(page_to_pfn); | |
2996 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |