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> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 LT |
23 | #include <linux/bootmem.h> |
24 | #include <linux/compiler.h> | |
9f158333 | 25 | #include <linux/kernel.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/suspend.h> | |
28 | #include <linux/pagevec.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/slab.h> | |
5a3135c2 | 31 | #include <linux/oom.h> |
1da177e4 LT |
32 | #include <linux/notifier.h> |
33 | #include <linux/topology.h> | |
34 | #include <linux/sysctl.h> | |
35 | #include <linux/cpu.h> | |
36 | #include <linux/cpuset.h> | |
bdc8cb98 | 37 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
38 | #include <linux/nodemask.h> |
39 | #include <linux/vmalloc.h> | |
4be38e35 | 40 | #include <linux/mempolicy.h> |
6811378e | 41 | #include <linux/stop_machine.h> |
c713216d MG |
42 | #include <linux/sort.h> |
43 | #include <linux/pfn.h> | |
3fcfab16 | 44 | #include <linux/backing-dev.h> |
933e312e | 45 | #include <linux/fault-inject.h> |
a5d76b54 | 46 | #include <linux/page-isolation.h> |
8a9f3ccd | 47 | #include <linux/memcontrol.h> |
3ac7fe5a | 48 | #include <linux/debugobjects.h> |
1da177e4 LT |
49 | |
50 | #include <asm/tlbflush.h> | |
ac924c60 | 51 | #include <asm/div64.h> |
1da177e4 LT |
52 | #include "internal.h" |
53 | ||
54 | /* | |
13808910 | 55 | * Array of node states. |
1da177e4 | 56 | */ |
13808910 CL |
57 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
58 | [N_POSSIBLE] = NODE_MASK_ALL, | |
59 | [N_ONLINE] = { { [0] = 1UL } }, | |
60 | #ifndef CONFIG_NUMA | |
61 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
62 | #ifdef CONFIG_HIGHMEM | |
63 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
64 | #endif | |
65 | [N_CPU] = { { [0] = 1UL } }, | |
66 | #endif /* NUMA */ | |
67 | }; | |
68 | EXPORT_SYMBOL(node_states); | |
69 | ||
6c231b7b | 70 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 71 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 72 | long nr_swap_pages; |
8ad4b1fb | 73 | int percpu_pagelist_fraction; |
1da177e4 | 74 | |
d9c23400 MG |
75 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
76 | int pageblock_order __read_mostly; | |
77 | #endif | |
78 | ||
d98c7a09 | 79 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 80 | |
1da177e4 LT |
81 | /* |
82 | * results with 256, 32 in the lowmem_reserve sysctl: | |
83 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
84 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
85 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
86 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
87 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
88 | * |
89 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
90 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 91 | */ |
2f1b6248 | 92 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 93 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 94 | 256, |
4b51d669 | 95 | #endif |
fb0e7942 | 96 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 97 | 256, |
fb0e7942 | 98 | #endif |
e53ef38d | 99 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 100 | 32, |
e53ef38d | 101 | #endif |
2a1e274a | 102 | 32, |
2f1b6248 | 103 | }; |
1da177e4 LT |
104 | |
105 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 106 | |
15ad7cdc | 107 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 108 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 109 | "DMA", |
4b51d669 | 110 | #endif |
fb0e7942 | 111 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 112 | "DMA32", |
fb0e7942 | 113 | #endif |
2f1b6248 | 114 | "Normal", |
e53ef38d | 115 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 116 | "HighMem", |
e53ef38d | 117 | #endif |
2a1e274a | 118 | "Movable", |
2f1b6248 CL |
119 | }; |
120 | ||
1da177e4 LT |
121 | int min_free_kbytes = 1024; |
122 | ||
86356ab1 YG |
123 | unsigned long __meminitdata nr_kernel_pages; |
124 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 125 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 126 | |
c713216d MG |
127 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
128 | /* | |
183ff22b | 129 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
130 | * ranges of memory (RAM) that may be registered with add_active_range(). |
131 | * Ranges passed to add_active_range() will be merged if possible | |
132 | * so the number of times add_active_range() can be called is | |
133 | * related to the number of nodes and the number of holes | |
134 | */ | |
135 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
136 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
137 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
138 | #else | |
139 | #if MAX_NUMNODES >= 32 | |
140 | /* If there can be many nodes, allow up to 50 holes per node */ | |
141 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
142 | #else | |
143 | /* By default, allow up to 256 distinct regions */ | |
144 | #define MAX_ACTIVE_REGIONS 256 | |
145 | #endif | |
146 | #endif | |
147 | ||
98011f56 JB |
148 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
149 | static int __meminitdata nr_nodemap_entries; | |
150 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
151 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c | 152 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
98011f56 JB |
153 | static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES]; |
154 | static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES]; | |
fb01439c | 155 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ |
2a1e274a | 156 | unsigned long __initdata required_kernelcore; |
484f51f8 | 157 | static unsigned long __initdata required_movablecore; |
e228929b | 158 | unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
159 | |
160 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
161 | int movable_zone; | |
162 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
163 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
164 | ||
418508c1 MS |
165 | #if MAX_NUMNODES > 1 |
166 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
167 | EXPORT_SYMBOL(nr_node_ids); | |
168 | #endif | |
169 | ||
9ef9acb0 MG |
170 | int page_group_by_mobility_disabled __read_mostly; |
171 | ||
b2a0ac88 MG |
172 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
173 | { | |
174 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
175 | PB_migrate, PB_migrate_end); | |
176 | } | |
177 | ||
13e7444b | 178 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 179 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 180 | { |
bdc8cb98 DH |
181 | int ret = 0; |
182 | unsigned seq; | |
183 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 184 | |
bdc8cb98 DH |
185 | do { |
186 | seq = zone_span_seqbegin(zone); | |
187 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
188 | ret = 1; | |
189 | else if (pfn < zone->zone_start_pfn) | |
190 | ret = 1; | |
191 | } while (zone_span_seqretry(zone, seq)); | |
192 | ||
193 | return ret; | |
c6a57e19 DH |
194 | } |
195 | ||
196 | static int page_is_consistent(struct zone *zone, struct page *page) | |
197 | { | |
14e07298 | 198 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 199 | return 0; |
1da177e4 | 200 | if (zone != page_zone(page)) |
c6a57e19 DH |
201 | return 0; |
202 | ||
203 | return 1; | |
204 | } | |
205 | /* | |
206 | * Temporary debugging check for pages not lying within a given zone. | |
207 | */ | |
208 | static int bad_range(struct zone *zone, struct page *page) | |
209 | { | |
210 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 211 | return 1; |
c6a57e19 DH |
212 | if (!page_is_consistent(zone, page)) |
213 | return 1; | |
214 | ||
1da177e4 LT |
215 | return 0; |
216 | } | |
13e7444b NP |
217 | #else |
218 | static inline int bad_range(struct zone *zone, struct page *page) | |
219 | { | |
220 | return 0; | |
221 | } | |
222 | #endif | |
223 | ||
224abf92 | 224 | static void bad_page(struct page *page) |
1da177e4 | 225 | { |
9442ec9d HD |
226 | void *pc = page_get_page_cgroup(page); |
227 | ||
228 | printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG | |
229 | "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n", | |
224abf92 NP |
230 | current->comm, page, (int)(2*sizeof(unsigned long)), |
231 | (unsigned long)page->flags, page->mapping, | |
232 | page_mapcount(page), page_count(page)); | |
9442ec9d HD |
233 | if (pc) { |
234 | printk(KERN_EMERG "cgroup:%p\n", pc); | |
235 | page_reset_bad_cgroup(page); | |
236 | } | |
237 | printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
238 | KERN_EMERG "Backtrace:\n"); | |
1da177e4 | 239 | dump_stack(); |
dfa7e20c | 240 | page->flags &= ~PAGE_FLAGS_CLEAR_WHEN_BAD; |
1da177e4 LT |
241 | set_page_count(page, 0); |
242 | reset_page_mapcount(page); | |
243 | page->mapping = NULL; | |
9f158333 | 244 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
245 | } |
246 | ||
1da177e4 LT |
247 | /* |
248 | * Higher-order pages are called "compound pages". They are structured thusly: | |
249 | * | |
250 | * The first PAGE_SIZE page is called the "head page". | |
251 | * | |
252 | * The remaining PAGE_SIZE pages are called "tail pages". | |
253 | * | |
254 | * All pages have PG_compound set. All pages have their ->private pointing at | |
255 | * the head page (even the head page has this). | |
256 | * | |
41d78ba5 HD |
257 | * The first tail page's ->lru.next holds the address of the compound page's |
258 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
259 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 260 | */ |
d98c7a09 HD |
261 | |
262 | static void free_compound_page(struct page *page) | |
263 | { | |
d85f3385 | 264 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
265 | } |
266 | ||
01ad1c08 | 267 | void prep_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
268 | { |
269 | int i; | |
270 | int nr_pages = 1 << order; | |
271 | ||
33f2ef89 | 272 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 273 | set_compound_order(page, order); |
6d777953 | 274 | __SetPageHead(page); |
d85f3385 | 275 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
276 | struct page *p = page + i; |
277 | ||
d85f3385 | 278 | __SetPageTail(p); |
d85f3385 | 279 | p->first_page = page; |
1da177e4 LT |
280 | } |
281 | } | |
282 | ||
283 | static void destroy_compound_page(struct page *page, unsigned long order) | |
284 | { | |
285 | int i; | |
286 | int nr_pages = 1 << order; | |
287 | ||
d85f3385 | 288 | if (unlikely(compound_order(page) != order)) |
224abf92 | 289 | bad_page(page); |
1da177e4 | 290 | |
6d777953 | 291 | if (unlikely(!PageHead(page))) |
d85f3385 | 292 | bad_page(page); |
6d777953 | 293 | __ClearPageHead(page); |
d85f3385 | 294 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
295 | struct page *p = page + i; |
296 | ||
6d777953 | 297 | if (unlikely(!PageTail(p) | |
d85f3385 | 298 | (p->first_page != page))) |
224abf92 | 299 | bad_page(page); |
d85f3385 | 300 | __ClearPageTail(p); |
1da177e4 LT |
301 | } |
302 | } | |
1da177e4 | 303 | |
17cf4406 NP |
304 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
305 | { | |
306 | int i; | |
307 | ||
6626c5d5 AM |
308 | /* |
309 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
310 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
311 | */ | |
725d704e | 312 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
313 | for (i = 0; i < (1 << order); i++) |
314 | clear_highpage(page + i); | |
315 | } | |
316 | ||
6aa3001b AM |
317 | static inline void set_page_order(struct page *page, int order) |
318 | { | |
4c21e2f2 | 319 | set_page_private(page, order); |
676165a8 | 320 | __SetPageBuddy(page); |
1da177e4 LT |
321 | } |
322 | ||
323 | static inline void rmv_page_order(struct page *page) | |
324 | { | |
676165a8 | 325 | __ClearPageBuddy(page); |
4c21e2f2 | 326 | set_page_private(page, 0); |
1da177e4 LT |
327 | } |
328 | ||
329 | /* | |
330 | * Locate the struct page for both the matching buddy in our | |
331 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
332 | * | |
333 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
334 | * the following equation: | |
335 | * B2 = B1 ^ (1 << O) | |
336 | * For example, if the starting buddy (buddy2) is #8 its order | |
337 | * 1 buddy is #10: | |
338 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
339 | * | |
340 | * 2) Any buddy B will have an order O+1 parent P which | |
341 | * satisfies the following equation: | |
342 | * P = B & ~(1 << O) | |
343 | * | |
d6e05edc | 344 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
345 | */ |
346 | static inline struct page * | |
347 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
348 | { | |
349 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
350 | ||
351 | return page + (buddy_idx - page_idx); | |
352 | } | |
353 | ||
354 | static inline unsigned long | |
355 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
356 | { | |
357 | return (page_idx & ~(1 << order)); | |
358 | } | |
359 | ||
360 | /* | |
361 | * This function checks whether a page is free && is the buddy | |
362 | * we can do coalesce a page and its buddy if | |
13e7444b | 363 | * (a) the buddy is not in a hole && |
676165a8 | 364 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
365 | * (c) a page and its buddy have the same order && |
366 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
367 | * |
368 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
369 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 370 | * |
676165a8 | 371 | * For recording page's order, we use page_private(page). |
1da177e4 | 372 | */ |
cb2b95e1 AW |
373 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
374 | int order) | |
1da177e4 | 375 | { |
14e07298 | 376 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 377 | return 0; |
13e7444b | 378 | |
cb2b95e1 AW |
379 | if (page_zone_id(page) != page_zone_id(buddy)) |
380 | return 0; | |
381 | ||
382 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
383 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 384 | return 1; |
676165a8 | 385 | } |
6aa3001b | 386 | return 0; |
1da177e4 LT |
387 | } |
388 | ||
389 | /* | |
390 | * Freeing function for a buddy system allocator. | |
391 | * | |
392 | * The concept of a buddy system is to maintain direct-mapped table | |
393 | * (containing bit values) for memory blocks of various "orders". | |
394 | * The bottom level table contains the map for the smallest allocatable | |
395 | * units of memory (here, pages), and each level above it describes | |
396 | * pairs of units from the levels below, hence, "buddies". | |
397 | * At a high level, all that happens here is marking the table entry | |
398 | * at the bottom level available, and propagating the changes upward | |
399 | * as necessary, plus some accounting needed to play nicely with other | |
400 | * parts of the VM system. | |
401 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 402 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 403 | * order is recorded in page_private(page) field. |
1da177e4 LT |
404 | * So when we are allocating or freeing one, we can derive the state of the |
405 | * other. That is, if we allocate a small block, and both were | |
406 | * free, the remainder of the region must be split into blocks. | |
407 | * If a block is freed, and its buddy is also free, then this | |
408 | * triggers coalescing into a block of larger size. | |
409 | * | |
410 | * -- wli | |
411 | */ | |
412 | ||
48db57f8 | 413 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
414 | struct zone *zone, unsigned int order) |
415 | { | |
416 | unsigned long page_idx; | |
417 | int order_size = 1 << order; | |
b2a0ac88 | 418 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 419 | |
224abf92 | 420 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
421 | destroy_compound_page(page, order); |
422 | ||
423 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
424 | ||
725d704e NP |
425 | VM_BUG_ON(page_idx & (order_size - 1)); |
426 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 427 | |
d23ad423 | 428 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
429 | while (order < MAX_ORDER-1) { |
430 | unsigned long combined_idx; | |
1da177e4 LT |
431 | struct page *buddy; |
432 | ||
1da177e4 | 433 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 434 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 435 | break; |
13e7444b | 436 | |
3c82d0ce | 437 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 438 | list_del(&buddy->lru); |
b2a0ac88 | 439 | zone->free_area[order].nr_free--; |
1da177e4 | 440 | rmv_page_order(buddy); |
13e7444b | 441 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
442 | page = page + (combined_idx - page_idx); |
443 | page_idx = combined_idx; | |
444 | order++; | |
445 | } | |
446 | set_page_order(page, order); | |
b2a0ac88 MG |
447 | list_add(&page->lru, |
448 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
449 | zone->free_area[order].nr_free++; |
450 | } | |
451 | ||
224abf92 | 452 | static inline int free_pages_check(struct page *page) |
1da177e4 | 453 | { |
92be2e33 NP |
454 | if (unlikely(page_mapcount(page) | |
455 | (page->mapping != NULL) | | |
9442ec9d | 456 | (page_get_page_cgroup(page) != NULL) | |
92be2e33 | 457 | (page_count(page) != 0) | |
dfa7e20c | 458 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) |
224abf92 | 459 | bad_page(page); |
1da177e4 | 460 | if (PageDirty(page)) |
242e5468 | 461 | __ClearPageDirty(page); |
689bcebf HD |
462 | /* |
463 | * For now, we report if PG_reserved was found set, but do not | |
464 | * clear it, and do not free the page. But we shall soon need | |
465 | * to do more, for when the ZERO_PAGE count wraps negative. | |
466 | */ | |
467 | return PageReserved(page); | |
1da177e4 LT |
468 | } |
469 | ||
470 | /* | |
471 | * Frees a list of pages. | |
472 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 473 | * count is the number of pages to free. |
1da177e4 LT |
474 | * |
475 | * If the zone was previously in an "all pages pinned" state then look to | |
476 | * see if this freeing clears that state. | |
477 | * | |
478 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
479 | * pinned" detection logic. | |
480 | */ | |
48db57f8 NP |
481 | static void free_pages_bulk(struct zone *zone, int count, |
482 | struct list_head *list, int order) | |
1da177e4 | 483 | { |
c54ad30c | 484 | spin_lock(&zone->lock); |
e815af95 | 485 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 486 | zone->pages_scanned = 0; |
48db57f8 NP |
487 | while (count--) { |
488 | struct page *page; | |
489 | ||
725d704e | 490 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 491 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 492 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 493 | list_del(&page->lru); |
48db57f8 | 494 | __free_one_page(page, zone, order); |
1da177e4 | 495 | } |
c54ad30c | 496 | spin_unlock(&zone->lock); |
1da177e4 LT |
497 | } |
498 | ||
48db57f8 | 499 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 500 | { |
006d22d9 | 501 | spin_lock(&zone->lock); |
e815af95 | 502 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 503 | zone->pages_scanned = 0; |
0798e519 | 504 | __free_one_page(page, zone, order); |
006d22d9 | 505 | spin_unlock(&zone->lock); |
48db57f8 NP |
506 | } |
507 | ||
508 | static void __free_pages_ok(struct page *page, unsigned int order) | |
509 | { | |
510 | unsigned long flags; | |
1da177e4 | 511 | int i; |
689bcebf | 512 | int reserved = 0; |
1da177e4 | 513 | |
1da177e4 | 514 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 515 | reserved += free_pages_check(page + i); |
689bcebf HD |
516 | if (reserved) |
517 | return; | |
518 | ||
3ac7fe5a | 519 | if (!PageHighMem(page)) { |
9858db50 | 520 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
521 | debug_check_no_obj_freed(page_address(page), |
522 | PAGE_SIZE << order); | |
523 | } | |
dafb1367 | 524 | arch_free_page(page, order); |
48db57f8 | 525 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 526 | |
c54ad30c | 527 | local_irq_save(flags); |
f8891e5e | 528 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 529 | free_one_page(page_zone(page), page, order); |
c54ad30c | 530 | local_irq_restore(flags); |
1da177e4 LT |
531 | } |
532 | ||
a226f6c8 DH |
533 | /* |
534 | * permit the bootmem allocator to evade page validation on high-order frees | |
535 | */ | |
0c0a4a51 | 536 | void __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
537 | { |
538 | if (order == 0) { | |
539 | __ClearPageReserved(page); | |
540 | set_page_count(page, 0); | |
7835e98b | 541 | set_page_refcounted(page); |
545b1ea9 | 542 | __free_page(page); |
a226f6c8 | 543 | } else { |
a226f6c8 DH |
544 | int loop; |
545 | ||
545b1ea9 | 546 | prefetchw(page); |
a226f6c8 DH |
547 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
548 | struct page *p = &page[loop]; | |
549 | ||
545b1ea9 NP |
550 | if (loop + 1 < BITS_PER_LONG) |
551 | prefetchw(p + 1); | |
a226f6c8 DH |
552 | __ClearPageReserved(p); |
553 | set_page_count(p, 0); | |
554 | } | |
555 | ||
7835e98b | 556 | set_page_refcounted(page); |
545b1ea9 | 557 | __free_pages(page, order); |
a226f6c8 DH |
558 | } |
559 | } | |
560 | ||
1da177e4 LT |
561 | |
562 | /* | |
563 | * The order of subdivision here is critical for the IO subsystem. | |
564 | * Please do not alter this order without good reasons and regression | |
565 | * testing. Specifically, as large blocks of memory are subdivided, | |
566 | * the order in which smaller blocks are delivered depends on the order | |
567 | * they're subdivided in this function. This is the primary factor | |
568 | * influencing the order in which pages are delivered to the IO | |
569 | * subsystem according to empirical testing, and this is also justified | |
570 | * by considering the behavior of a buddy system containing a single | |
571 | * large block of memory acted on by a series of small allocations. | |
572 | * This behavior is a critical factor in sglist merging's success. | |
573 | * | |
574 | * -- wli | |
575 | */ | |
085cc7d5 | 576 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
577 | int low, int high, struct free_area *area, |
578 | int migratetype) | |
1da177e4 LT |
579 | { |
580 | unsigned long size = 1 << high; | |
581 | ||
582 | while (high > low) { | |
583 | area--; | |
584 | high--; | |
585 | size >>= 1; | |
725d704e | 586 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 587 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
588 | area->nr_free++; |
589 | set_page_order(&page[size], high); | |
590 | } | |
1da177e4 LT |
591 | } |
592 | ||
1da177e4 LT |
593 | /* |
594 | * This page is about to be returned from the page allocator | |
595 | */ | |
17cf4406 | 596 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 597 | { |
92be2e33 NP |
598 | if (unlikely(page_mapcount(page) | |
599 | (page->mapping != NULL) | | |
9442ec9d | 600 | (page_get_page_cgroup(page) != NULL) | |
92be2e33 | 601 | (page_count(page) != 0) | |
dfa7e20c | 602 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) |
224abf92 | 603 | bad_page(page); |
1da177e4 | 604 | |
689bcebf HD |
605 | /* |
606 | * For now, we report if PG_reserved was found set, but do not | |
607 | * clear it, and do not allocate the page: as a safety net. | |
608 | */ | |
609 | if (PageReserved(page)) | |
610 | return 1; | |
611 | ||
0a128b2b | 612 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_reclaim | |
1da177e4 | 613 | 1 << PG_referenced | 1 << PG_arch_1 | |
5409bae0 | 614 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 615 | set_page_private(page, 0); |
7835e98b | 616 | set_page_refcounted(page); |
cc102509 NP |
617 | |
618 | arch_alloc_page(page, order); | |
1da177e4 | 619 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
620 | |
621 | if (gfp_flags & __GFP_ZERO) | |
622 | prep_zero_page(page, order, gfp_flags); | |
623 | ||
624 | if (order && (gfp_flags & __GFP_COMP)) | |
625 | prep_compound_page(page, order); | |
626 | ||
689bcebf | 627 | return 0; |
1da177e4 LT |
628 | } |
629 | ||
56fd56b8 MG |
630 | /* |
631 | * Go through the free lists for the given migratetype and remove | |
632 | * the smallest available page from the freelists | |
633 | */ | |
634 | static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
635 | int migratetype) | |
636 | { | |
637 | unsigned int current_order; | |
638 | struct free_area * area; | |
639 | struct page *page; | |
640 | ||
641 | /* Find a page of the appropriate size in the preferred list */ | |
642 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
643 | area = &(zone->free_area[current_order]); | |
644 | if (list_empty(&area->free_list[migratetype])) | |
645 | continue; | |
646 | ||
647 | page = list_entry(area->free_list[migratetype].next, | |
648 | struct page, lru); | |
649 | list_del(&page->lru); | |
650 | rmv_page_order(page); | |
651 | area->nr_free--; | |
652 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); | |
653 | expand(zone, page, order, current_order, area, migratetype); | |
654 | return page; | |
655 | } | |
656 | ||
657 | return NULL; | |
658 | } | |
659 | ||
660 | ||
b2a0ac88 MG |
661 | /* |
662 | * This array describes the order lists are fallen back to when | |
663 | * the free lists for the desirable migrate type are depleted | |
664 | */ | |
665 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
666 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
667 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
668 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
669 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
670 | }; |
671 | ||
c361be55 MG |
672 | /* |
673 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 674 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
675 | * boundary. If alignment is required, use move_freepages_block() |
676 | */ | |
677 | int move_freepages(struct zone *zone, | |
678 | struct page *start_page, struct page *end_page, | |
679 | int migratetype) | |
680 | { | |
681 | struct page *page; | |
682 | unsigned long order; | |
d100313f | 683 | int pages_moved = 0; |
c361be55 MG |
684 | |
685 | #ifndef CONFIG_HOLES_IN_ZONE | |
686 | /* | |
687 | * page_zone is not safe to call in this context when | |
688 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
689 | * anyway as we check zone boundaries in move_freepages_block(). | |
690 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 691 | * grouping pages by mobility |
c361be55 MG |
692 | */ |
693 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
694 | #endif | |
695 | ||
696 | for (page = start_page; page <= end_page;) { | |
697 | if (!pfn_valid_within(page_to_pfn(page))) { | |
698 | page++; | |
699 | continue; | |
700 | } | |
701 | ||
702 | if (!PageBuddy(page)) { | |
703 | page++; | |
704 | continue; | |
705 | } | |
706 | ||
707 | order = page_order(page); | |
708 | list_del(&page->lru); | |
709 | list_add(&page->lru, | |
710 | &zone->free_area[order].free_list[migratetype]); | |
711 | page += 1 << order; | |
d100313f | 712 | pages_moved += 1 << order; |
c361be55 MG |
713 | } |
714 | ||
d100313f | 715 | return pages_moved; |
c361be55 MG |
716 | } |
717 | ||
718 | int move_freepages_block(struct zone *zone, struct page *page, int migratetype) | |
719 | { | |
720 | unsigned long start_pfn, end_pfn; | |
721 | struct page *start_page, *end_page; | |
722 | ||
723 | start_pfn = page_to_pfn(page); | |
d9c23400 | 724 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 725 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
726 | end_page = start_page + pageblock_nr_pages - 1; |
727 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
728 | |
729 | /* Do not cross zone boundaries */ | |
730 | if (start_pfn < zone->zone_start_pfn) | |
731 | start_page = page; | |
732 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
733 | return 0; | |
734 | ||
735 | return move_freepages(zone, start_page, end_page, migratetype); | |
736 | } | |
737 | ||
b2a0ac88 MG |
738 | /* Remove an element from the buddy allocator from the fallback list */ |
739 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
740 | int start_migratetype) | |
741 | { | |
742 | struct free_area * area; | |
743 | int current_order; | |
744 | struct page *page; | |
745 | int migratetype, i; | |
746 | ||
747 | /* Find the largest possible block of pages in the other list */ | |
748 | for (current_order = MAX_ORDER-1; current_order >= order; | |
749 | --current_order) { | |
750 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
751 | migratetype = fallbacks[start_migratetype][i]; | |
752 | ||
56fd56b8 MG |
753 | /* MIGRATE_RESERVE handled later if necessary */ |
754 | if (migratetype == MIGRATE_RESERVE) | |
755 | continue; | |
e010487d | 756 | |
b2a0ac88 MG |
757 | area = &(zone->free_area[current_order]); |
758 | if (list_empty(&area->free_list[migratetype])) | |
759 | continue; | |
760 | ||
761 | page = list_entry(area->free_list[migratetype].next, | |
762 | struct page, lru); | |
763 | area->nr_free--; | |
764 | ||
765 | /* | |
c361be55 | 766 | * If breaking a large block of pages, move all free |
46dafbca MG |
767 | * pages to the preferred allocation list. If falling |
768 | * back for a reclaimable kernel allocation, be more | |
769 | * agressive about taking ownership of free pages | |
b2a0ac88 | 770 | */ |
d9c23400 | 771 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
46dafbca MG |
772 | start_migratetype == MIGRATE_RECLAIMABLE) { |
773 | unsigned long pages; | |
774 | pages = move_freepages_block(zone, page, | |
775 | start_migratetype); | |
776 | ||
777 | /* Claim the whole block if over half of it is free */ | |
d9c23400 | 778 | if (pages >= (1 << (pageblock_order-1))) |
46dafbca MG |
779 | set_pageblock_migratetype(page, |
780 | start_migratetype); | |
781 | ||
b2a0ac88 | 782 | migratetype = start_migratetype; |
c361be55 | 783 | } |
b2a0ac88 MG |
784 | |
785 | /* Remove the page from the freelists */ | |
786 | list_del(&page->lru); | |
787 | rmv_page_order(page); | |
788 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
789 | -(1UL << order)); | |
790 | ||
d9c23400 | 791 | if (current_order == pageblock_order) |
b2a0ac88 MG |
792 | set_pageblock_migratetype(page, |
793 | start_migratetype); | |
794 | ||
795 | expand(zone, page, order, current_order, area, migratetype); | |
796 | return page; | |
797 | } | |
798 | } | |
799 | ||
56fd56b8 MG |
800 | /* Use MIGRATE_RESERVE rather than fail an allocation */ |
801 | return __rmqueue_smallest(zone, order, MIGRATE_RESERVE); | |
b2a0ac88 MG |
802 | } |
803 | ||
56fd56b8 | 804 | /* |
1da177e4 LT |
805 | * Do the hard work of removing an element from the buddy allocator. |
806 | * Call me with the zone->lock already held. | |
807 | */ | |
b2a0ac88 MG |
808 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
809 | int migratetype) | |
1da177e4 | 810 | { |
1da177e4 LT |
811 | struct page *page; |
812 | ||
56fd56b8 | 813 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 814 | |
56fd56b8 MG |
815 | if (unlikely(!page)) |
816 | page = __rmqueue_fallback(zone, order, migratetype); | |
b2a0ac88 MG |
817 | |
818 | return page; | |
1da177e4 LT |
819 | } |
820 | ||
821 | /* | |
822 | * Obtain a specified number of elements from the buddy allocator, all under | |
823 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
824 | * Returns the number of new pages which were placed at *list. | |
825 | */ | |
826 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
827 | unsigned long count, struct list_head *list, |
828 | int migratetype) | |
1da177e4 | 829 | { |
1da177e4 | 830 | int i; |
1da177e4 | 831 | |
c54ad30c | 832 | spin_lock(&zone->lock); |
1da177e4 | 833 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 834 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 835 | if (unlikely(page == NULL)) |
1da177e4 | 836 | break; |
81eabcbe MG |
837 | |
838 | /* | |
839 | * Split buddy pages returned by expand() are received here | |
840 | * in physical page order. The page is added to the callers and | |
841 | * list and the list head then moves forward. From the callers | |
842 | * perspective, the linked list is ordered by page number in | |
843 | * some conditions. This is useful for IO devices that can | |
844 | * merge IO requests if the physical pages are ordered | |
845 | * properly. | |
846 | */ | |
535131e6 MG |
847 | list_add(&page->lru, list); |
848 | set_page_private(page, migratetype); | |
81eabcbe | 849 | list = &page->lru; |
1da177e4 | 850 | } |
c54ad30c | 851 | spin_unlock(&zone->lock); |
085cc7d5 | 852 | return i; |
1da177e4 LT |
853 | } |
854 | ||
4ae7c039 | 855 | #ifdef CONFIG_NUMA |
8fce4d8e | 856 | /* |
4037d452 CL |
857 | * Called from the vmstat counter updater to drain pagesets of this |
858 | * currently executing processor on remote nodes after they have | |
859 | * expired. | |
860 | * | |
879336c3 CL |
861 | * Note that this function must be called with the thread pinned to |
862 | * a single processor. | |
8fce4d8e | 863 | */ |
4037d452 | 864 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 865 | { |
4ae7c039 | 866 | unsigned long flags; |
4037d452 | 867 | int to_drain; |
4ae7c039 | 868 | |
4037d452 CL |
869 | local_irq_save(flags); |
870 | if (pcp->count >= pcp->batch) | |
871 | to_drain = pcp->batch; | |
872 | else | |
873 | to_drain = pcp->count; | |
874 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
875 | pcp->count -= to_drain; | |
876 | local_irq_restore(flags); | |
4ae7c039 CL |
877 | } |
878 | #endif | |
879 | ||
9f8f2172 CL |
880 | /* |
881 | * Drain pages of the indicated processor. | |
882 | * | |
883 | * The processor must either be the current processor and the | |
884 | * thread pinned to the current processor or a processor that | |
885 | * is not online. | |
886 | */ | |
887 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 888 | { |
c54ad30c | 889 | unsigned long flags; |
1da177e4 | 890 | struct zone *zone; |
1da177e4 LT |
891 | |
892 | for_each_zone(zone) { | |
893 | struct per_cpu_pageset *pset; | |
3dfa5721 | 894 | struct per_cpu_pages *pcp; |
1da177e4 | 895 | |
f2e12bb2 CL |
896 | if (!populated_zone(zone)) |
897 | continue; | |
898 | ||
e7c8d5c9 | 899 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
900 | |
901 | pcp = &pset->pcp; | |
902 | local_irq_save(flags); | |
903 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
904 | pcp->count = 0; | |
905 | local_irq_restore(flags); | |
1da177e4 LT |
906 | } |
907 | } | |
1da177e4 | 908 | |
9f8f2172 CL |
909 | /* |
910 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
911 | */ | |
912 | void drain_local_pages(void *arg) | |
913 | { | |
914 | drain_pages(smp_processor_id()); | |
915 | } | |
916 | ||
917 | /* | |
918 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
919 | */ | |
920 | void drain_all_pages(void) | |
921 | { | |
15c8b6c1 | 922 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
923 | } |
924 | ||
296699de | 925 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
926 | |
927 | void mark_free_pages(struct zone *zone) | |
928 | { | |
f623f0db RW |
929 | unsigned long pfn, max_zone_pfn; |
930 | unsigned long flags; | |
b2a0ac88 | 931 | int order, t; |
1da177e4 LT |
932 | struct list_head *curr; |
933 | ||
934 | if (!zone->spanned_pages) | |
935 | return; | |
936 | ||
937 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
938 | |
939 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
940 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
941 | if (pfn_valid(pfn)) { | |
942 | struct page *page = pfn_to_page(pfn); | |
943 | ||
7be98234 RW |
944 | if (!swsusp_page_is_forbidden(page)) |
945 | swsusp_unset_page_free(page); | |
f623f0db | 946 | } |
1da177e4 | 947 | |
b2a0ac88 MG |
948 | for_each_migratetype_order(order, t) { |
949 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 950 | unsigned long i; |
1da177e4 | 951 | |
f623f0db RW |
952 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
953 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 954 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 955 | } |
b2a0ac88 | 956 | } |
1da177e4 LT |
957 | spin_unlock_irqrestore(&zone->lock, flags); |
958 | } | |
e2c55dc8 | 959 | #endif /* CONFIG_PM */ |
1da177e4 | 960 | |
1da177e4 LT |
961 | /* |
962 | * Free a 0-order page | |
963 | */ | |
920c7a5d | 964 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
965 | { |
966 | struct zone *zone = page_zone(page); | |
967 | struct per_cpu_pages *pcp; | |
968 | unsigned long flags; | |
969 | ||
1da177e4 LT |
970 | if (PageAnon(page)) |
971 | page->mapping = NULL; | |
224abf92 | 972 | if (free_pages_check(page)) |
689bcebf HD |
973 | return; |
974 | ||
3ac7fe5a | 975 | if (!PageHighMem(page)) { |
9858db50 | 976 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); |
3ac7fe5a TG |
977 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); |
978 | } | |
dafb1367 | 979 | arch_free_page(page, 0); |
689bcebf HD |
980 | kernel_map_pages(page, 1, 0); |
981 | ||
3dfa5721 | 982 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
1da177e4 | 983 | local_irq_save(flags); |
f8891e5e | 984 | __count_vm_event(PGFREE); |
3dfa5721 CL |
985 | if (cold) |
986 | list_add_tail(&page->lru, &pcp->list); | |
987 | else | |
988 | list_add(&page->lru, &pcp->list); | |
535131e6 | 989 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 990 | pcp->count++; |
48db57f8 NP |
991 | if (pcp->count >= pcp->high) { |
992 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
993 | pcp->count -= pcp->batch; | |
994 | } | |
1da177e4 LT |
995 | local_irq_restore(flags); |
996 | put_cpu(); | |
997 | } | |
998 | ||
920c7a5d | 999 | void free_hot_page(struct page *page) |
1da177e4 LT |
1000 | { |
1001 | free_hot_cold_page(page, 0); | |
1002 | } | |
1003 | ||
920c7a5d | 1004 | void free_cold_page(struct page *page) |
1da177e4 LT |
1005 | { |
1006 | free_hot_cold_page(page, 1); | |
1007 | } | |
1008 | ||
8dfcc9ba NP |
1009 | /* |
1010 | * split_page takes a non-compound higher-order page, and splits it into | |
1011 | * n (1<<order) sub-pages: page[0..n] | |
1012 | * Each sub-page must be freed individually. | |
1013 | * | |
1014 | * Note: this is probably too low level an operation for use in drivers. | |
1015 | * Please consult with lkml before using this in your driver. | |
1016 | */ | |
1017 | void split_page(struct page *page, unsigned int order) | |
1018 | { | |
1019 | int i; | |
1020 | ||
725d704e NP |
1021 | VM_BUG_ON(PageCompound(page)); |
1022 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
1023 | for (i = 1; i < (1 << order); i++) |
1024 | set_page_refcounted(page + i); | |
8dfcc9ba | 1025 | } |
8dfcc9ba | 1026 | |
1da177e4 LT |
1027 | /* |
1028 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1029 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1030 | * or two. | |
1031 | */ | |
18ea7e71 | 1032 | static struct page *buffered_rmqueue(struct zone *preferred_zone, |
a74609fa | 1033 | struct zone *zone, int order, gfp_t gfp_flags) |
1da177e4 LT |
1034 | { |
1035 | unsigned long flags; | |
689bcebf | 1036 | struct page *page; |
1da177e4 | 1037 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1038 | int cpu; |
64c5e135 | 1039 | int migratetype = allocflags_to_migratetype(gfp_flags); |
1da177e4 | 1040 | |
689bcebf | 1041 | again: |
a74609fa | 1042 | cpu = get_cpu(); |
48db57f8 | 1043 | if (likely(order == 0)) { |
1da177e4 LT |
1044 | struct per_cpu_pages *pcp; |
1045 | ||
3dfa5721 | 1046 | pcp = &zone_pcp(zone, cpu)->pcp; |
1da177e4 | 1047 | local_irq_save(flags); |
a74609fa | 1048 | if (!pcp->count) { |
941c7105 | 1049 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 1050 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
1051 | if (unlikely(!pcp->count)) |
1052 | goto failed; | |
1da177e4 | 1053 | } |
b92a6edd | 1054 | |
535131e6 | 1055 | /* Find a page of the appropriate migrate type */ |
3dfa5721 CL |
1056 | if (cold) { |
1057 | list_for_each_entry_reverse(page, &pcp->list, lru) | |
1058 | if (page_private(page) == migratetype) | |
1059 | break; | |
1060 | } else { | |
1061 | list_for_each_entry(page, &pcp->list, lru) | |
1062 | if (page_private(page) == migratetype) | |
1063 | break; | |
1064 | } | |
535131e6 | 1065 | |
b92a6edd MG |
1066 | /* Allocate more to the pcp list if necessary */ |
1067 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 MG |
1068 | pcp->count += rmqueue_bulk(zone, 0, |
1069 | pcp->batch, &pcp->list, migratetype); | |
1070 | page = list_entry(pcp->list.next, struct page, lru); | |
535131e6 | 1071 | } |
b92a6edd MG |
1072 | |
1073 | list_del(&page->lru); | |
1074 | pcp->count--; | |
7fb1d9fc | 1075 | } else { |
1da177e4 | 1076 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1077 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1078 | spin_unlock(&zone->lock); |
1079 | if (!page) | |
1080 | goto failed; | |
1da177e4 LT |
1081 | } |
1082 | ||
f8891e5e | 1083 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1084 | zone_statistics(preferred_zone, zone); |
a74609fa NP |
1085 | local_irq_restore(flags); |
1086 | put_cpu(); | |
1da177e4 | 1087 | |
725d704e | 1088 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1089 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1090 | goto again; |
1da177e4 | 1091 | return page; |
a74609fa NP |
1092 | |
1093 | failed: | |
1094 | local_irq_restore(flags); | |
1095 | put_cpu(); | |
1096 | return NULL; | |
1da177e4 LT |
1097 | } |
1098 | ||
7fb1d9fc | 1099 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
1100 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
1101 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
1102 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
1103 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1104 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1105 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1106 | |
933e312e AM |
1107 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1108 | ||
1109 | static struct fail_page_alloc_attr { | |
1110 | struct fault_attr attr; | |
1111 | ||
1112 | u32 ignore_gfp_highmem; | |
1113 | u32 ignore_gfp_wait; | |
54114994 | 1114 | u32 min_order; |
933e312e AM |
1115 | |
1116 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1117 | ||
1118 | struct dentry *ignore_gfp_highmem_file; | |
1119 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1120 | struct dentry *min_order_file; |
933e312e AM |
1121 | |
1122 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1123 | ||
1124 | } fail_page_alloc = { | |
1125 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1126 | .ignore_gfp_wait = 1, |
1127 | .ignore_gfp_highmem = 1, | |
54114994 | 1128 | .min_order = 1, |
933e312e AM |
1129 | }; |
1130 | ||
1131 | static int __init setup_fail_page_alloc(char *str) | |
1132 | { | |
1133 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1134 | } | |
1135 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1136 | ||
1137 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1138 | { | |
54114994 AM |
1139 | if (order < fail_page_alloc.min_order) |
1140 | return 0; | |
933e312e AM |
1141 | if (gfp_mask & __GFP_NOFAIL) |
1142 | return 0; | |
1143 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1144 | return 0; | |
1145 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1146 | return 0; | |
1147 | ||
1148 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1149 | } | |
1150 | ||
1151 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1152 | ||
1153 | static int __init fail_page_alloc_debugfs(void) | |
1154 | { | |
1155 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1156 | struct dentry *dir; | |
1157 | int err; | |
1158 | ||
1159 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1160 | "fail_page_alloc"); | |
1161 | if (err) | |
1162 | return err; | |
1163 | dir = fail_page_alloc.attr.dentries.dir; | |
1164 | ||
1165 | fail_page_alloc.ignore_gfp_wait_file = | |
1166 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1167 | &fail_page_alloc.ignore_gfp_wait); | |
1168 | ||
1169 | fail_page_alloc.ignore_gfp_highmem_file = | |
1170 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1171 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1172 | fail_page_alloc.min_order_file = |
1173 | debugfs_create_u32("min-order", mode, dir, | |
1174 | &fail_page_alloc.min_order); | |
933e312e AM |
1175 | |
1176 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1177 | !fail_page_alloc.ignore_gfp_highmem_file || |
1178 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1179 | err = -ENOMEM; |
1180 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1181 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1182 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1183 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1184 | } | |
1185 | ||
1186 | return err; | |
1187 | } | |
1188 | ||
1189 | late_initcall(fail_page_alloc_debugfs); | |
1190 | ||
1191 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1192 | ||
1193 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1194 | ||
1195 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1196 | { | |
1197 | return 0; | |
1198 | } | |
1199 | ||
1200 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1201 | ||
1da177e4 LT |
1202 | /* |
1203 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1204 | * of the allocation. | |
1205 | */ | |
1206 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1207 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1208 | { |
1209 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1210 | long min = mark; |
1211 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1212 | int o; |
1213 | ||
7fb1d9fc | 1214 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1215 | min -= min / 2; |
7fb1d9fc | 1216 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1217 | min -= min / 4; |
1218 | ||
1219 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1220 | return 0; | |
1221 | for (o = 0; o < order; o++) { | |
1222 | /* At the next order, this order's pages become unavailable */ | |
1223 | free_pages -= z->free_area[o].nr_free << o; | |
1224 | ||
1225 | /* Require fewer higher order pages to be free */ | |
1226 | min >>= 1; | |
1227 | ||
1228 | if (free_pages <= min) | |
1229 | return 0; | |
1230 | } | |
1231 | return 1; | |
1232 | } | |
1233 | ||
9276b1bc PJ |
1234 | #ifdef CONFIG_NUMA |
1235 | /* | |
1236 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1237 | * skip over zones that are not allowed by the cpuset, or that have | |
1238 | * been recently (in last second) found to be nearly full. See further | |
1239 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1240 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1241 | * |
1242 | * If the zonelist cache is present in the passed in zonelist, then | |
1243 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1244 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1245 | * |
1246 | * If the zonelist cache is not available for this zonelist, does | |
1247 | * nothing and returns NULL. | |
1248 | * | |
1249 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1250 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1251 | * | |
1252 | * We hold off even calling zlc_setup, until after we've checked the | |
1253 | * first zone in the zonelist, on the theory that most allocations will | |
1254 | * be satisfied from that first zone, so best to examine that zone as | |
1255 | * quickly as we can. | |
1256 | */ | |
1257 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1258 | { | |
1259 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1260 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1261 | ||
1262 | zlc = zonelist->zlcache_ptr; | |
1263 | if (!zlc) | |
1264 | return NULL; | |
1265 | ||
f05111f5 | 1266 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1267 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1268 | zlc->last_full_zap = jiffies; | |
1269 | } | |
1270 | ||
1271 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1272 | &cpuset_current_mems_allowed : | |
37b07e41 | 1273 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1274 | return allowednodes; |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1279 | * if it is worth looking at further for free memory: | |
1280 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1281 | * bit set in the zonelist_cache fullzones BITMAP). | |
1282 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1283 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1284 | * Return true (non-zero) if zone is worth looking at further, or | |
1285 | * else return false (zero) if it is not. | |
1286 | * | |
1287 | * This check -ignores- the distinction between various watermarks, | |
1288 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1289 | * found to be full for any variation of these watermarks, it will | |
1290 | * be considered full for up to one second by all requests, unless | |
1291 | * we are so low on memory on all allowed nodes that we are forced | |
1292 | * into the second scan of the zonelist. | |
1293 | * | |
1294 | * In the second scan we ignore this zonelist cache and exactly | |
1295 | * apply the watermarks to all zones, even it is slower to do so. | |
1296 | * We are low on memory in the second scan, and should leave no stone | |
1297 | * unturned looking for a free page. | |
1298 | */ | |
dd1a239f | 1299 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1300 | nodemask_t *allowednodes) |
1301 | { | |
1302 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1303 | int i; /* index of *z in zonelist zones */ | |
1304 | int n; /* node that zone *z is on */ | |
1305 | ||
1306 | zlc = zonelist->zlcache_ptr; | |
1307 | if (!zlc) | |
1308 | return 1; | |
1309 | ||
dd1a239f | 1310 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1311 | n = zlc->z_to_n[i]; |
1312 | ||
1313 | /* This zone is worth trying if it is allowed but not full */ | |
1314 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1319 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1320 | * from that zone don't waste time re-examining it. | |
1321 | */ | |
dd1a239f | 1322 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1323 | { |
1324 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1325 | int i; /* index of *z in zonelist zones */ | |
1326 | ||
1327 | zlc = zonelist->zlcache_ptr; | |
1328 | if (!zlc) | |
1329 | return; | |
1330 | ||
dd1a239f | 1331 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1332 | |
1333 | set_bit(i, zlc->fullzones); | |
1334 | } | |
1335 | ||
1336 | #else /* CONFIG_NUMA */ | |
1337 | ||
1338 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1339 | { | |
1340 | return NULL; | |
1341 | } | |
1342 | ||
dd1a239f | 1343 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1344 | nodemask_t *allowednodes) |
1345 | { | |
1346 | return 1; | |
1347 | } | |
1348 | ||
dd1a239f | 1349 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1350 | { |
1351 | } | |
1352 | #endif /* CONFIG_NUMA */ | |
1353 | ||
7fb1d9fc | 1354 | /* |
0798e519 | 1355 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1356 | * a page. |
1357 | */ | |
1358 | static struct page * | |
19770b32 | 1359 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
54a6eb5c | 1360 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags) |
753ee728 | 1361 | { |
dd1a239f | 1362 | struct zoneref *z; |
7fb1d9fc | 1363 | struct page *page = NULL; |
54a6eb5c | 1364 | int classzone_idx; |
18ea7e71 | 1365 | struct zone *zone, *preferred_zone; |
9276b1bc PJ |
1366 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1367 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1368 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1369 | |
19770b32 MG |
1370 | (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask, |
1371 | &preferred_zone); | |
7eb54824 AW |
1372 | if (!preferred_zone) |
1373 | return NULL; | |
1374 | ||
19770b32 | 1375 | classzone_idx = zone_idx(preferred_zone); |
7fb1d9fc | 1376 | |
9276b1bc | 1377 | zonelist_scan: |
7fb1d9fc | 1378 | /* |
9276b1bc | 1379 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1380 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1381 | */ | |
19770b32 MG |
1382 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1383 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1384 | if (NUMA_BUILD && zlc_active && |
1385 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1386 | continue; | |
7fb1d9fc | 1387 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1388 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1389 | goto try_next_zone; |
7fb1d9fc RS |
1390 | |
1391 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1392 | unsigned long mark; |
1393 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1394 | mark = zone->pages_min; |
3148890b | 1395 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1396 | mark = zone->pages_low; |
3148890b | 1397 | else |
1192d526 | 1398 | mark = zone->pages_high; |
0798e519 PJ |
1399 | if (!zone_watermark_ok(zone, order, mark, |
1400 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1401 | if (!zone_reclaim_mode || |
1192d526 | 1402 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1403 | goto this_zone_full; |
0798e519 | 1404 | } |
7fb1d9fc RS |
1405 | } |
1406 | ||
18ea7e71 | 1407 | page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask); |
0798e519 | 1408 | if (page) |
7fb1d9fc | 1409 | break; |
9276b1bc PJ |
1410 | this_zone_full: |
1411 | if (NUMA_BUILD) | |
1412 | zlc_mark_zone_full(zonelist, z); | |
1413 | try_next_zone: | |
1414 | if (NUMA_BUILD && !did_zlc_setup) { | |
1415 | /* we do zlc_setup after the first zone is tried */ | |
1416 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1417 | zlc_active = 1; | |
1418 | did_zlc_setup = 1; | |
1419 | } | |
54a6eb5c | 1420 | } |
9276b1bc PJ |
1421 | |
1422 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1423 | /* Disable zlc cache for second zonelist scan */ | |
1424 | zlc_active = 0; | |
1425 | goto zonelist_scan; | |
1426 | } | |
7fb1d9fc | 1427 | return page; |
753ee728 MH |
1428 | } |
1429 | ||
1da177e4 LT |
1430 | /* |
1431 | * This is the 'heart' of the zoned buddy allocator. | |
1432 | */ | |
e4048e5d | 1433 | struct page * |
19770b32 MG |
1434 | __alloc_pages_internal(gfp_t gfp_mask, unsigned int order, |
1435 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1da177e4 | 1436 | { |
260b2367 | 1437 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
54a6eb5c | 1438 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
dd1a239f MG |
1439 | struct zoneref *z; |
1440 | struct zone *zone; | |
1da177e4 LT |
1441 | struct page *page; |
1442 | struct reclaim_state reclaim_state; | |
1443 | struct task_struct *p = current; | |
1da177e4 | 1444 | int do_retry; |
7fb1d9fc | 1445 | int alloc_flags; |
a41f24ea NA |
1446 | unsigned long did_some_progress; |
1447 | unsigned long pages_reclaimed = 0; | |
1da177e4 LT |
1448 | |
1449 | might_sleep_if(wait); | |
1450 | ||
933e312e AM |
1451 | if (should_fail_alloc_page(gfp_mask, order)) |
1452 | return NULL; | |
1453 | ||
6b1de916 | 1454 | restart: |
dd1a239f | 1455 | z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1456 | |
dd1a239f | 1457 | if (unlikely(!z->zone)) { |
523b9458 CL |
1458 | /* |
1459 | * Happens if we have an empty zonelist as a result of | |
1460 | * GFP_THISNODE being used on a memoryless node | |
1461 | */ | |
1da177e4 LT |
1462 | return NULL; |
1463 | } | |
6b1de916 | 1464 | |
19770b32 | 1465 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
54a6eb5c | 1466 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1467 | if (page) |
1468 | goto got_pg; | |
1da177e4 | 1469 | |
952f3b51 CL |
1470 | /* |
1471 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1472 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1473 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1474 | * using a larger set of nodes after it has established that the | |
1475 | * allowed per node queues are empty and that nodes are | |
1476 | * over allocated. | |
1477 | */ | |
1478 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1479 | goto nopage; | |
1480 | ||
dd1a239f MG |
1481 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
1482 | wakeup_kswapd(zone, order); | |
1da177e4 | 1483 | |
9bf2229f | 1484 | /* |
7fb1d9fc RS |
1485 | * OK, we're below the kswapd watermark and have kicked background |
1486 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1487 | * to how we want to proceed. | |
1488 | * | |
1489 | * The caller may dip into page reserves a bit more if the caller | |
1490 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1491 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1492 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1493 | */ |
3148890b | 1494 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1495 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1496 | alloc_flags |= ALLOC_HARDER; | |
1497 | if (gfp_mask & __GFP_HIGH) | |
1498 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1499 | if (wait) |
1500 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1501 | |
1502 | /* | |
1503 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1504 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1505 | * |
1506 | * This is the last chance, in general, before the goto nopage. | |
1507 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1508 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1509 | */ |
19770b32 | 1510 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
54a6eb5c | 1511 | high_zoneidx, alloc_flags); |
7fb1d9fc RS |
1512 | if (page) |
1513 | goto got_pg; | |
1da177e4 LT |
1514 | |
1515 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1516 | |
b43a57bb | 1517 | rebalance: |
b84a35be NP |
1518 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1519 | && !in_interrupt()) { | |
1520 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1521 | nofail_alloc: |
b84a35be | 1522 | /* go through the zonelist yet again, ignoring mins */ |
19770b32 | 1523 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
54a6eb5c | 1524 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1525 | if (page) |
1526 | goto got_pg; | |
885036d3 | 1527 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1528 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1529 | goto nofail_alloc; |
1530 | } | |
1da177e4 LT |
1531 | } |
1532 | goto nopage; | |
1533 | } | |
1534 | ||
1535 | /* Atomic allocations - we can't balance anything */ | |
1536 | if (!wait) | |
1537 | goto nopage; | |
1538 | ||
1da177e4 LT |
1539 | cond_resched(); |
1540 | ||
1541 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1542 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1543 | p->flags |= PF_MEMALLOC; |
1544 | reclaim_state.reclaimed_slab = 0; | |
1545 | p->reclaim_state = &reclaim_state; | |
1546 | ||
dac1d27b | 1547 | did_some_progress = try_to_free_pages(zonelist, order, gfp_mask); |
1da177e4 LT |
1548 | |
1549 | p->reclaim_state = NULL; | |
1550 | p->flags &= ~PF_MEMALLOC; | |
1551 | ||
1552 | cond_resched(); | |
1553 | ||
e2c55dc8 | 1554 | if (order != 0) |
9f8f2172 | 1555 | drain_all_pages(); |
e2c55dc8 | 1556 | |
1da177e4 | 1557 | if (likely(did_some_progress)) { |
19770b32 | 1558 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
54a6eb5c | 1559 | zonelist, high_zoneidx, alloc_flags); |
7fb1d9fc RS |
1560 | if (page) |
1561 | goto got_pg; | |
1da177e4 | 1562 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
dd1a239f | 1563 | if (!try_set_zone_oom(zonelist, gfp_mask)) { |
ff0ceb9d DR |
1564 | schedule_timeout_uninterruptible(1); |
1565 | goto restart; | |
1566 | } | |
1567 | ||
1da177e4 LT |
1568 | /* |
1569 | * Go through the zonelist yet one more time, keep | |
1570 | * very high watermark here, this is only to catch | |
1571 | * a parallel oom killing, we must fail if we're still | |
1572 | * under heavy pressure. | |
1573 | */ | |
19770b32 MG |
1574 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, |
1575 | order, zonelist, high_zoneidx, | |
1576 | ALLOC_WMARK_HIGH|ALLOC_CPUSET); | |
ff0ceb9d | 1577 | if (page) { |
dd1a239f | 1578 | clear_zonelist_oom(zonelist, gfp_mask); |
7fb1d9fc | 1579 | goto got_pg; |
ff0ceb9d | 1580 | } |
1da177e4 | 1581 | |
a8bbf72a | 1582 | /* The OOM killer will not help higher order allocs so fail */ |
ff0ceb9d | 1583 | if (order > PAGE_ALLOC_COSTLY_ORDER) { |
dd1a239f | 1584 | clear_zonelist_oom(zonelist, gfp_mask); |
a8bbf72a | 1585 | goto nopage; |
ff0ceb9d | 1586 | } |
a8bbf72a | 1587 | |
9b0f8b04 | 1588 | out_of_memory(zonelist, gfp_mask, order); |
dd1a239f | 1589 | clear_zonelist_oom(zonelist, gfp_mask); |
1da177e4 LT |
1590 | goto restart; |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * Don't let big-order allocations loop unless the caller explicitly | |
1595 | * requests that. Wait for some write requests to complete then retry. | |
1596 | * | |
a41f24ea NA |
1597 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER |
1598 | * means __GFP_NOFAIL, but that may not be true in other | |
ab857d09 | 1599 | * implementations. |
a41f24ea NA |
1600 | * |
1601 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1602 | * specified, then we retry until we no longer reclaim any pages | |
1603 | * (above), or we've reclaimed an order of pages at least as | |
1604 | * large as the allocation's order. In both cases, if the | |
1605 | * allocation still fails, we stop retrying. | |
1da177e4 | 1606 | */ |
a41f24ea | 1607 | pages_reclaimed += did_some_progress; |
1da177e4 LT |
1608 | do_retry = 0; |
1609 | if (!(gfp_mask & __GFP_NORETRY)) { | |
a41f24ea | 1610 | if (order <= PAGE_ALLOC_COSTLY_ORDER) { |
1da177e4 | 1611 | do_retry = 1; |
a41f24ea NA |
1612 | } else { |
1613 | if (gfp_mask & __GFP_REPEAT && | |
1614 | pages_reclaimed < (1 << order)) | |
1615 | do_retry = 1; | |
1616 | } | |
1da177e4 LT |
1617 | if (gfp_mask & __GFP_NOFAIL) |
1618 | do_retry = 1; | |
1619 | } | |
1620 | if (do_retry) { | |
3fcfab16 | 1621 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1622 | goto rebalance; |
1623 | } | |
1624 | ||
1625 | nopage: | |
1626 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1627 | printk(KERN_WARNING "%s: page allocation failure." | |
1628 | " order:%d, mode:0x%x\n", | |
1629 | p->comm, order, gfp_mask); | |
1630 | dump_stack(); | |
578c2fd6 | 1631 | show_mem(); |
1da177e4 | 1632 | } |
1da177e4 | 1633 | got_pg: |
1da177e4 LT |
1634 | return page; |
1635 | } | |
e4048e5d | 1636 | EXPORT_SYMBOL(__alloc_pages_internal); |
1da177e4 LT |
1637 | |
1638 | /* | |
1639 | * Common helper functions. | |
1640 | */ | |
920c7a5d | 1641 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1642 | { |
1643 | struct page * page; | |
1644 | page = alloc_pages(gfp_mask, order); | |
1645 | if (!page) | |
1646 | return 0; | |
1647 | return (unsigned long) page_address(page); | |
1648 | } | |
1649 | ||
1650 | EXPORT_SYMBOL(__get_free_pages); | |
1651 | ||
920c7a5d | 1652 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1653 | { |
1654 | struct page * page; | |
1655 | ||
1656 | /* | |
1657 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1658 | * a highmem page | |
1659 | */ | |
725d704e | 1660 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1661 | |
1662 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1663 | if (page) | |
1664 | return (unsigned long) page_address(page); | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | EXPORT_SYMBOL(get_zeroed_page); | |
1669 | ||
1670 | void __pagevec_free(struct pagevec *pvec) | |
1671 | { | |
1672 | int i = pagevec_count(pvec); | |
1673 | ||
1674 | while (--i >= 0) | |
1675 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1676 | } | |
1677 | ||
920c7a5d | 1678 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 1679 | { |
b5810039 | 1680 | if (put_page_testzero(page)) { |
1da177e4 LT |
1681 | if (order == 0) |
1682 | free_hot_page(page); | |
1683 | else | |
1684 | __free_pages_ok(page, order); | |
1685 | } | |
1686 | } | |
1687 | ||
1688 | EXPORT_SYMBOL(__free_pages); | |
1689 | ||
920c7a5d | 1690 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
1691 | { |
1692 | if (addr != 0) { | |
725d704e | 1693 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1694 | __free_pages(virt_to_page((void *)addr), order); |
1695 | } | |
1696 | } | |
1697 | ||
1698 | EXPORT_SYMBOL(free_pages); | |
1699 | ||
1da177e4 LT |
1700 | static unsigned int nr_free_zone_pages(int offset) |
1701 | { | |
dd1a239f | 1702 | struct zoneref *z; |
54a6eb5c MG |
1703 | struct zone *zone; |
1704 | ||
e310fd43 | 1705 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
1706 | unsigned int sum = 0; |
1707 | ||
0e88460d | 1708 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 1709 | |
54a6eb5c | 1710 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 MB |
1711 | unsigned long size = zone->present_pages; |
1712 | unsigned long high = zone->pages_high; | |
1713 | if (size > high) | |
1714 | sum += size - high; | |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | return sum; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1722 | */ | |
1723 | unsigned int nr_free_buffer_pages(void) | |
1724 | { | |
af4ca457 | 1725 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1726 | } |
c2f1a551 | 1727 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1728 | |
1729 | /* | |
1730 | * Amount of free RAM allocatable within all zones | |
1731 | */ | |
1732 | unsigned int nr_free_pagecache_pages(void) | |
1733 | { | |
2a1e274a | 1734 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1735 | } |
08e0f6a9 CL |
1736 | |
1737 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1738 | { |
08e0f6a9 | 1739 | if (NUMA_BUILD) |
25ba77c1 | 1740 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1741 | } |
1da177e4 | 1742 | |
1da177e4 LT |
1743 | void si_meminfo(struct sysinfo *val) |
1744 | { | |
1745 | val->totalram = totalram_pages; | |
1746 | val->sharedram = 0; | |
d23ad423 | 1747 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1748 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1749 | val->totalhigh = totalhigh_pages; |
1750 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1751 | val->mem_unit = PAGE_SIZE; |
1752 | } | |
1753 | ||
1754 | EXPORT_SYMBOL(si_meminfo); | |
1755 | ||
1756 | #ifdef CONFIG_NUMA | |
1757 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1758 | { | |
1759 | pg_data_t *pgdat = NODE_DATA(nid); | |
1760 | ||
1761 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1762 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1763 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1764 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1765 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1766 | NR_FREE_PAGES); | |
98d2b0eb CL |
1767 | #else |
1768 | val->totalhigh = 0; | |
1769 | val->freehigh = 0; | |
1770 | #endif | |
1da177e4 LT |
1771 | val->mem_unit = PAGE_SIZE; |
1772 | } | |
1773 | #endif | |
1774 | ||
1775 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1776 | ||
1777 | /* | |
1778 | * Show free area list (used inside shift_scroll-lock stuff) | |
1779 | * We also calculate the percentage fragmentation. We do this by counting the | |
1780 | * memory on each free list with the exception of the first item on the list. | |
1781 | */ | |
1782 | void show_free_areas(void) | |
1783 | { | |
c7241913 | 1784 | int cpu; |
1da177e4 LT |
1785 | struct zone *zone; |
1786 | ||
1787 | for_each_zone(zone) { | |
c7241913 | 1788 | if (!populated_zone(zone)) |
1da177e4 | 1789 | continue; |
c7241913 JS |
1790 | |
1791 | show_node(zone); | |
1792 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1793 | |
6b482c67 | 1794 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1795 | struct per_cpu_pageset *pageset; |
1796 | ||
e7c8d5c9 | 1797 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1798 | |
3dfa5721 CL |
1799 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
1800 | cpu, pageset->pcp.high, | |
1801 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
1802 | } |
1803 | } | |
1804 | ||
a25700a5 | 1805 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1806 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1807 | global_page_state(NR_ACTIVE), |
1808 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1809 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1810 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1811 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1812 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1813 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1814 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1815 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1816 | global_page_state(NR_PAGETABLE), |
1817 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1818 | |
1819 | for_each_zone(zone) { | |
1820 | int i; | |
1821 | ||
c7241913 JS |
1822 | if (!populated_zone(zone)) |
1823 | continue; | |
1824 | ||
1da177e4 LT |
1825 | show_node(zone); |
1826 | printk("%s" | |
1827 | " free:%lukB" | |
1828 | " min:%lukB" | |
1829 | " low:%lukB" | |
1830 | " high:%lukB" | |
1831 | " active:%lukB" | |
1832 | " inactive:%lukB" | |
1833 | " present:%lukB" | |
1834 | " pages_scanned:%lu" | |
1835 | " all_unreclaimable? %s" | |
1836 | "\n", | |
1837 | zone->name, | |
d23ad423 | 1838 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1839 | K(zone->pages_min), |
1840 | K(zone->pages_low), | |
1841 | K(zone->pages_high), | |
c8785385 CL |
1842 | K(zone_page_state(zone, NR_ACTIVE)), |
1843 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1844 | K(zone->present_pages), |
1845 | zone->pages_scanned, | |
e815af95 | 1846 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
1847 | ); |
1848 | printk("lowmem_reserve[]:"); | |
1849 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1850 | printk(" %lu", zone->lowmem_reserve[i]); | |
1851 | printk("\n"); | |
1852 | } | |
1853 | ||
1854 | for_each_zone(zone) { | |
8f9de51a | 1855 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1856 | |
c7241913 JS |
1857 | if (!populated_zone(zone)) |
1858 | continue; | |
1859 | ||
1da177e4 LT |
1860 | show_node(zone); |
1861 | printk("%s: ", zone->name); | |
1da177e4 LT |
1862 | |
1863 | spin_lock_irqsave(&zone->lock, flags); | |
1864 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1865 | nr[order] = zone->free_area[order].nr_free; |
1866 | total += nr[order] << order; | |
1da177e4 LT |
1867 | } |
1868 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1869 | for (order = 0; order < MAX_ORDER; order++) |
1870 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1871 | printk("= %lukB\n", K(total)); |
1872 | } | |
1873 | ||
e6f3602d LW |
1874 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
1875 | ||
1da177e4 LT |
1876 | show_swap_cache_info(); |
1877 | } | |
1878 | ||
19770b32 MG |
1879 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
1880 | { | |
1881 | zoneref->zone = zone; | |
1882 | zoneref->zone_idx = zone_idx(zone); | |
1883 | } | |
1884 | ||
1da177e4 LT |
1885 | /* |
1886 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1887 | * |
1888 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1889 | */ |
f0c0b2b8 KH |
1890 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1891 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1892 | { |
1a93205b CL |
1893 | struct zone *zone; |
1894 | ||
98d2b0eb | 1895 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1896 | zone_type++; |
02a68a5e CL |
1897 | |
1898 | do { | |
2f6726e5 | 1899 | zone_type--; |
070f8032 | 1900 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1901 | if (populated_zone(zone)) { |
dd1a239f MG |
1902 | zoneref_set_zone(zone, |
1903 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 1904 | check_highest_zone(zone_type); |
1da177e4 | 1905 | } |
02a68a5e | 1906 | |
2f6726e5 | 1907 | } while (zone_type); |
070f8032 | 1908 | return nr_zones; |
1da177e4 LT |
1909 | } |
1910 | ||
f0c0b2b8 KH |
1911 | |
1912 | /* | |
1913 | * zonelist_order: | |
1914 | * 0 = automatic detection of better ordering. | |
1915 | * 1 = order by ([node] distance, -zonetype) | |
1916 | * 2 = order by (-zonetype, [node] distance) | |
1917 | * | |
1918 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1919 | * the same zonelist. So only NUMA can configure this param. | |
1920 | */ | |
1921 | #define ZONELIST_ORDER_DEFAULT 0 | |
1922 | #define ZONELIST_ORDER_NODE 1 | |
1923 | #define ZONELIST_ORDER_ZONE 2 | |
1924 | ||
1925 | /* zonelist order in the kernel. | |
1926 | * set_zonelist_order() will set this to NODE or ZONE. | |
1927 | */ | |
1928 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1929 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1930 | ||
1931 | ||
1da177e4 | 1932 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1933 | /* The value user specified ....changed by config */ |
1934 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1935 | /* string for sysctl */ | |
1936 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1937 | char numa_zonelist_order[16] = "default"; | |
1938 | ||
1939 | /* | |
1940 | * interface for configure zonelist ordering. | |
1941 | * command line option "numa_zonelist_order" | |
1942 | * = "[dD]efault - default, automatic configuration. | |
1943 | * = "[nN]ode - order by node locality, then by zone within node | |
1944 | * = "[zZ]one - order by zone, then by locality within zone | |
1945 | */ | |
1946 | ||
1947 | static int __parse_numa_zonelist_order(char *s) | |
1948 | { | |
1949 | if (*s == 'd' || *s == 'D') { | |
1950 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1951 | } else if (*s == 'n' || *s == 'N') { | |
1952 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
1953 | } else if (*s == 'z' || *s == 'Z') { | |
1954 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
1955 | } else { | |
1956 | printk(KERN_WARNING | |
1957 | "Ignoring invalid numa_zonelist_order value: " | |
1958 | "%s\n", s); | |
1959 | return -EINVAL; | |
1960 | } | |
1961 | return 0; | |
1962 | } | |
1963 | ||
1964 | static __init int setup_numa_zonelist_order(char *s) | |
1965 | { | |
1966 | if (s) | |
1967 | return __parse_numa_zonelist_order(s); | |
1968 | return 0; | |
1969 | } | |
1970 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
1971 | ||
1972 | /* | |
1973 | * sysctl handler for numa_zonelist_order | |
1974 | */ | |
1975 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
1976 | struct file *file, void __user *buffer, size_t *length, | |
1977 | loff_t *ppos) | |
1978 | { | |
1979 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
1980 | int ret; | |
1981 | ||
1982 | if (write) | |
1983 | strncpy(saved_string, (char*)table->data, | |
1984 | NUMA_ZONELIST_ORDER_LEN); | |
1985 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
1986 | if (ret) | |
1987 | return ret; | |
1988 | if (write) { | |
1989 | int oldval = user_zonelist_order; | |
1990 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
1991 | /* | |
1992 | * bogus value. restore saved string | |
1993 | */ | |
1994 | strncpy((char*)table->data, saved_string, | |
1995 | NUMA_ZONELIST_ORDER_LEN); | |
1996 | user_zonelist_order = oldval; | |
1997 | } else if (oldval != user_zonelist_order) | |
1998 | build_all_zonelists(); | |
1999 | } | |
2000 | return 0; | |
2001 | } | |
2002 | ||
2003 | ||
1da177e4 | 2004 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
2005 | static int node_load[MAX_NUMNODES]; |
2006 | ||
1da177e4 | 2007 | /** |
4dc3b16b | 2008 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2009 | * @node: node whose fallback list we're appending |
2010 | * @used_node_mask: nodemask_t of already used nodes | |
2011 | * | |
2012 | * We use a number of factors to determine which is the next node that should | |
2013 | * appear on a given node's fallback list. The node should not have appeared | |
2014 | * already in @node's fallback list, and it should be the next closest node | |
2015 | * according to the distance array (which contains arbitrary distance values | |
2016 | * from each node to each node in the system), and should also prefer nodes | |
2017 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2018 | * on them otherwise. | |
2019 | * It returns -1 if no node is found. | |
2020 | */ | |
f0c0b2b8 | 2021 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2022 | { |
4cf808eb | 2023 | int n, val; |
1da177e4 LT |
2024 | int min_val = INT_MAX; |
2025 | int best_node = -1; | |
c5f59f08 | 2026 | node_to_cpumask_ptr(tmp, 0); |
1da177e4 | 2027 | |
4cf808eb LT |
2028 | /* Use the local node if we haven't already */ |
2029 | if (!node_isset(node, *used_node_mask)) { | |
2030 | node_set(node, *used_node_mask); | |
2031 | return node; | |
2032 | } | |
1da177e4 | 2033 | |
37b07e41 | 2034 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2035 | |
2036 | /* Don't want a node to appear more than once */ | |
2037 | if (node_isset(n, *used_node_mask)) | |
2038 | continue; | |
2039 | ||
1da177e4 LT |
2040 | /* Use the distance array to find the distance */ |
2041 | val = node_distance(node, n); | |
2042 | ||
4cf808eb LT |
2043 | /* Penalize nodes under us ("prefer the next node") */ |
2044 | val += (n < node); | |
2045 | ||
1da177e4 | 2046 | /* Give preference to headless and unused nodes */ |
c5f59f08 MT |
2047 | node_to_cpumask_ptr_next(tmp, n); |
2048 | if (!cpus_empty(*tmp)) | |
1da177e4 LT |
2049 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2050 | ||
2051 | /* Slight preference for less loaded node */ | |
2052 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2053 | val += node_load[n]; | |
2054 | ||
2055 | if (val < min_val) { | |
2056 | min_val = val; | |
2057 | best_node = n; | |
2058 | } | |
2059 | } | |
2060 | ||
2061 | if (best_node >= 0) | |
2062 | node_set(best_node, *used_node_mask); | |
2063 | ||
2064 | return best_node; | |
2065 | } | |
2066 | ||
f0c0b2b8 KH |
2067 | |
2068 | /* | |
2069 | * Build zonelists ordered by node and zones within node. | |
2070 | * This results in maximum locality--normal zone overflows into local | |
2071 | * DMA zone, if any--but risks exhausting DMA zone. | |
2072 | */ | |
2073 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2074 | { |
f0c0b2b8 | 2075 | int j; |
1da177e4 | 2076 | struct zonelist *zonelist; |
f0c0b2b8 | 2077 | |
54a6eb5c | 2078 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2079 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2080 | ; |
2081 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2082 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2083 | zonelist->_zonerefs[j].zone = NULL; |
2084 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2085 | } |
2086 | ||
523b9458 CL |
2087 | /* |
2088 | * Build gfp_thisnode zonelists | |
2089 | */ | |
2090 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2091 | { | |
523b9458 CL |
2092 | int j; |
2093 | struct zonelist *zonelist; | |
2094 | ||
54a6eb5c MG |
2095 | zonelist = &pgdat->node_zonelists[1]; |
2096 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2097 | zonelist->_zonerefs[j].zone = NULL; |
2098 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2099 | } |
2100 | ||
f0c0b2b8 KH |
2101 | /* |
2102 | * Build zonelists ordered by zone and nodes within zones. | |
2103 | * This results in conserving DMA zone[s] until all Normal memory is | |
2104 | * exhausted, but results in overflowing to remote node while memory | |
2105 | * may still exist in local DMA zone. | |
2106 | */ | |
2107 | static int node_order[MAX_NUMNODES]; | |
2108 | ||
2109 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2110 | { | |
f0c0b2b8 KH |
2111 | int pos, j, node; |
2112 | int zone_type; /* needs to be signed */ | |
2113 | struct zone *z; | |
2114 | struct zonelist *zonelist; | |
2115 | ||
54a6eb5c MG |
2116 | zonelist = &pgdat->node_zonelists[0]; |
2117 | pos = 0; | |
2118 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2119 | for (j = 0; j < nr_nodes; j++) { | |
2120 | node = node_order[j]; | |
2121 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2122 | if (populated_zone(z)) { | |
dd1a239f MG |
2123 | zoneref_set_zone(z, |
2124 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2125 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2126 | } |
2127 | } | |
f0c0b2b8 | 2128 | } |
dd1a239f MG |
2129 | zonelist->_zonerefs[pos].zone = NULL; |
2130 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2131 | } |
2132 | ||
2133 | static int default_zonelist_order(void) | |
2134 | { | |
2135 | int nid, zone_type; | |
2136 | unsigned long low_kmem_size,total_size; | |
2137 | struct zone *z; | |
2138 | int average_size; | |
2139 | /* | |
2140 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2141 | * If they are really small and used heavily, the system can fall | |
2142 | * into OOM very easily. | |
2143 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2144 | */ | |
2145 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2146 | low_kmem_size = 0; | |
2147 | total_size = 0; | |
2148 | for_each_online_node(nid) { | |
2149 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2150 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2151 | if (populated_zone(z)) { | |
2152 | if (zone_type < ZONE_NORMAL) | |
2153 | low_kmem_size += z->present_pages; | |
2154 | total_size += z->present_pages; | |
2155 | } | |
2156 | } | |
2157 | } | |
2158 | if (!low_kmem_size || /* there are no DMA area. */ | |
2159 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2160 | return ZONELIST_ORDER_NODE; | |
2161 | /* | |
2162 | * look into each node's config. | |
2163 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2164 | * local memory, NODE_ORDER may be suitable. | |
2165 | */ | |
37b07e41 LS |
2166 | average_size = total_size / |
2167 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2168 | for_each_online_node(nid) { |
2169 | low_kmem_size = 0; | |
2170 | total_size = 0; | |
2171 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2172 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2173 | if (populated_zone(z)) { | |
2174 | if (zone_type < ZONE_NORMAL) | |
2175 | low_kmem_size += z->present_pages; | |
2176 | total_size += z->present_pages; | |
2177 | } | |
2178 | } | |
2179 | if (low_kmem_size && | |
2180 | total_size > average_size && /* ignore small node */ | |
2181 | low_kmem_size > total_size * 70/100) | |
2182 | return ZONELIST_ORDER_NODE; | |
2183 | } | |
2184 | return ZONELIST_ORDER_ZONE; | |
2185 | } | |
2186 | ||
2187 | static void set_zonelist_order(void) | |
2188 | { | |
2189 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2190 | current_zonelist_order = default_zonelist_order(); | |
2191 | else | |
2192 | current_zonelist_order = user_zonelist_order; | |
2193 | } | |
2194 | ||
2195 | static void build_zonelists(pg_data_t *pgdat) | |
2196 | { | |
2197 | int j, node, load; | |
2198 | enum zone_type i; | |
1da177e4 | 2199 | nodemask_t used_mask; |
f0c0b2b8 KH |
2200 | int local_node, prev_node; |
2201 | struct zonelist *zonelist; | |
2202 | int order = current_zonelist_order; | |
1da177e4 LT |
2203 | |
2204 | /* initialize zonelists */ | |
523b9458 | 2205 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2206 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2207 | zonelist->_zonerefs[0].zone = NULL; |
2208 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2209 | } |
2210 | ||
2211 | /* NUMA-aware ordering of nodes */ | |
2212 | local_node = pgdat->node_id; | |
2213 | load = num_online_nodes(); | |
2214 | prev_node = local_node; | |
2215 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2216 | |
2217 | memset(node_load, 0, sizeof(node_load)); | |
2218 | memset(node_order, 0, sizeof(node_order)); | |
2219 | j = 0; | |
2220 | ||
1da177e4 | 2221 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2222 | int distance = node_distance(local_node, node); |
2223 | ||
2224 | /* | |
2225 | * If another node is sufficiently far away then it is better | |
2226 | * to reclaim pages in a zone before going off node. | |
2227 | */ | |
2228 | if (distance > RECLAIM_DISTANCE) | |
2229 | zone_reclaim_mode = 1; | |
2230 | ||
1da177e4 LT |
2231 | /* |
2232 | * We don't want to pressure a particular node. | |
2233 | * So adding penalty to the first node in same | |
2234 | * distance group to make it round-robin. | |
2235 | */ | |
9eeff239 | 2236 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2237 | node_load[node] = load; |
2238 | ||
1da177e4 LT |
2239 | prev_node = node; |
2240 | load--; | |
f0c0b2b8 KH |
2241 | if (order == ZONELIST_ORDER_NODE) |
2242 | build_zonelists_in_node_order(pgdat, node); | |
2243 | else | |
2244 | node_order[j++] = node; /* remember order */ | |
2245 | } | |
1da177e4 | 2246 | |
f0c0b2b8 KH |
2247 | if (order == ZONELIST_ORDER_ZONE) { |
2248 | /* calculate node order -- i.e., DMA last! */ | |
2249 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2250 | } |
523b9458 CL |
2251 | |
2252 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2253 | } |
2254 | ||
9276b1bc | 2255 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2256 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2257 | { |
54a6eb5c MG |
2258 | struct zonelist *zonelist; |
2259 | struct zonelist_cache *zlc; | |
dd1a239f | 2260 | struct zoneref *z; |
9276b1bc | 2261 | |
54a6eb5c MG |
2262 | zonelist = &pgdat->node_zonelists[0]; |
2263 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2264 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2265 | for (z = zonelist->_zonerefs; z->zone; z++) |
2266 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2267 | } |
2268 | ||
f0c0b2b8 | 2269 | |
1da177e4 LT |
2270 | #else /* CONFIG_NUMA */ |
2271 | ||
f0c0b2b8 KH |
2272 | static void set_zonelist_order(void) |
2273 | { | |
2274 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2275 | } | |
2276 | ||
2277 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2278 | { |
19655d34 | 2279 | int node, local_node; |
54a6eb5c MG |
2280 | enum zone_type j; |
2281 | struct zonelist *zonelist; | |
1da177e4 LT |
2282 | |
2283 | local_node = pgdat->node_id; | |
1da177e4 | 2284 | |
54a6eb5c MG |
2285 | zonelist = &pgdat->node_zonelists[0]; |
2286 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2287 | |
54a6eb5c MG |
2288 | /* |
2289 | * Now we build the zonelist so that it contains the zones | |
2290 | * of all the other nodes. | |
2291 | * We don't want to pressure a particular node, so when | |
2292 | * building the zones for node N, we make sure that the | |
2293 | * zones coming right after the local ones are those from | |
2294 | * node N+1 (modulo N) | |
2295 | */ | |
2296 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2297 | if (!node_online(node)) | |
2298 | continue; | |
2299 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2300 | MAX_NR_ZONES - 1); | |
1da177e4 | 2301 | } |
54a6eb5c MG |
2302 | for (node = 0; node < local_node; node++) { |
2303 | if (!node_online(node)) | |
2304 | continue; | |
2305 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2306 | MAX_NR_ZONES - 1); | |
2307 | } | |
2308 | ||
dd1a239f MG |
2309 | zonelist->_zonerefs[j].zone = NULL; |
2310 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
2311 | } |
2312 | ||
9276b1bc | 2313 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2314 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2315 | { |
54a6eb5c | 2316 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
2317 | } |
2318 | ||
1da177e4 LT |
2319 | #endif /* CONFIG_NUMA */ |
2320 | ||
6811378e | 2321 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2322 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2323 | { |
6811378e | 2324 | int nid; |
9276b1bc PJ |
2325 | |
2326 | for_each_online_node(nid) { | |
7ea1530a CL |
2327 | pg_data_t *pgdat = NODE_DATA(nid); |
2328 | ||
2329 | build_zonelists(pgdat); | |
2330 | build_zonelist_cache(pgdat); | |
9276b1bc | 2331 | } |
6811378e YG |
2332 | return 0; |
2333 | } | |
2334 | ||
f0c0b2b8 | 2335 | void build_all_zonelists(void) |
6811378e | 2336 | { |
f0c0b2b8 KH |
2337 | set_zonelist_order(); |
2338 | ||
6811378e | 2339 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2340 | __build_all_zonelists(NULL); |
68ad8df4 | 2341 | mminit_verify_zonelist(); |
6811378e YG |
2342 | cpuset_init_current_mems_allowed(); |
2343 | } else { | |
183ff22b | 2344 | /* we have to stop all cpus to guarantee there is no user |
6811378e YG |
2345 | of zonelist */ |
2346 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2347 | /* cpuset refresh routine should be here */ | |
2348 | } | |
bd1e22b8 | 2349 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2350 | /* |
2351 | * Disable grouping by mobility if the number of pages in the | |
2352 | * system is too low to allow the mechanism to work. It would be | |
2353 | * more accurate, but expensive to check per-zone. This check is | |
2354 | * made on memory-hotadd so a system can start with mobility | |
2355 | * disabled and enable it later | |
2356 | */ | |
d9c23400 | 2357 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2358 | page_group_by_mobility_disabled = 1; |
2359 | else | |
2360 | page_group_by_mobility_disabled = 0; | |
2361 | ||
2362 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2363 | "Total pages: %ld\n", | |
f0c0b2b8 KH |
2364 | num_online_nodes(), |
2365 | zonelist_order_name[current_zonelist_order], | |
9ef9acb0 | 2366 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2367 | vm_total_pages); |
2368 | #ifdef CONFIG_NUMA | |
2369 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2370 | #endif | |
1da177e4 LT |
2371 | } |
2372 | ||
2373 | /* | |
2374 | * Helper functions to size the waitqueue hash table. | |
2375 | * Essentially these want to choose hash table sizes sufficiently | |
2376 | * large so that collisions trying to wait on pages are rare. | |
2377 | * But in fact, the number of active page waitqueues on typical | |
2378 | * systems is ridiculously low, less than 200. So this is even | |
2379 | * conservative, even though it seems large. | |
2380 | * | |
2381 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2382 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2383 | */ | |
2384 | #define PAGES_PER_WAITQUEUE 256 | |
2385 | ||
cca448fe | 2386 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2387 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2388 | { |
2389 | unsigned long size = 1; | |
2390 | ||
2391 | pages /= PAGES_PER_WAITQUEUE; | |
2392 | ||
2393 | while (size < pages) | |
2394 | size <<= 1; | |
2395 | ||
2396 | /* | |
2397 | * Once we have dozens or even hundreds of threads sleeping | |
2398 | * on IO we've got bigger problems than wait queue collision. | |
2399 | * Limit the size of the wait table to a reasonable size. | |
2400 | */ | |
2401 | size = min(size, 4096UL); | |
2402 | ||
2403 | return max(size, 4UL); | |
2404 | } | |
cca448fe YG |
2405 | #else |
2406 | /* | |
2407 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2408 | * a suitable size for its wait_table. So we use the maximum size now. | |
2409 | * | |
2410 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2411 | * | |
2412 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2413 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2414 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2415 | * | |
2416 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2417 | * or more by the traditional way. (See above). It equals: | |
2418 | * | |
2419 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2420 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2421 | * powerpc (64K page size) : = (32G +16M)byte. | |
2422 | */ | |
2423 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2424 | { | |
2425 | return 4096UL; | |
2426 | } | |
2427 | #endif | |
1da177e4 LT |
2428 | |
2429 | /* | |
2430 | * This is an integer logarithm so that shifts can be used later | |
2431 | * to extract the more random high bits from the multiplicative | |
2432 | * hash function before the remainder is taken. | |
2433 | */ | |
2434 | static inline unsigned long wait_table_bits(unsigned long size) | |
2435 | { | |
2436 | return ffz(~size); | |
2437 | } | |
2438 | ||
2439 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2440 | ||
56fd56b8 | 2441 | /* |
d9c23400 | 2442 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
56fd56b8 MG |
2443 | * of blocks reserved is based on zone->pages_min. The memory within the |
2444 | * reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
2445 | * higher will lead to a bigger reserve which will get freed as contiguous | |
2446 | * blocks as reclaim kicks in | |
2447 | */ | |
2448 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2449 | { | |
2450 | unsigned long start_pfn, pfn, end_pfn; | |
2451 | struct page *page; | |
2452 | unsigned long reserve, block_migratetype; | |
2453 | ||
2454 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2455 | start_pfn = zone->zone_start_pfn; | |
2456 | end_pfn = start_pfn + zone->spanned_pages; | |
d9c23400 MG |
2457 | reserve = roundup(zone->pages_min, pageblock_nr_pages) >> |
2458 | pageblock_order; | |
56fd56b8 | 2459 | |
d9c23400 | 2460 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2461 | if (!pfn_valid(pfn)) |
2462 | continue; | |
2463 | page = pfn_to_page(pfn); | |
2464 | ||
2465 | /* Blocks with reserved pages will never free, skip them. */ | |
2466 | if (PageReserved(page)) | |
2467 | continue; | |
2468 | ||
2469 | block_migratetype = get_pageblock_migratetype(page); | |
2470 | ||
2471 | /* If this block is reserved, account for it */ | |
2472 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2473 | reserve--; | |
2474 | continue; | |
2475 | } | |
2476 | ||
2477 | /* Suitable for reserving if this block is movable */ | |
2478 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2479 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2480 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2481 | reserve--; | |
2482 | continue; | |
2483 | } | |
2484 | ||
2485 | /* | |
2486 | * If the reserve is met and this is a previous reserved block, | |
2487 | * take it back | |
2488 | */ | |
2489 | if (block_migratetype == MIGRATE_RESERVE) { | |
2490 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2491 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2492 | } | |
2493 | } | |
2494 | } | |
ac0e5b7a | 2495 | |
1da177e4 LT |
2496 | /* |
2497 | * Initially all pages are reserved - free ones are freed | |
2498 | * up by free_all_bootmem() once the early boot process is | |
2499 | * done. Non-atomic initialization, single-pass. | |
2500 | */ | |
c09b4240 | 2501 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2502 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2503 | { |
1da177e4 | 2504 | struct page *page; |
29751f69 AW |
2505 | unsigned long end_pfn = start_pfn + size; |
2506 | unsigned long pfn; | |
86051ca5 | 2507 | struct zone *z; |
1da177e4 | 2508 | |
86051ca5 | 2509 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 2510 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2511 | /* |
2512 | * There can be holes in boot-time mem_map[]s | |
2513 | * handed to this function. They do not | |
2514 | * exist on hotplugged memory. | |
2515 | */ | |
2516 | if (context == MEMMAP_EARLY) { | |
2517 | if (!early_pfn_valid(pfn)) | |
2518 | continue; | |
2519 | if (!early_pfn_in_nid(pfn, nid)) | |
2520 | continue; | |
2521 | } | |
d41dee36 AW |
2522 | page = pfn_to_page(pfn); |
2523 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 2524 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 2525 | init_page_count(page); |
1da177e4 LT |
2526 | reset_page_mapcount(page); |
2527 | SetPageReserved(page); | |
b2a0ac88 MG |
2528 | /* |
2529 | * Mark the block movable so that blocks are reserved for | |
2530 | * movable at startup. This will force kernel allocations | |
2531 | * to reserve their blocks rather than leaking throughout | |
2532 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2533 | * kernel allocations are made. Later some blocks near |
2534 | * the start are marked MIGRATE_RESERVE by | |
2535 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
2536 | * |
2537 | * bitmap is created for zone's valid pfn range. but memmap | |
2538 | * can be created for invalid pages (for alignment) | |
2539 | * check here not to call set_pageblock_migratetype() against | |
2540 | * pfn out of zone. | |
b2a0ac88 | 2541 | */ |
86051ca5 KH |
2542 | if ((z->zone_start_pfn <= pfn) |
2543 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
2544 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 2545 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2546 | |
1da177e4 LT |
2547 | INIT_LIST_HEAD(&page->lru); |
2548 | #ifdef WANT_PAGE_VIRTUAL | |
2549 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2550 | if (!is_highmem_idx(zone)) | |
3212c6be | 2551 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2552 | #endif |
1da177e4 LT |
2553 | } |
2554 | } | |
2555 | ||
1e548deb | 2556 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 2557 | { |
b2a0ac88 MG |
2558 | int order, t; |
2559 | for_each_migratetype_order(order, t) { | |
2560 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2561 | zone->free_area[order].nr_free = 0; |
2562 | } | |
2563 | } | |
2564 | ||
2565 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2566 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2567 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2568 | #endif |
2569 | ||
1d6f4e60 | 2570 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2571 | { |
2572 | int batch; | |
2573 | ||
2574 | /* | |
2575 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2576 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2577 | * |
2578 | * OK, so we don't know how big the cache is. So guess. | |
2579 | */ | |
2580 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2581 | if (batch * PAGE_SIZE > 512 * 1024) |
2582 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2583 | batch /= 4; /* We effectively *= 4 below */ |
2584 | if (batch < 1) | |
2585 | batch = 1; | |
2586 | ||
2587 | /* | |
0ceaacc9 NP |
2588 | * Clamp the batch to a 2^n - 1 value. Having a power |
2589 | * of 2 value was found to be more likely to have | |
2590 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2591 | * |
0ceaacc9 NP |
2592 | * For example if 2 tasks are alternately allocating |
2593 | * batches of pages, one task can end up with a lot | |
2594 | * of pages of one half of the possible page colors | |
2595 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2596 | */ |
0ceaacc9 | 2597 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2598 | |
e7c8d5c9 CL |
2599 | return batch; |
2600 | } | |
2601 | ||
2caaad41 CL |
2602 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2603 | { | |
2604 | struct per_cpu_pages *pcp; | |
2605 | ||
1c6fe946 MD |
2606 | memset(p, 0, sizeof(*p)); |
2607 | ||
3dfa5721 | 2608 | pcp = &p->pcp; |
2caaad41 | 2609 | pcp->count = 0; |
2caaad41 CL |
2610 | pcp->high = 6 * batch; |
2611 | pcp->batch = max(1UL, 1 * batch); | |
2612 | INIT_LIST_HEAD(&pcp->list); | |
2caaad41 CL |
2613 | } |
2614 | ||
8ad4b1fb RS |
2615 | /* |
2616 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2617 | * to the value high for the pageset p. | |
2618 | */ | |
2619 | ||
2620 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2621 | unsigned long high) | |
2622 | { | |
2623 | struct per_cpu_pages *pcp; | |
2624 | ||
3dfa5721 | 2625 | pcp = &p->pcp; |
8ad4b1fb RS |
2626 | pcp->high = high; |
2627 | pcp->batch = max(1UL, high/4); | |
2628 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2629 | pcp->batch = PAGE_SHIFT * 8; | |
2630 | } | |
2631 | ||
2632 | ||
e7c8d5c9 CL |
2633 | #ifdef CONFIG_NUMA |
2634 | /* | |
2caaad41 CL |
2635 | * Boot pageset table. One per cpu which is going to be used for all |
2636 | * zones and all nodes. The parameters will be set in such a way | |
2637 | * that an item put on a list will immediately be handed over to | |
2638 | * the buddy list. This is safe since pageset manipulation is done | |
2639 | * with interrupts disabled. | |
2640 | * | |
2641 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2642 | * |
2643 | * The boot_pagesets must be kept even after bootup is complete for | |
2644 | * unused processors and/or zones. They do play a role for bootstrapping | |
2645 | * hotplugged processors. | |
2646 | * | |
2647 | * zoneinfo_show() and maybe other functions do | |
2648 | * not check if the processor is online before following the pageset pointer. | |
2649 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2650 | */ |
88a2a4ac | 2651 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2652 | |
2653 | /* | |
2654 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2655 | * per cpu pageset array in struct zone. |
2656 | */ | |
6292d9aa | 2657 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2658 | { |
2659 | struct zone *zone, *dzone; | |
37c0708d CL |
2660 | int node = cpu_to_node(cpu); |
2661 | ||
2662 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 CL |
2663 | |
2664 | for_each_zone(zone) { | |
e7c8d5c9 | 2665 | |
66a55030 CL |
2666 | if (!populated_zone(zone)) |
2667 | continue; | |
2668 | ||
23316bc8 | 2669 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2670 | GFP_KERNEL, node); |
23316bc8 | 2671 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2672 | goto bad; |
e7c8d5c9 | 2673 | |
23316bc8 | 2674 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2675 | |
2676 | if (percpu_pagelist_fraction) | |
2677 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2678 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2679 | } |
2680 | ||
2681 | return 0; | |
2682 | bad: | |
2683 | for_each_zone(dzone) { | |
64191688 AM |
2684 | if (!populated_zone(dzone)) |
2685 | continue; | |
e7c8d5c9 CL |
2686 | if (dzone == zone) |
2687 | break; | |
23316bc8 NP |
2688 | kfree(zone_pcp(dzone, cpu)); |
2689 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2690 | } |
2691 | return -ENOMEM; | |
2692 | } | |
2693 | ||
2694 | static inline void free_zone_pagesets(int cpu) | |
2695 | { | |
e7c8d5c9 CL |
2696 | struct zone *zone; |
2697 | ||
2698 | for_each_zone(zone) { | |
2699 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2700 | ||
f3ef9ead DR |
2701 | /* Free per_cpu_pageset if it is slab allocated */ |
2702 | if (pset != &boot_pageset[cpu]) | |
2703 | kfree(pset); | |
e7c8d5c9 | 2704 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2705 | } |
e7c8d5c9 CL |
2706 | } |
2707 | ||
9c7b216d | 2708 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2709 | unsigned long action, |
2710 | void *hcpu) | |
2711 | { | |
2712 | int cpu = (long)hcpu; | |
2713 | int ret = NOTIFY_OK; | |
2714 | ||
2715 | switch (action) { | |
ce421c79 | 2716 | case CPU_UP_PREPARE: |
8bb78442 | 2717 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2718 | if (process_zones(cpu)) |
2719 | ret = NOTIFY_BAD; | |
2720 | break; | |
2721 | case CPU_UP_CANCELED: | |
8bb78442 | 2722 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2723 | case CPU_DEAD: |
8bb78442 | 2724 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2725 | free_zone_pagesets(cpu); |
2726 | break; | |
2727 | default: | |
2728 | break; | |
e7c8d5c9 CL |
2729 | } |
2730 | return ret; | |
2731 | } | |
2732 | ||
74b85f37 | 2733 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2734 | { &pageset_cpuup_callback, NULL, 0 }; |
2735 | ||
78d9955b | 2736 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2737 | { |
2738 | int err; | |
2739 | ||
2740 | /* Initialize per_cpu_pageset for cpu 0. | |
2741 | * A cpuup callback will do this for every cpu | |
2742 | * as it comes online | |
2743 | */ | |
2744 | err = process_zones(smp_processor_id()); | |
2745 | BUG_ON(err); | |
2746 | register_cpu_notifier(&pageset_notifier); | |
2747 | } | |
2748 | ||
2749 | #endif | |
2750 | ||
577a32f6 | 2751 | static noinline __init_refok |
cca448fe | 2752 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2753 | { |
2754 | int i; | |
2755 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2756 | size_t alloc_size; |
ed8ece2e DH |
2757 | |
2758 | /* | |
2759 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2760 | * per zone. | |
2761 | */ | |
02b694de YG |
2762 | zone->wait_table_hash_nr_entries = |
2763 | wait_table_hash_nr_entries(zone_size_pages); | |
2764 | zone->wait_table_bits = | |
2765 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2766 | alloc_size = zone->wait_table_hash_nr_entries |
2767 | * sizeof(wait_queue_head_t); | |
2768 | ||
cd94b9db | 2769 | if (!slab_is_available()) { |
cca448fe YG |
2770 | zone->wait_table = (wait_queue_head_t *) |
2771 | alloc_bootmem_node(pgdat, alloc_size); | |
2772 | } else { | |
2773 | /* | |
2774 | * This case means that a zone whose size was 0 gets new memory | |
2775 | * via memory hot-add. | |
2776 | * But it may be the case that a new node was hot-added. In | |
2777 | * this case vmalloc() will not be able to use this new node's | |
2778 | * memory - this wait_table must be initialized to use this new | |
2779 | * node itself as well. | |
2780 | * To use this new node's memory, further consideration will be | |
2781 | * necessary. | |
2782 | */ | |
8691f3a7 | 2783 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
2784 | } |
2785 | if (!zone->wait_table) | |
2786 | return -ENOMEM; | |
ed8ece2e | 2787 | |
02b694de | 2788 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2789 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2790 | |
2791 | return 0; | |
ed8ece2e DH |
2792 | } |
2793 | ||
c09b4240 | 2794 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2795 | { |
2796 | int cpu; | |
2797 | unsigned long batch = zone_batchsize(zone); | |
2798 | ||
2799 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2800 | #ifdef CONFIG_NUMA | |
2801 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2802 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2803 | setup_pageset(&boot_pageset[cpu],0); |
2804 | #else | |
2805 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2806 | #endif | |
2807 | } | |
f5335c0f AB |
2808 | if (zone->present_pages) |
2809 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2810 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2811 | } |
2812 | ||
718127cc YG |
2813 | __meminit int init_currently_empty_zone(struct zone *zone, |
2814 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2815 | unsigned long size, |
2816 | enum memmap_context context) | |
ed8ece2e DH |
2817 | { |
2818 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2819 | int ret; |
2820 | ret = zone_wait_table_init(zone, size); | |
2821 | if (ret) | |
2822 | return ret; | |
ed8ece2e DH |
2823 | pgdat->nr_zones = zone_idx(zone) + 1; |
2824 | ||
ed8ece2e DH |
2825 | zone->zone_start_pfn = zone_start_pfn; |
2826 | ||
708614e6 MG |
2827 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
2828 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
2829 | pgdat->node_id, | |
2830 | (unsigned long)zone_idx(zone), | |
2831 | zone_start_pfn, (zone_start_pfn + size)); | |
2832 | ||
1e548deb | 2833 | zone_init_free_lists(zone); |
718127cc YG |
2834 | |
2835 | return 0; | |
ed8ece2e DH |
2836 | } |
2837 | ||
c713216d MG |
2838 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2839 | /* | |
2840 | * Basic iterator support. Return the first range of PFNs for a node | |
2841 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2842 | */ | |
a3142c8e | 2843 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2844 | { |
2845 | int i; | |
2846 | ||
2847 | for (i = 0; i < nr_nodemap_entries; i++) | |
2848 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2849 | return i; | |
2850 | ||
2851 | return -1; | |
2852 | } | |
2853 | ||
2854 | /* | |
2855 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 2856 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 2857 | */ |
a3142c8e | 2858 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2859 | { |
2860 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2861 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2862 | return index; | |
2863 | ||
2864 | return -1; | |
2865 | } | |
2866 | ||
2867 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2868 | /* | |
2869 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2870 | * Architectures may implement their own version but if add_active_range() | |
2871 | * was used and there are no special requirements, this is a convenient | |
2872 | * alternative | |
2873 | */ | |
6f076f5d | 2874 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2875 | { |
2876 | int i; | |
2877 | ||
2878 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2879 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2880 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2881 | ||
2882 | if (start_pfn <= pfn && pfn < end_pfn) | |
2883 | return early_node_map[i].nid; | |
2884 | } | |
2885 | ||
2886 | return 0; | |
2887 | } | |
2888 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2889 | ||
2890 | /* Basic iterator support to walk early_node_map[] */ | |
2891 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2892 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2893 | i = next_active_region_index_in_nid(i, nid)) | |
2894 | ||
2895 | /** | |
2896 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2897 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2898 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2899 | * |
2900 | * If an architecture guarantees that all ranges registered with | |
2901 | * add_active_ranges() contain no holes and may be freed, this | |
2902 | * this function may be used instead of calling free_bootmem() manually. | |
2903 | */ | |
2904 | void __init free_bootmem_with_active_regions(int nid, | |
2905 | unsigned long max_low_pfn) | |
2906 | { | |
2907 | int i; | |
2908 | ||
2909 | for_each_active_range_index_in_nid(i, nid) { | |
2910 | unsigned long size_pages = 0; | |
2911 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2912 | ||
2913 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2914 | continue; | |
2915 | ||
2916 | if (end_pfn > max_low_pfn) | |
2917 | end_pfn = max_low_pfn; | |
2918 | ||
2919 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2920 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2921 | PFN_PHYS(early_node_map[i].start_pfn), | |
2922 | size_pages << PAGE_SHIFT); | |
2923 | } | |
2924 | } | |
2925 | ||
b5bc6c0e YL |
2926 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
2927 | { | |
2928 | int i; | |
d52d53b8 | 2929 | int ret; |
b5bc6c0e | 2930 | |
d52d53b8 YL |
2931 | for_each_active_range_index_in_nid(i, nid) { |
2932 | ret = work_fn(early_node_map[i].start_pfn, | |
2933 | early_node_map[i].end_pfn, data); | |
2934 | if (ret) | |
2935 | break; | |
2936 | } | |
b5bc6c0e | 2937 | } |
c713216d MG |
2938 | /** |
2939 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2940 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2941 | * |
2942 | * If an architecture guarantees that all ranges registered with | |
2943 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2944 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2945 | */ |
2946 | void __init sparse_memory_present_with_active_regions(int nid) | |
2947 | { | |
2948 | int i; | |
2949 | ||
2950 | for_each_active_range_index_in_nid(i, nid) | |
2951 | memory_present(early_node_map[i].nid, | |
2952 | early_node_map[i].start_pfn, | |
2953 | early_node_map[i].end_pfn); | |
2954 | } | |
2955 | ||
fb01439c MG |
2956 | /** |
2957 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2958 | * @nid: The nid of the node to push the boundary for | |
2959 | * @start_pfn: The start pfn of the node | |
2960 | * @end_pfn: The end pfn of the node | |
2961 | * | |
2962 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2963 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2964 | * be hotplugged even though no physical memory exists. This function allows | |
2965 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2966 | * be used later. | |
2967 | */ | |
2968 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2969 | void __init push_node_boundaries(unsigned int nid, | |
2970 | unsigned long start_pfn, unsigned long end_pfn) | |
2971 | { | |
6b74ab97 MG |
2972 | mminit_dprintk(MMINIT_TRACE, "zoneboundary", |
2973 | "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
fb01439c MG |
2974 | nid, start_pfn, end_pfn); |
2975 | ||
2976 | /* Initialise the boundary for this node if necessary */ | |
2977 | if (node_boundary_end_pfn[nid] == 0) | |
2978 | node_boundary_start_pfn[nid] = -1UL; | |
2979 | ||
2980 | /* Update the boundaries */ | |
2981 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2982 | node_boundary_start_pfn[nid] = start_pfn; | |
2983 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2984 | node_boundary_end_pfn[nid] = end_pfn; | |
2985 | } | |
2986 | ||
2987 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2988 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2989 | unsigned long *start_pfn, unsigned long *end_pfn) |
2990 | { | |
6b74ab97 MG |
2991 | mminit_dprintk(MMINIT_TRACE, "zoneboundary", |
2992 | "Entering account_node_boundary(%u, %lu, %lu)\n", | |
fb01439c MG |
2993 | nid, *start_pfn, *end_pfn); |
2994 | ||
2995 | /* Return if boundary information has not been provided */ | |
2996 | if (node_boundary_end_pfn[nid] == 0) | |
2997 | return; | |
2998 | ||
2999 | /* Check the boundaries and update if necessary */ | |
3000 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
3001 | *start_pfn = node_boundary_start_pfn[nid]; | |
3002 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
3003 | *end_pfn = node_boundary_end_pfn[nid]; | |
3004 | } | |
3005 | #else | |
3006 | void __init push_node_boundaries(unsigned int nid, | |
3007 | unsigned long start_pfn, unsigned long end_pfn) {} | |
3008 | ||
98011f56 | 3009 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
3010 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
3011 | #endif | |
3012 | ||
3013 | ||
c713216d MG |
3014 | /** |
3015 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3016 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3017 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3018 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3019 | * |
3020 | * It returns the start and end page frame of a node based on information | |
3021 | * provided by an arch calling add_active_range(). If called for a node | |
3022 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3023 | * PFNs will be 0. |
c713216d | 3024 | */ |
a3142c8e | 3025 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3026 | unsigned long *start_pfn, unsigned long *end_pfn) |
3027 | { | |
3028 | int i; | |
3029 | *start_pfn = -1UL; | |
3030 | *end_pfn = 0; | |
3031 | ||
3032 | for_each_active_range_index_in_nid(i, nid) { | |
3033 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3034 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3035 | } | |
3036 | ||
633c0666 | 3037 | if (*start_pfn == -1UL) |
c713216d | 3038 | *start_pfn = 0; |
fb01439c MG |
3039 | |
3040 | /* Push the node boundaries out if requested */ | |
3041 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
3042 | } |
3043 | ||
2a1e274a MG |
3044 | /* |
3045 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3046 | * assumption is made that zones within a node are ordered in monotonic | |
3047 | * increasing memory addresses so that the "highest" populated zone is used | |
3048 | */ | |
3049 | void __init find_usable_zone_for_movable(void) | |
3050 | { | |
3051 | int zone_index; | |
3052 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3053 | if (zone_index == ZONE_MOVABLE) | |
3054 | continue; | |
3055 | ||
3056 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3057 | arch_zone_lowest_possible_pfn[zone_index]) | |
3058 | break; | |
3059 | } | |
3060 | ||
3061 | VM_BUG_ON(zone_index == -1); | |
3062 | movable_zone = zone_index; | |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3067 | * because it is sized independant of architecture. Unlike the other zones, | |
3068 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3069 | * in each node depending on the size of each node and how evenly kernelcore | |
3070 | * is distributed. This helper function adjusts the zone ranges | |
3071 | * provided by the architecture for a given node by using the end of the | |
3072 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3073 | * zones within a node are in order of monotonic increases memory addresses | |
3074 | */ | |
3075 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
3076 | unsigned long zone_type, | |
3077 | unsigned long node_start_pfn, | |
3078 | unsigned long node_end_pfn, | |
3079 | unsigned long *zone_start_pfn, | |
3080 | unsigned long *zone_end_pfn) | |
3081 | { | |
3082 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3083 | if (zone_movable_pfn[nid]) { | |
3084 | /* Size ZONE_MOVABLE */ | |
3085 | if (zone_type == ZONE_MOVABLE) { | |
3086 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3087 | *zone_end_pfn = min(node_end_pfn, | |
3088 | arch_zone_highest_possible_pfn[movable_zone]); | |
3089 | ||
3090 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3091 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3092 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3093 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3094 | ||
3095 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3096 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3097 | *zone_start_pfn = *zone_end_pfn; | |
3098 | } | |
3099 | } | |
3100 | ||
c713216d MG |
3101 | /* |
3102 | * Return the number of pages a zone spans in a node, including holes | |
3103 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3104 | */ | |
6ea6e688 | 3105 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3106 | unsigned long zone_type, |
3107 | unsigned long *ignored) | |
3108 | { | |
3109 | unsigned long node_start_pfn, node_end_pfn; | |
3110 | unsigned long zone_start_pfn, zone_end_pfn; | |
3111 | ||
3112 | /* Get the start and end of the node and zone */ | |
3113 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3114 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3115 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3116 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3117 | node_start_pfn, node_end_pfn, | |
3118 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3119 | |
3120 | /* Check that this node has pages within the zone's required range */ | |
3121 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3122 | return 0; | |
3123 | ||
3124 | /* Move the zone boundaries inside the node if necessary */ | |
3125 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3126 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3127 | ||
3128 | /* Return the spanned pages */ | |
3129 | return zone_end_pfn - zone_start_pfn; | |
3130 | } | |
3131 | ||
3132 | /* | |
3133 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3134 | * then all holes in the requested range will be accounted for. |
c713216d | 3135 | */ |
a3142c8e | 3136 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3137 | unsigned long range_start_pfn, |
3138 | unsigned long range_end_pfn) | |
3139 | { | |
3140 | int i = 0; | |
3141 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3142 | unsigned long start_pfn; | |
3143 | ||
3144 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3145 | i = first_active_region_index_in_nid(nid); | |
3146 | if (i == -1) | |
3147 | return 0; | |
3148 | ||
b5445f95 MG |
3149 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3150 | ||
9c7cd687 MG |
3151 | /* Account for ranges before physical memory on this node */ |
3152 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3153 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3154 | |
3155 | /* Find all holes for the zone within the node */ | |
3156 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3157 | ||
3158 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3159 | if (prev_end_pfn >= range_end_pfn) | |
3160 | break; | |
3161 | ||
3162 | /* Make sure the end of the zone is not within the hole */ | |
3163 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3164 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3165 | ||
3166 | /* Update the hole size cound and move on */ | |
3167 | if (start_pfn > range_start_pfn) { | |
3168 | BUG_ON(prev_end_pfn > start_pfn); | |
3169 | hole_pages += start_pfn - prev_end_pfn; | |
3170 | } | |
3171 | prev_end_pfn = early_node_map[i].end_pfn; | |
3172 | } | |
3173 | ||
9c7cd687 MG |
3174 | /* Account for ranges past physical memory on this node */ |
3175 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3176 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3177 | max(range_start_pfn, prev_end_pfn); |
3178 | ||
c713216d MG |
3179 | return hole_pages; |
3180 | } | |
3181 | ||
3182 | /** | |
3183 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3184 | * @start_pfn: The start PFN to start searching for holes | |
3185 | * @end_pfn: The end PFN to stop searching for holes | |
3186 | * | |
88ca3b94 | 3187 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3188 | */ |
3189 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3190 | unsigned long end_pfn) | |
3191 | { | |
3192 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3193 | } | |
3194 | ||
3195 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3196 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3197 | unsigned long zone_type, |
3198 | unsigned long *ignored) | |
3199 | { | |
9c7cd687 MG |
3200 | unsigned long node_start_pfn, node_end_pfn; |
3201 | unsigned long zone_start_pfn, zone_end_pfn; | |
3202 | ||
3203 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3204 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3205 | node_start_pfn); | |
3206 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3207 | node_end_pfn); | |
3208 | ||
2a1e274a MG |
3209 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3210 | node_start_pfn, node_end_pfn, | |
3211 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3212 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3213 | } |
0e0b864e | 3214 | |
c713216d | 3215 | #else |
6ea6e688 | 3216 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3217 | unsigned long zone_type, |
3218 | unsigned long *zones_size) | |
3219 | { | |
3220 | return zones_size[zone_type]; | |
3221 | } | |
3222 | ||
6ea6e688 | 3223 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3224 | unsigned long zone_type, |
3225 | unsigned long *zholes_size) | |
3226 | { | |
3227 | if (!zholes_size) | |
3228 | return 0; | |
3229 | ||
3230 | return zholes_size[zone_type]; | |
3231 | } | |
0e0b864e | 3232 | |
c713216d MG |
3233 | #endif |
3234 | ||
a3142c8e | 3235 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3236 | unsigned long *zones_size, unsigned long *zholes_size) |
3237 | { | |
3238 | unsigned long realtotalpages, totalpages = 0; | |
3239 | enum zone_type i; | |
3240 | ||
3241 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3242 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3243 | zones_size); | |
3244 | pgdat->node_spanned_pages = totalpages; | |
3245 | ||
3246 | realtotalpages = totalpages; | |
3247 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3248 | realtotalpages -= | |
3249 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3250 | zholes_size); | |
3251 | pgdat->node_present_pages = realtotalpages; | |
3252 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3253 | realtotalpages); | |
3254 | } | |
3255 | ||
835c134e MG |
3256 | #ifndef CONFIG_SPARSEMEM |
3257 | /* | |
3258 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3259 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3260 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3261 | * round what is now in bits to nearest long in bits, then return it in |
3262 | * bytes. | |
3263 | */ | |
3264 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3265 | { | |
3266 | unsigned long usemapsize; | |
3267 | ||
d9c23400 MG |
3268 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3269 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3270 | usemapsize *= NR_PAGEBLOCK_BITS; |
3271 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3272 | ||
3273 | return usemapsize / 8; | |
3274 | } | |
3275 | ||
3276 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3277 | struct zone *zone, unsigned long zonesize) | |
3278 | { | |
3279 | unsigned long usemapsize = usemap_size(zonesize); | |
3280 | zone->pageblock_flags = NULL; | |
3281 | if (usemapsize) { | |
3282 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3283 | memset(zone->pageblock_flags, 0, usemapsize); | |
3284 | } | |
3285 | } | |
3286 | #else | |
3287 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3288 | struct zone *zone, unsigned long zonesize) {} | |
3289 | #endif /* CONFIG_SPARSEMEM */ | |
3290 | ||
d9c23400 | 3291 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3292 | |
3293 | /* Return a sensible default order for the pageblock size. */ | |
3294 | static inline int pageblock_default_order(void) | |
3295 | { | |
3296 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3297 | return HUGETLB_PAGE_ORDER; | |
3298 | ||
3299 | return MAX_ORDER-1; | |
3300 | } | |
3301 | ||
d9c23400 MG |
3302 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3303 | static inline void __init set_pageblock_order(unsigned int order) | |
3304 | { | |
3305 | /* Check that pageblock_nr_pages has not already been setup */ | |
3306 | if (pageblock_order) | |
3307 | return; | |
3308 | ||
3309 | /* | |
3310 | * Assume the largest contiguous order of interest is a huge page. | |
3311 | * This value may be variable depending on boot parameters on IA64 | |
3312 | */ | |
3313 | pageblock_order = order; | |
3314 | } | |
3315 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3316 | ||
ba72cb8c MG |
3317 | /* |
3318 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3319 | * and pageblock_default_order() are unused as pageblock_order is set | |
3320 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3321 | * pageblock_order based on the kernel config | |
3322 | */ | |
3323 | static inline int pageblock_default_order(unsigned int order) | |
3324 | { | |
3325 | return MAX_ORDER-1; | |
3326 | } | |
d9c23400 MG |
3327 | #define set_pageblock_order(x) do {} while (0) |
3328 | ||
3329 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3330 | ||
1da177e4 LT |
3331 | /* |
3332 | * Set up the zone data structures: | |
3333 | * - mark all pages reserved | |
3334 | * - mark all memory queues empty | |
3335 | * - clear the memory bitmaps | |
3336 | */ | |
b5a0e011 | 3337 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3338 | unsigned long *zones_size, unsigned long *zholes_size) |
3339 | { | |
2f1b6248 | 3340 | enum zone_type j; |
ed8ece2e | 3341 | int nid = pgdat->node_id; |
1da177e4 | 3342 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3343 | int ret; |
1da177e4 | 3344 | |
208d54e5 | 3345 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3346 | pgdat->nr_zones = 0; |
3347 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3348 | pgdat->kswapd_max_order = 0; | |
3349 | ||
3350 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3351 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3352 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 3353 | |
c713216d MG |
3354 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3355 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3356 | zholes_size); | |
1da177e4 | 3357 | |
0e0b864e MG |
3358 | /* |
3359 | * Adjust realsize so that it accounts for how much memory | |
3360 | * is used by this zone for memmap. This affects the watermark | |
3361 | * and per-cpu initialisations | |
3362 | */ | |
f7232154 JW |
3363 | memmap_pages = |
3364 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
3365 | if (realsize >= memmap_pages) { |
3366 | realsize -= memmap_pages; | |
6b74ab97 MG |
3367 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3368 | "%s zone: %lu pages used for memmap\n", | |
0e0b864e MG |
3369 | zone_names[j], memmap_pages); |
3370 | } else | |
3371 | printk(KERN_WARNING | |
3372 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3373 | zone_names[j], memmap_pages, realsize); | |
3374 | ||
6267276f CL |
3375 | /* Account for reserved pages */ |
3376 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3377 | realsize -= dma_reserve; |
6b74ab97 MG |
3378 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3379 | "%s zone: %lu pages reserved\n", | |
6267276f | 3380 | zone_names[0], dma_reserve); |
0e0b864e MG |
3381 | } |
3382 | ||
98d2b0eb | 3383 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3384 | nr_kernel_pages += realsize; |
3385 | nr_all_pages += realsize; | |
3386 | ||
3387 | zone->spanned_pages = size; | |
3388 | zone->present_pages = realsize; | |
9614634f | 3389 | #ifdef CONFIG_NUMA |
d5f541ed | 3390 | zone->node = nid; |
8417bba4 | 3391 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3392 | / 100; |
0ff38490 | 3393 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3394 | #endif |
1da177e4 LT |
3395 | zone->name = zone_names[j]; |
3396 | spin_lock_init(&zone->lock); | |
3397 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3398 | zone_seqlock_init(zone); |
1da177e4 | 3399 | zone->zone_pgdat = pgdat; |
1da177e4 | 3400 | |
3bb1a852 | 3401 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3402 | |
ed8ece2e | 3403 | zone_pcp_init(zone); |
1da177e4 LT |
3404 | INIT_LIST_HEAD(&zone->active_list); |
3405 | INIT_LIST_HEAD(&zone->inactive_list); | |
3406 | zone->nr_scan_active = 0; | |
3407 | zone->nr_scan_inactive = 0; | |
2244b95a | 3408 | zap_zone_vm_stats(zone); |
e815af95 | 3409 | zone->flags = 0; |
1da177e4 LT |
3410 | if (!size) |
3411 | continue; | |
3412 | ||
ba72cb8c | 3413 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3414 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3415 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3416 | size, MEMMAP_EARLY); | |
718127cc | 3417 | BUG_ON(ret); |
76cdd58e | 3418 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 3419 | zone_start_pfn += size; |
1da177e4 LT |
3420 | } |
3421 | } | |
3422 | ||
577a32f6 | 3423 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3424 | { |
1da177e4 LT |
3425 | /* Skip empty nodes */ |
3426 | if (!pgdat->node_spanned_pages) | |
3427 | return; | |
3428 | ||
d41dee36 | 3429 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3430 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3431 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3432 | unsigned long size, start, end; |
d41dee36 AW |
3433 | struct page *map; |
3434 | ||
e984bb43 BP |
3435 | /* |
3436 | * The zone's endpoints aren't required to be MAX_ORDER | |
3437 | * aligned but the node_mem_map endpoints must be in order | |
3438 | * for the buddy allocator to function correctly. | |
3439 | */ | |
3440 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3441 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3442 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3443 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3444 | map = alloc_remap(pgdat->node_id, size); |
3445 | if (!map) | |
3446 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3447 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3448 | } |
12d810c1 | 3449 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3450 | /* |
3451 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3452 | */ | |
c713216d | 3453 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3454 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3455 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3456 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3457 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3458 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3459 | } | |
1da177e4 | 3460 | #endif |
d41dee36 | 3461 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3462 | } |
3463 | ||
9109fb7b JW |
3464 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
3465 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 3466 | { |
9109fb7b JW |
3467 | pg_data_t *pgdat = NODE_DATA(nid); |
3468 | ||
1da177e4 LT |
3469 | pgdat->node_id = nid; |
3470 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3471 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3472 | |
3473 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
3474 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
3475 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3476 | nid, (unsigned long)pgdat, | |
3477 | (unsigned long)pgdat->node_mem_map); | |
3478 | #endif | |
1da177e4 LT |
3479 | |
3480 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3481 | } | |
3482 | ||
c713216d | 3483 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3484 | |
3485 | #if MAX_NUMNODES > 1 | |
3486 | /* | |
3487 | * Figure out the number of possible node ids. | |
3488 | */ | |
3489 | static void __init setup_nr_node_ids(void) | |
3490 | { | |
3491 | unsigned int node; | |
3492 | unsigned int highest = 0; | |
3493 | ||
3494 | for_each_node_mask(node, node_possible_map) | |
3495 | highest = node; | |
3496 | nr_node_ids = highest + 1; | |
3497 | } | |
3498 | #else | |
3499 | static inline void setup_nr_node_ids(void) | |
3500 | { | |
3501 | } | |
3502 | #endif | |
3503 | ||
c713216d MG |
3504 | /** |
3505 | * add_active_range - Register a range of PFNs backed by physical memory | |
3506 | * @nid: The node ID the range resides on | |
3507 | * @start_pfn: The start PFN of the available physical memory | |
3508 | * @end_pfn: The end PFN of the available physical memory | |
3509 | * | |
3510 | * These ranges are stored in an early_node_map[] and later used by | |
3511 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3512 | * range spans a memory hole, it is up to the architecture to ensure | |
3513 | * the memory is not freed by the bootmem allocator. If possible | |
3514 | * the range being registered will be merged with existing ranges. | |
3515 | */ | |
3516 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3517 | unsigned long end_pfn) | |
3518 | { | |
3519 | int i; | |
3520 | ||
6b74ab97 MG |
3521 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
3522 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3523 | "%d entries of %d used\n", | |
3524 | nid, start_pfn, end_pfn, | |
3525 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 3526 | |
2dbb51c4 MG |
3527 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
3528 | ||
c713216d MG |
3529 | /* Merge with existing active regions if possible */ |
3530 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3531 | if (early_node_map[i].nid != nid) | |
3532 | continue; | |
3533 | ||
3534 | /* Skip if an existing region covers this new one */ | |
3535 | if (start_pfn >= early_node_map[i].start_pfn && | |
3536 | end_pfn <= early_node_map[i].end_pfn) | |
3537 | return; | |
3538 | ||
3539 | /* Merge forward if suitable */ | |
3540 | if (start_pfn <= early_node_map[i].end_pfn && | |
3541 | end_pfn > early_node_map[i].end_pfn) { | |
3542 | early_node_map[i].end_pfn = end_pfn; | |
3543 | return; | |
3544 | } | |
3545 | ||
3546 | /* Merge backward if suitable */ | |
3547 | if (start_pfn < early_node_map[i].end_pfn && | |
3548 | end_pfn >= early_node_map[i].start_pfn) { | |
3549 | early_node_map[i].start_pfn = start_pfn; | |
3550 | return; | |
3551 | } | |
3552 | } | |
3553 | ||
3554 | /* Check that early_node_map is large enough */ | |
3555 | if (i >= MAX_ACTIVE_REGIONS) { | |
3556 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3557 | MAX_ACTIVE_REGIONS); | |
3558 | return; | |
3559 | } | |
3560 | ||
3561 | early_node_map[i].nid = nid; | |
3562 | early_node_map[i].start_pfn = start_pfn; | |
3563 | early_node_map[i].end_pfn = end_pfn; | |
3564 | nr_nodemap_entries = i + 1; | |
3565 | } | |
3566 | ||
3567 | /** | |
cc1050ba | 3568 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 3569 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
3570 | * @start_pfn: The new PFN of the range |
3571 | * @end_pfn: The new PFN of the range | |
c713216d MG |
3572 | * |
3573 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
3574 | * The map is kept near the end physical page range that has already been |
3575 | * registered. This function allows an arch to shrink an existing registered | |
3576 | * range. | |
c713216d | 3577 | */ |
cc1050ba YL |
3578 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
3579 | unsigned long end_pfn) | |
c713216d | 3580 | { |
cc1a9d86 YL |
3581 | int i, j; |
3582 | int removed = 0; | |
c713216d | 3583 | |
cc1050ba YL |
3584 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
3585 | nid, start_pfn, end_pfn); | |
3586 | ||
c713216d | 3587 | /* Find the old active region end and shrink */ |
cc1a9d86 | 3588 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
3589 | if (early_node_map[i].start_pfn >= start_pfn && |
3590 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 3591 | /* clear it */ |
cc1050ba | 3592 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
3593 | early_node_map[i].end_pfn = 0; |
3594 | removed = 1; | |
3595 | continue; | |
3596 | } | |
cc1050ba YL |
3597 | if (early_node_map[i].start_pfn < start_pfn && |
3598 | early_node_map[i].end_pfn > start_pfn) { | |
3599 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
3600 | early_node_map[i].end_pfn = start_pfn; | |
3601 | if (temp_end_pfn > end_pfn) | |
3602 | add_active_range(nid, end_pfn, temp_end_pfn); | |
3603 | continue; | |
3604 | } | |
3605 | if (early_node_map[i].start_pfn >= start_pfn && | |
3606 | early_node_map[i].end_pfn > end_pfn && | |
3607 | early_node_map[i].start_pfn < end_pfn) { | |
3608 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 3609 | continue; |
c713216d | 3610 | } |
cc1a9d86 YL |
3611 | } |
3612 | ||
3613 | if (!removed) | |
3614 | return; | |
3615 | ||
3616 | /* remove the blank ones */ | |
3617 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
3618 | if (early_node_map[i].nid != nid) | |
3619 | continue; | |
3620 | if (early_node_map[i].end_pfn) | |
3621 | continue; | |
3622 | /* we found it, get rid of it */ | |
3623 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
3624 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
3625 | sizeof(early_node_map[j])); | |
3626 | j = nr_nodemap_entries - 1; | |
3627 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
3628 | nr_nodemap_entries--; | |
3629 | } | |
c713216d MG |
3630 | } |
3631 | ||
3632 | /** | |
3633 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3634 | * |
c713216d MG |
3635 | * During discovery, it may be found that a table like SRAT is invalid |
3636 | * and an alternative discovery method must be used. This function removes | |
3637 | * all currently registered regions. | |
3638 | */ | |
88ca3b94 | 3639 | void __init remove_all_active_ranges(void) |
c713216d MG |
3640 | { |
3641 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3642 | nr_nodemap_entries = 0; | |
fb01439c MG |
3643 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3644 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3645 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3646 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3647 | } |
3648 | ||
3649 | /* Compare two active node_active_regions */ | |
3650 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3651 | { | |
3652 | struct node_active_region *arange = (struct node_active_region *)a; | |
3653 | struct node_active_region *brange = (struct node_active_region *)b; | |
3654 | ||
3655 | /* Done this way to avoid overflows */ | |
3656 | if (arange->start_pfn > brange->start_pfn) | |
3657 | return 1; | |
3658 | if (arange->start_pfn < brange->start_pfn) | |
3659 | return -1; | |
3660 | ||
3661 | return 0; | |
3662 | } | |
3663 | ||
3664 | /* sort the node_map by start_pfn */ | |
3665 | static void __init sort_node_map(void) | |
3666 | { | |
3667 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3668 | sizeof(struct node_active_region), | |
3669 | cmp_node_active_region, NULL); | |
3670 | } | |
3671 | ||
a6af2bc3 | 3672 | /* Find the lowest pfn for a node */ |
2bc0d261 | 3673 | unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
3674 | { |
3675 | int i; | |
a6af2bc3 | 3676 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3677 | |
c713216d MG |
3678 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3679 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3680 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3681 | |
a6af2bc3 MG |
3682 | if (min_pfn == ULONG_MAX) { |
3683 | printk(KERN_WARNING | |
2bc0d261 | 3684 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
3685 | return 0; |
3686 | } | |
3687 | ||
3688 | return min_pfn; | |
c713216d MG |
3689 | } |
3690 | ||
3691 | /** | |
3692 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3693 | * | |
3694 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3695 | * add_active_range(). |
c713216d MG |
3696 | */ |
3697 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3698 | { | |
3699 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3700 | } | |
3701 | ||
3702 | /** | |
3703 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3704 | * | |
3705 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3706 | * add_active_range(). |
c713216d MG |
3707 | */ |
3708 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3709 | { | |
3710 | int i; | |
3711 | unsigned long max_pfn = 0; | |
3712 | ||
3713 | for (i = 0; i < nr_nodemap_entries; i++) | |
3714 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3715 | ||
3716 | return max_pfn; | |
3717 | } | |
3718 | ||
37b07e41 LS |
3719 | /* |
3720 | * early_calculate_totalpages() | |
3721 | * Sum pages in active regions for movable zone. | |
3722 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3723 | */ | |
484f51f8 | 3724 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
3725 | { |
3726 | int i; | |
3727 | unsigned long totalpages = 0; | |
3728 | ||
37b07e41 LS |
3729 | for (i = 0; i < nr_nodemap_entries; i++) { |
3730 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3731 | early_node_map[i].start_pfn; |
37b07e41 LS |
3732 | totalpages += pages; |
3733 | if (pages) | |
3734 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3735 | } | |
3736 | return totalpages; | |
7e63efef MG |
3737 | } |
3738 | ||
2a1e274a MG |
3739 | /* |
3740 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3741 | * is spread evenly between nodes as long as the nodes have enough | |
3742 | * memory. When they don't, some nodes will have more kernelcore than | |
3743 | * others | |
3744 | */ | |
3745 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3746 | { | |
3747 | int i, nid; | |
3748 | unsigned long usable_startpfn; | |
3749 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3750 | unsigned long totalpages = early_calculate_totalpages(); |
3751 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3752 | |
7e63efef MG |
3753 | /* |
3754 | * If movablecore was specified, calculate what size of | |
3755 | * kernelcore that corresponds so that memory usable for | |
3756 | * any allocation type is evenly spread. If both kernelcore | |
3757 | * and movablecore are specified, then the value of kernelcore | |
3758 | * will be used for required_kernelcore if it's greater than | |
3759 | * what movablecore would have allowed. | |
3760 | */ | |
3761 | if (required_movablecore) { | |
7e63efef MG |
3762 | unsigned long corepages; |
3763 | ||
3764 | /* | |
3765 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3766 | * was requested by the user | |
3767 | */ | |
3768 | required_movablecore = | |
3769 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3770 | corepages = totalpages - required_movablecore; | |
3771 | ||
3772 | required_kernelcore = max(required_kernelcore, corepages); | |
3773 | } | |
3774 | ||
2a1e274a MG |
3775 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3776 | if (!required_kernelcore) | |
3777 | return; | |
3778 | ||
3779 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3780 | find_usable_zone_for_movable(); | |
3781 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3782 | ||
3783 | restart: | |
3784 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3785 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3786 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3787 | /* |
3788 | * Recalculate kernelcore_node if the division per node | |
3789 | * now exceeds what is necessary to satisfy the requested | |
3790 | * amount of memory for the kernel | |
3791 | */ | |
3792 | if (required_kernelcore < kernelcore_node) | |
3793 | kernelcore_node = required_kernelcore / usable_nodes; | |
3794 | ||
3795 | /* | |
3796 | * As the map is walked, we track how much memory is usable | |
3797 | * by the kernel using kernelcore_remaining. When it is | |
3798 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3799 | */ | |
3800 | kernelcore_remaining = kernelcore_node; | |
3801 | ||
3802 | /* Go through each range of PFNs within this node */ | |
3803 | for_each_active_range_index_in_nid(i, nid) { | |
3804 | unsigned long start_pfn, end_pfn; | |
3805 | unsigned long size_pages; | |
3806 | ||
3807 | start_pfn = max(early_node_map[i].start_pfn, | |
3808 | zone_movable_pfn[nid]); | |
3809 | end_pfn = early_node_map[i].end_pfn; | |
3810 | if (start_pfn >= end_pfn) | |
3811 | continue; | |
3812 | ||
3813 | /* Account for what is only usable for kernelcore */ | |
3814 | if (start_pfn < usable_startpfn) { | |
3815 | unsigned long kernel_pages; | |
3816 | kernel_pages = min(end_pfn, usable_startpfn) | |
3817 | - start_pfn; | |
3818 | ||
3819 | kernelcore_remaining -= min(kernel_pages, | |
3820 | kernelcore_remaining); | |
3821 | required_kernelcore -= min(kernel_pages, | |
3822 | required_kernelcore); | |
3823 | ||
3824 | /* Continue if range is now fully accounted */ | |
3825 | if (end_pfn <= usable_startpfn) { | |
3826 | ||
3827 | /* | |
3828 | * Push zone_movable_pfn to the end so | |
3829 | * that if we have to rebalance | |
3830 | * kernelcore across nodes, we will | |
3831 | * not double account here | |
3832 | */ | |
3833 | zone_movable_pfn[nid] = end_pfn; | |
3834 | continue; | |
3835 | } | |
3836 | start_pfn = usable_startpfn; | |
3837 | } | |
3838 | ||
3839 | /* | |
3840 | * The usable PFN range for ZONE_MOVABLE is from | |
3841 | * start_pfn->end_pfn. Calculate size_pages as the | |
3842 | * number of pages used as kernelcore | |
3843 | */ | |
3844 | size_pages = end_pfn - start_pfn; | |
3845 | if (size_pages > kernelcore_remaining) | |
3846 | size_pages = kernelcore_remaining; | |
3847 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3848 | ||
3849 | /* | |
3850 | * Some kernelcore has been met, update counts and | |
3851 | * break if the kernelcore for this node has been | |
3852 | * satisified | |
3853 | */ | |
3854 | required_kernelcore -= min(required_kernelcore, | |
3855 | size_pages); | |
3856 | kernelcore_remaining -= size_pages; | |
3857 | if (!kernelcore_remaining) | |
3858 | break; | |
3859 | } | |
3860 | } | |
3861 | ||
3862 | /* | |
3863 | * If there is still required_kernelcore, we do another pass with one | |
3864 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3865 | * along on the nodes that still have memory until kernelcore is | |
3866 | * satisified | |
3867 | */ | |
3868 | usable_nodes--; | |
3869 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3870 | goto restart; | |
3871 | ||
3872 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3873 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3874 | zone_movable_pfn[nid] = | |
3875 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3876 | } | |
3877 | ||
37b07e41 LS |
3878 | /* Any regular memory on that node ? */ |
3879 | static void check_for_regular_memory(pg_data_t *pgdat) | |
3880 | { | |
3881 | #ifdef CONFIG_HIGHMEM | |
3882 | enum zone_type zone_type; | |
3883 | ||
3884 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
3885 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3886 | if (zone->present_pages) | |
3887 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
3888 | } | |
3889 | #endif | |
3890 | } | |
3891 | ||
c713216d MG |
3892 | /** |
3893 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3894 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3895 | * |
3896 | * This will call free_area_init_node() for each active node in the system. | |
3897 | * Using the page ranges provided by add_active_range(), the size of each | |
3898 | * zone in each node and their holes is calculated. If the maximum PFN | |
3899 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3900 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3901 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3902 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3903 | * at arch_max_dma_pfn. | |
3904 | */ | |
3905 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3906 | { | |
3907 | unsigned long nid; | |
3908 | enum zone_type i; | |
3909 | ||
a6af2bc3 MG |
3910 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3911 | sort_node_map(); | |
3912 | ||
c713216d MG |
3913 | /* Record where the zone boundaries are */ |
3914 | memset(arch_zone_lowest_possible_pfn, 0, | |
3915 | sizeof(arch_zone_lowest_possible_pfn)); | |
3916 | memset(arch_zone_highest_possible_pfn, 0, | |
3917 | sizeof(arch_zone_highest_possible_pfn)); | |
3918 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3919 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3920 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3921 | if (i == ZONE_MOVABLE) |
3922 | continue; | |
c713216d MG |
3923 | arch_zone_lowest_possible_pfn[i] = |
3924 | arch_zone_highest_possible_pfn[i-1]; | |
3925 | arch_zone_highest_possible_pfn[i] = | |
3926 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3927 | } | |
2a1e274a MG |
3928 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3929 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3930 | ||
3931 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3932 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3933 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3934 | |
c713216d MG |
3935 | /* Print out the zone ranges */ |
3936 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3937 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3938 | if (i == ZONE_MOVABLE) | |
3939 | continue; | |
5dab8ec1 | 3940 | printk(" %-8s %0#10lx -> %0#10lx\n", |
c713216d MG |
3941 | zone_names[i], |
3942 | arch_zone_lowest_possible_pfn[i], | |
3943 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3944 | } |
3945 | ||
3946 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3947 | printk("Movable zone start PFN for each node\n"); | |
3948 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3949 | if (zone_movable_pfn[i]) | |
3950 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3951 | } | |
c713216d MG |
3952 | |
3953 | /* Print out the early_node_map[] */ | |
3954 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3955 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 3956 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
3957 | early_node_map[i].start_pfn, |
3958 | early_node_map[i].end_pfn); | |
3959 | ||
3960 | /* Initialise every node */ | |
708614e6 | 3961 | mminit_verify_pageflags_layout(); |
8ef82866 | 3962 | setup_nr_node_ids(); |
c713216d MG |
3963 | for_each_online_node(nid) { |
3964 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 3965 | free_area_init_node(nid, NULL, |
c713216d | 3966 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
3967 | |
3968 | /* Any memory on that node */ | |
3969 | if (pgdat->node_present_pages) | |
3970 | node_set_state(nid, N_HIGH_MEMORY); | |
3971 | check_for_regular_memory(pgdat); | |
c713216d MG |
3972 | } |
3973 | } | |
2a1e274a | 3974 | |
7e63efef | 3975 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3976 | { |
3977 | unsigned long long coremem; | |
3978 | if (!p) | |
3979 | return -EINVAL; | |
3980 | ||
3981 | coremem = memparse(p, &p); | |
7e63efef | 3982 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3983 | |
7e63efef | 3984 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3985 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3986 | ||
3987 | return 0; | |
3988 | } | |
ed7ed365 | 3989 | |
7e63efef MG |
3990 | /* |
3991 | * kernelcore=size sets the amount of memory for use for allocations that | |
3992 | * cannot be reclaimed or migrated. | |
3993 | */ | |
3994 | static int __init cmdline_parse_kernelcore(char *p) | |
3995 | { | |
3996 | return cmdline_parse_core(p, &required_kernelcore); | |
3997 | } | |
3998 | ||
3999 | /* | |
4000 | * movablecore=size sets the amount of memory for use for allocations that | |
4001 | * can be reclaimed or migrated. | |
4002 | */ | |
4003 | static int __init cmdline_parse_movablecore(char *p) | |
4004 | { | |
4005 | return cmdline_parse_core(p, &required_movablecore); | |
4006 | } | |
4007 | ||
ed7ed365 | 4008 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4009 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4010 | |
c713216d MG |
4011 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4012 | ||
0e0b864e | 4013 | /** |
88ca3b94 RD |
4014 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4015 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4016 | * |
4017 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4018 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4019 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4020 | * function may optionally be used to account for unfreeable pages in the |
4021 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4022 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4023 | */ |
4024 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4025 | { | |
4026 | dma_reserve = new_dma_reserve; | |
4027 | } | |
4028 | ||
93b7504e | 4029 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
b61bfa3c | 4030 | struct pglist_data contig_page_data = { .bdata = &bootmem_node_data[0] }; |
1da177e4 | 4031 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4032 | #endif |
1da177e4 LT |
4033 | |
4034 | void __init free_area_init(unsigned long *zones_size) | |
4035 | { | |
9109fb7b | 4036 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4037 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4038 | } | |
1da177e4 | 4039 | |
1da177e4 LT |
4040 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4041 | unsigned long action, void *hcpu) | |
4042 | { | |
4043 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4044 | |
8bb78442 | 4045 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4046 | drain_pages(cpu); |
4047 | ||
4048 | /* | |
4049 | * Spill the event counters of the dead processor | |
4050 | * into the current processors event counters. | |
4051 | * This artificially elevates the count of the current | |
4052 | * processor. | |
4053 | */ | |
f8891e5e | 4054 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4055 | |
4056 | /* | |
4057 | * Zero the differential counters of the dead processor | |
4058 | * so that the vm statistics are consistent. | |
4059 | * | |
4060 | * This is only okay since the processor is dead and cannot | |
4061 | * race with what we are doing. | |
4062 | */ | |
2244b95a | 4063 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4064 | } |
4065 | return NOTIFY_OK; | |
4066 | } | |
1da177e4 LT |
4067 | |
4068 | void __init page_alloc_init(void) | |
4069 | { | |
4070 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4071 | } | |
4072 | ||
cb45b0e9 HA |
4073 | /* |
4074 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4075 | * or min_free_kbytes changes. | |
4076 | */ | |
4077 | static void calculate_totalreserve_pages(void) | |
4078 | { | |
4079 | struct pglist_data *pgdat; | |
4080 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4081 | enum zone_type i, j; |
cb45b0e9 HA |
4082 | |
4083 | for_each_online_pgdat(pgdat) { | |
4084 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4085 | struct zone *zone = pgdat->node_zones + i; | |
4086 | unsigned long max = 0; | |
4087 | ||
4088 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4089 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4090 | if (zone->lowmem_reserve[j] > max) | |
4091 | max = zone->lowmem_reserve[j]; | |
4092 | } | |
4093 | ||
4094 | /* we treat pages_high as reserved pages. */ | |
4095 | max += zone->pages_high; | |
4096 | ||
4097 | if (max > zone->present_pages) | |
4098 | max = zone->present_pages; | |
4099 | reserve_pages += max; | |
4100 | } | |
4101 | } | |
4102 | totalreserve_pages = reserve_pages; | |
4103 | } | |
4104 | ||
1da177e4 LT |
4105 | /* |
4106 | * setup_per_zone_lowmem_reserve - called whenever | |
4107 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4108 | * has a correct pages reserved value, so an adequate number of | |
4109 | * pages are left in the zone after a successful __alloc_pages(). | |
4110 | */ | |
4111 | static void setup_per_zone_lowmem_reserve(void) | |
4112 | { | |
4113 | struct pglist_data *pgdat; | |
2f6726e5 | 4114 | enum zone_type j, idx; |
1da177e4 | 4115 | |
ec936fc5 | 4116 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4117 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4118 | struct zone *zone = pgdat->node_zones + j; | |
4119 | unsigned long present_pages = zone->present_pages; | |
4120 | ||
4121 | zone->lowmem_reserve[j] = 0; | |
4122 | ||
2f6726e5 CL |
4123 | idx = j; |
4124 | while (idx) { | |
1da177e4 LT |
4125 | struct zone *lower_zone; |
4126 | ||
2f6726e5 CL |
4127 | idx--; |
4128 | ||
1da177e4 LT |
4129 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4130 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4131 | ||
4132 | lower_zone = pgdat->node_zones + idx; | |
4133 | lower_zone->lowmem_reserve[j] = present_pages / | |
4134 | sysctl_lowmem_reserve_ratio[idx]; | |
4135 | present_pages += lower_zone->present_pages; | |
4136 | } | |
4137 | } | |
4138 | } | |
cb45b0e9 HA |
4139 | |
4140 | /* update totalreserve_pages */ | |
4141 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4142 | } |
4143 | ||
88ca3b94 RD |
4144 | /** |
4145 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
4146 | * | |
4147 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
4148 | * with respect to min_free_kbytes. | |
1da177e4 | 4149 | */ |
3947be19 | 4150 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
4151 | { |
4152 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4153 | unsigned long lowmem_pages = 0; | |
4154 | struct zone *zone; | |
4155 | unsigned long flags; | |
4156 | ||
4157 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4158 | for_each_zone(zone) { | |
4159 | if (!is_highmem(zone)) | |
4160 | lowmem_pages += zone->present_pages; | |
4161 | } | |
4162 | ||
4163 | for_each_zone(zone) { | |
ac924c60 AM |
4164 | u64 tmp; |
4165 | ||
1da177e4 | 4166 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
4167 | tmp = (u64)pages_min * zone->present_pages; |
4168 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4169 | if (is_highmem(zone)) { |
4170 | /* | |
669ed175 NP |
4171 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4172 | * need highmem pages, so cap pages_min to a small | |
4173 | * value here. | |
4174 | * | |
4175 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
4176 | * deltas controls asynch page reclaim, and so should | |
4177 | * not be capped for highmem. | |
1da177e4 LT |
4178 | */ |
4179 | int min_pages; | |
4180 | ||
4181 | min_pages = zone->present_pages / 1024; | |
4182 | if (min_pages < SWAP_CLUSTER_MAX) | |
4183 | min_pages = SWAP_CLUSTER_MAX; | |
4184 | if (min_pages > 128) | |
4185 | min_pages = 128; | |
4186 | zone->pages_min = min_pages; | |
4187 | } else { | |
669ed175 NP |
4188 | /* |
4189 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4190 | * proportionate to the zone's size. |
4191 | */ | |
669ed175 | 4192 | zone->pages_min = tmp; |
1da177e4 LT |
4193 | } |
4194 | ||
ac924c60 AM |
4195 | zone->pages_low = zone->pages_min + (tmp >> 2); |
4196 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
56fd56b8 | 4197 | setup_zone_migrate_reserve(zone); |
1da177e4 LT |
4198 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
4199 | } | |
cb45b0e9 HA |
4200 | |
4201 | /* update totalreserve_pages */ | |
4202 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4203 | } |
4204 | ||
4205 | /* | |
4206 | * Initialise min_free_kbytes. | |
4207 | * | |
4208 | * For small machines we want it small (128k min). For large machines | |
4209 | * we want it large (64MB max). But it is not linear, because network | |
4210 | * bandwidth does not increase linearly with machine size. We use | |
4211 | * | |
4212 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4213 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4214 | * | |
4215 | * which yields | |
4216 | * | |
4217 | * 16MB: 512k | |
4218 | * 32MB: 724k | |
4219 | * 64MB: 1024k | |
4220 | * 128MB: 1448k | |
4221 | * 256MB: 2048k | |
4222 | * 512MB: 2896k | |
4223 | * 1024MB: 4096k | |
4224 | * 2048MB: 5792k | |
4225 | * 4096MB: 8192k | |
4226 | * 8192MB: 11584k | |
4227 | * 16384MB: 16384k | |
4228 | */ | |
4229 | static int __init init_per_zone_pages_min(void) | |
4230 | { | |
4231 | unsigned long lowmem_kbytes; | |
4232 | ||
4233 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4234 | ||
4235 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4236 | if (min_free_kbytes < 128) | |
4237 | min_free_kbytes = 128; | |
4238 | if (min_free_kbytes > 65536) | |
4239 | min_free_kbytes = 65536; | |
4240 | setup_per_zone_pages_min(); | |
4241 | setup_per_zone_lowmem_reserve(); | |
4242 | return 0; | |
4243 | } | |
4244 | module_init(init_per_zone_pages_min) | |
4245 | ||
4246 | /* | |
4247 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4248 | * that we can call two helper functions whenever min_free_kbytes | |
4249 | * changes. | |
4250 | */ | |
4251 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4252 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4253 | { | |
4254 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
4255 | if (write) |
4256 | setup_per_zone_pages_min(); | |
1da177e4 LT |
4257 | return 0; |
4258 | } | |
4259 | ||
9614634f CL |
4260 | #ifdef CONFIG_NUMA |
4261 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4262 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4263 | { | |
4264 | struct zone *zone; | |
4265 | int rc; | |
4266 | ||
4267 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4268 | if (rc) | |
4269 | return rc; | |
4270 | ||
4271 | for_each_zone(zone) | |
8417bba4 | 4272 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4273 | sysctl_min_unmapped_ratio) / 100; |
4274 | return 0; | |
4275 | } | |
0ff38490 CL |
4276 | |
4277 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4278 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4279 | { | |
4280 | struct zone *zone; | |
4281 | int rc; | |
4282 | ||
4283 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4284 | if (rc) | |
4285 | return rc; | |
4286 | ||
4287 | for_each_zone(zone) | |
4288 | zone->min_slab_pages = (zone->present_pages * | |
4289 | sysctl_min_slab_ratio) / 100; | |
4290 | return 0; | |
4291 | } | |
9614634f CL |
4292 | #endif |
4293 | ||
1da177e4 LT |
4294 | /* |
4295 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4296 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4297 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4298 | * | |
4299 | * The reserve ratio obviously has absolutely no relation with the | |
4300 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
4301 | * if in function of the boot time zone sizes. | |
4302 | */ | |
4303 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4304 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4305 | { | |
4306 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4307 | setup_per_zone_lowmem_reserve(); | |
4308 | return 0; | |
4309 | } | |
4310 | ||
8ad4b1fb RS |
4311 | /* |
4312 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4313 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4314 | * can have before it gets flushed back to buddy allocator. | |
4315 | */ | |
4316 | ||
4317 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4318 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4319 | { | |
4320 | struct zone *zone; | |
4321 | unsigned int cpu; | |
4322 | int ret; | |
4323 | ||
4324 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4325 | if (!write || (ret == -EINVAL)) | |
4326 | return ret; | |
4327 | for_each_zone(zone) { | |
4328 | for_each_online_cpu(cpu) { | |
4329 | unsigned long high; | |
4330 | high = zone->present_pages / percpu_pagelist_fraction; | |
4331 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4332 | } | |
4333 | } | |
4334 | return 0; | |
4335 | } | |
4336 | ||
f034b5d4 | 4337 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4338 | |
4339 | #ifdef CONFIG_NUMA | |
4340 | static int __init set_hashdist(char *str) | |
4341 | { | |
4342 | if (!str) | |
4343 | return 0; | |
4344 | hashdist = simple_strtoul(str, &str, 0); | |
4345 | return 1; | |
4346 | } | |
4347 | __setup("hashdist=", set_hashdist); | |
4348 | #endif | |
4349 | ||
4350 | /* | |
4351 | * allocate a large system hash table from bootmem | |
4352 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4353 | * quantity of entries | |
4354 | * - limit is the number of hash buckets, not the total allocation size | |
4355 | */ | |
4356 | void *__init alloc_large_system_hash(const char *tablename, | |
4357 | unsigned long bucketsize, | |
4358 | unsigned long numentries, | |
4359 | int scale, | |
4360 | int flags, | |
4361 | unsigned int *_hash_shift, | |
4362 | unsigned int *_hash_mask, | |
4363 | unsigned long limit) | |
4364 | { | |
4365 | unsigned long long max = limit; | |
4366 | unsigned long log2qty, size; | |
4367 | void *table = NULL; | |
4368 | ||
4369 | /* allow the kernel cmdline to have a say */ | |
4370 | if (!numentries) { | |
4371 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4372 | numentries = nr_kernel_pages; |
1da177e4 LT |
4373 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4374 | numentries >>= 20 - PAGE_SHIFT; | |
4375 | numentries <<= 20 - PAGE_SHIFT; | |
4376 | ||
4377 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4378 | if (scale > PAGE_SHIFT) | |
4379 | numentries >>= (scale - PAGE_SHIFT); | |
4380 | else | |
4381 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4382 | |
4383 | /* Make sure we've got at least a 0-order allocation.. */ | |
4384 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4385 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4386 | } |
6e692ed3 | 4387 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4388 | |
4389 | /* limit allocation size to 1/16 total memory by default */ | |
4390 | if (max == 0) { | |
4391 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4392 | do_div(max, bucketsize); | |
4393 | } | |
4394 | ||
4395 | if (numentries > max) | |
4396 | numentries = max; | |
4397 | ||
f0d1b0b3 | 4398 | log2qty = ilog2(numentries); |
1da177e4 LT |
4399 | |
4400 | do { | |
4401 | size = bucketsize << log2qty; | |
4402 | if (flags & HASH_EARLY) | |
4403 | table = alloc_bootmem(size); | |
4404 | else if (hashdist) | |
4405 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4406 | else { | |
2309f9e6 | 4407 | unsigned long order = get_order(size); |
1da177e4 | 4408 | table = (void*) __get_free_pages(GFP_ATOMIC, order); |
1037b83b ED |
4409 | /* |
4410 | * If bucketsize is not a power-of-two, we may free | |
4411 | * some pages at the end of hash table. | |
4412 | */ | |
4413 | if (table) { | |
4414 | unsigned long alloc_end = (unsigned long)table + | |
4415 | (PAGE_SIZE << order); | |
4416 | unsigned long used = (unsigned long)table + | |
4417 | PAGE_ALIGN(size); | |
4418 | split_page(virt_to_page(table), order); | |
4419 | while (used < alloc_end) { | |
4420 | free_page(used); | |
4421 | used += PAGE_SIZE; | |
4422 | } | |
4423 | } | |
1da177e4 LT |
4424 | } |
4425 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4426 | ||
4427 | if (!table) | |
4428 | panic("Failed to allocate %s hash table\n", tablename); | |
4429 | ||
b49ad484 | 4430 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4431 | tablename, |
4432 | (1U << log2qty), | |
f0d1b0b3 | 4433 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4434 | size); |
4435 | ||
4436 | if (_hash_shift) | |
4437 | *_hash_shift = log2qty; | |
4438 | if (_hash_mask) | |
4439 | *_hash_mask = (1 << log2qty) - 1; | |
4440 | ||
4441 | return table; | |
4442 | } | |
a117e66e KH |
4443 | |
4444 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
4445 | struct page *pfn_to_page(unsigned long pfn) |
4446 | { | |
67de6482 | 4447 | return __pfn_to_page(pfn); |
a117e66e KH |
4448 | } |
4449 | unsigned long page_to_pfn(struct page *page) | |
4450 | { | |
67de6482 | 4451 | return __page_to_pfn(page); |
a117e66e | 4452 | } |
a117e66e KH |
4453 | EXPORT_SYMBOL(pfn_to_page); |
4454 | EXPORT_SYMBOL(page_to_pfn); | |
4455 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 4456 | |
835c134e MG |
4457 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4458 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4459 | unsigned long pfn) | |
4460 | { | |
4461 | #ifdef CONFIG_SPARSEMEM | |
4462 | return __pfn_to_section(pfn)->pageblock_flags; | |
4463 | #else | |
4464 | return zone->pageblock_flags; | |
4465 | #endif /* CONFIG_SPARSEMEM */ | |
4466 | } | |
4467 | ||
4468 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4469 | { | |
4470 | #ifdef CONFIG_SPARSEMEM | |
4471 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4472 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4473 | #else |
4474 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4475 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4476 | #endif /* CONFIG_SPARSEMEM */ |
4477 | } | |
4478 | ||
4479 | /** | |
d9c23400 | 4480 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4481 | * @page: The page within the block of interest |
4482 | * @start_bitidx: The first bit of interest to retrieve | |
4483 | * @end_bitidx: The last bit of interest | |
4484 | * returns pageblock_bits flags | |
4485 | */ | |
4486 | unsigned long get_pageblock_flags_group(struct page *page, | |
4487 | int start_bitidx, int end_bitidx) | |
4488 | { | |
4489 | struct zone *zone; | |
4490 | unsigned long *bitmap; | |
4491 | unsigned long pfn, bitidx; | |
4492 | unsigned long flags = 0; | |
4493 | unsigned long value = 1; | |
4494 | ||
4495 | zone = page_zone(page); | |
4496 | pfn = page_to_pfn(page); | |
4497 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4498 | bitidx = pfn_to_bitidx(zone, pfn); | |
4499 | ||
4500 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4501 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4502 | flags |= value; | |
6220ec78 | 4503 | |
835c134e MG |
4504 | return flags; |
4505 | } | |
4506 | ||
4507 | /** | |
d9c23400 | 4508 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4509 | * @page: The page within the block of interest |
4510 | * @start_bitidx: The first bit of interest | |
4511 | * @end_bitidx: The last bit of interest | |
4512 | * @flags: The flags to set | |
4513 | */ | |
4514 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4515 | int start_bitidx, int end_bitidx) | |
4516 | { | |
4517 | struct zone *zone; | |
4518 | unsigned long *bitmap; | |
4519 | unsigned long pfn, bitidx; | |
4520 | unsigned long value = 1; | |
4521 | ||
4522 | zone = page_zone(page); | |
4523 | pfn = page_to_pfn(page); | |
4524 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4525 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
4526 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
4527 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
4528 | |
4529 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4530 | if (flags & value) | |
4531 | __set_bit(bitidx + start_bitidx, bitmap); | |
4532 | else | |
4533 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4534 | } | |
a5d76b54 KH |
4535 | |
4536 | /* | |
4537 | * This is designed as sub function...plz see page_isolation.c also. | |
4538 | * set/clear page block's type to be ISOLATE. | |
4539 | * page allocater never alloc memory from ISOLATE block. | |
4540 | */ | |
4541 | ||
4542 | int set_migratetype_isolate(struct page *page) | |
4543 | { | |
4544 | struct zone *zone; | |
4545 | unsigned long flags; | |
4546 | int ret = -EBUSY; | |
4547 | ||
4548 | zone = page_zone(page); | |
4549 | spin_lock_irqsave(&zone->lock, flags); | |
4550 | /* | |
4551 | * In future, more migrate types will be able to be isolation target. | |
4552 | */ | |
4553 | if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE) | |
4554 | goto out; | |
4555 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
4556 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
4557 | ret = 0; | |
4558 | out: | |
4559 | spin_unlock_irqrestore(&zone->lock, flags); | |
4560 | if (!ret) | |
9f8f2172 | 4561 | drain_all_pages(); |
a5d76b54 KH |
4562 | return ret; |
4563 | } | |
4564 | ||
4565 | void unset_migratetype_isolate(struct page *page) | |
4566 | { | |
4567 | struct zone *zone; | |
4568 | unsigned long flags; | |
4569 | zone = page_zone(page); | |
4570 | spin_lock_irqsave(&zone->lock, flags); | |
4571 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
4572 | goto out; | |
4573 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4574 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4575 | out: | |
4576 | spin_unlock_irqrestore(&zone->lock, flags); | |
4577 | } | |
0c0e6195 KH |
4578 | |
4579 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
4580 | /* | |
4581 | * All pages in the range must be isolated before calling this. | |
4582 | */ | |
4583 | void | |
4584 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
4585 | { | |
4586 | struct page *page; | |
4587 | struct zone *zone; | |
4588 | int order, i; | |
4589 | unsigned long pfn; | |
4590 | unsigned long flags; | |
4591 | /* find the first valid pfn */ | |
4592 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
4593 | if (pfn_valid(pfn)) | |
4594 | break; | |
4595 | if (pfn == end_pfn) | |
4596 | return; | |
4597 | zone = page_zone(pfn_to_page(pfn)); | |
4598 | spin_lock_irqsave(&zone->lock, flags); | |
4599 | pfn = start_pfn; | |
4600 | while (pfn < end_pfn) { | |
4601 | if (!pfn_valid(pfn)) { | |
4602 | pfn++; | |
4603 | continue; | |
4604 | } | |
4605 | page = pfn_to_page(pfn); | |
4606 | BUG_ON(page_count(page)); | |
4607 | BUG_ON(!PageBuddy(page)); | |
4608 | order = page_order(page); | |
4609 | #ifdef CONFIG_DEBUG_VM | |
4610 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
4611 | pfn, 1 << order, end_pfn); | |
4612 | #endif | |
4613 | list_del(&page->lru); | |
4614 | rmv_page_order(page); | |
4615 | zone->free_area[order].nr_free--; | |
4616 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
4617 | - (1UL << order)); | |
4618 | for (i = 0; i < (1 << order); i++) | |
4619 | SetPageReserved((page+i)); | |
4620 | pfn += (1 << order); | |
4621 | } | |
4622 | spin_unlock_irqrestore(&zone->lock, flags); | |
4623 | } | |
4624 | #endif |