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> |
b1eeab67 | 26 | #include <linux/kmemcheck.h> |
1da177e4 LT |
27 | #include <linux/module.h> |
28 | #include <linux/suspend.h> | |
29 | #include <linux/pagevec.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/slab.h> | |
5a3135c2 | 32 | #include <linux/oom.h> |
1da177e4 LT |
33 | #include <linux/notifier.h> |
34 | #include <linux/topology.h> | |
35 | #include <linux/sysctl.h> | |
36 | #include <linux/cpu.h> | |
37 | #include <linux/cpuset.h> | |
bdc8cb98 | 38 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
39 | #include <linux/nodemask.h> |
40 | #include <linux/vmalloc.h> | |
4be38e35 | 41 | #include <linux/mempolicy.h> |
6811378e | 42 | #include <linux/stop_machine.h> |
c713216d MG |
43 | #include <linux/sort.h> |
44 | #include <linux/pfn.h> | |
3fcfab16 | 45 | #include <linux/backing-dev.h> |
933e312e | 46 | #include <linux/fault-inject.h> |
a5d76b54 | 47 | #include <linux/page-isolation.h> |
52d4b9ac | 48 | #include <linux/page_cgroup.h> |
3ac7fe5a | 49 | #include <linux/debugobjects.h> |
dbb1f81c | 50 | #include <linux/kmemleak.h> |
925cc71e | 51 | #include <linux/memory.h> |
0d3d062a | 52 | #include <trace/events/kmem.h> |
1da177e4 LT |
53 | |
54 | #include <asm/tlbflush.h> | |
ac924c60 | 55 | #include <asm/div64.h> |
1da177e4 LT |
56 | #include "internal.h" |
57 | ||
58 | /* | |
13808910 | 59 | * Array of node states. |
1da177e4 | 60 | */ |
13808910 CL |
61 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
62 | [N_POSSIBLE] = NODE_MASK_ALL, | |
63 | [N_ONLINE] = { { [0] = 1UL } }, | |
64 | #ifndef CONFIG_NUMA | |
65 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
66 | #ifdef CONFIG_HIGHMEM | |
67 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
68 | #endif | |
69 | [N_CPU] = { { [0] = 1UL } }, | |
70 | #endif /* NUMA */ | |
71 | }; | |
72 | EXPORT_SYMBOL(node_states); | |
73 | ||
6c231b7b | 74 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 75 | unsigned long totalreserve_pages __read_mostly; |
8ad4b1fb | 76 | int percpu_pagelist_fraction; |
dcce284a | 77 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 78 | |
d9c23400 MG |
79 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
80 | int pageblock_order __read_mostly; | |
81 | #endif | |
82 | ||
d98c7a09 | 83 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 84 | |
1da177e4 LT |
85 | /* |
86 | * results with 256, 32 in the lowmem_reserve sysctl: | |
87 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
88 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
89 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
90 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
91 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
92 | * |
93 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
94 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 95 | */ |
2f1b6248 | 96 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 97 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 98 | 256, |
4b51d669 | 99 | #endif |
fb0e7942 | 100 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 101 | 256, |
fb0e7942 | 102 | #endif |
e53ef38d | 103 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 104 | 32, |
e53ef38d | 105 | #endif |
2a1e274a | 106 | 32, |
2f1b6248 | 107 | }; |
1da177e4 LT |
108 | |
109 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 110 | |
15ad7cdc | 111 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 112 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 113 | "DMA", |
4b51d669 | 114 | #endif |
fb0e7942 | 115 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 116 | "DMA32", |
fb0e7942 | 117 | #endif |
2f1b6248 | 118 | "Normal", |
e53ef38d | 119 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 120 | "HighMem", |
e53ef38d | 121 | #endif |
2a1e274a | 122 | "Movable", |
2f1b6248 CL |
123 | }; |
124 | ||
1da177e4 LT |
125 | int min_free_kbytes = 1024; |
126 | ||
2c85f51d JB |
127 | static unsigned long __meminitdata nr_kernel_pages; |
128 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 129 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 130 | |
c713216d MG |
131 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
132 | /* | |
183ff22b | 133 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
134 | * ranges of memory (RAM) that may be registered with add_active_range(). |
135 | * Ranges passed to add_active_range() will be merged if possible | |
136 | * so the number of times add_active_range() can be called is | |
137 | * related to the number of nodes and the number of holes | |
138 | */ | |
139 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
140 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
141 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
142 | #else | |
143 | #if MAX_NUMNODES >= 32 | |
144 | /* If there can be many nodes, allow up to 50 holes per node */ | |
145 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
146 | #else | |
147 | /* By default, allow up to 256 distinct regions */ | |
148 | #define MAX_ACTIVE_REGIONS 256 | |
149 | #endif | |
150 | #endif | |
151 | ||
98011f56 JB |
152 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
153 | static int __meminitdata nr_nodemap_entries; | |
154 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
155 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
b69a7288 | 156 | static unsigned long __initdata required_kernelcore; |
484f51f8 | 157 | static unsigned long __initdata required_movablecore; |
b69a7288 | 158 | static 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; | |
62bc62a8 | 167 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 168 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 169 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
170 | #endif |
171 | ||
9ef9acb0 MG |
172 | int page_group_by_mobility_disabled __read_mostly; |
173 | ||
b2a0ac88 MG |
174 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
175 | { | |
49255c61 MG |
176 | |
177 | if (unlikely(page_group_by_mobility_disabled)) | |
178 | migratetype = MIGRATE_UNMOVABLE; | |
179 | ||
b2a0ac88 MG |
180 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
181 | PB_migrate, PB_migrate_end); | |
182 | } | |
183 | ||
7f33d49a RW |
184 | bool oom_killer_disabled __read_mostly; |
185 | ||
13e7444b | 186 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 187 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 188 | { |
bdc8cb98 DH |
189 | int ret = 0; |
190 | unsigned seq; | |
191 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 192 | |
bdc8cb98 DH |
193 | do { |
194 | seq = zone_span_seqbegin(zone); | |
195 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
196 | ret = 1; | |
197 | else if (pfn < zone->zone_start_pfn) | |
198 | ret = 1; | |
199 | } while (zone_span_seqretry(zone, seq)); | |
200 | ||
201 | return ret; | |
c6a57e19 DH |
202 | } |
203 | ||
204 | static int page_is_consistent(struct zone *zone, struct page *page) | |
205 | { | |
14e07298 | 206 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 207 | return 0; |
1da177e4 | 208 | if (zone != page_zone(page)) |
c6a57e19 DH |
209 | return 0; |
210 | ||
211 | return 1; | |
212 | } | |
213 | /* | |
214 | * Temporary debugging check for pages not lying within a given zone. | |
215 | */ | |
216 | static int bad_range(struct zone *zone, struct page *page) | |
217 | { | |
218 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 219 | return 1; |
c6a57e19 DH |
220 | if (!page_is_consistent(zone, page)) |
221 | return 1; | |
222 | ||
1da177e4 LT |
223 | return 0; |
224 | } | |
13e7444b NP |
225 | #else |
226 | static inline int bad_range(struct zone *zone, struct page *page) | |
227 | { | |
228 | return 0; | |
229 | } | |
230 | #endif | |
231 | ||
224abf92 | 232 | static void bad_page(struct page *page) |
1da177e4 | 233 | { |
d936cf9b HD |
234 | static unsigned long resume; |
235 | static unsigned long nr_shown; | |
236 | static unsigned long nr_unshown; | |
237 | ||
2a7684a2 WF |
238 | /* Don't complain about poisoned pages */ |
239 | if (PageHWPoison(page)) { | |
240 | __ClearPageBuddy(page); | |
241 | return; | |
242 | } | |
243 | ||
d936cf9b HD |
244 | /* |
245 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
246 | * or allow a steady drip of one report per second. | |
247 | */ | |
248 | if (nr_shown == 60) { | |
249 | if (time_before(jiffies, resume)) { | |
250 | nr_unshown++; | |
251 | goto out; | |
252 | } | |
253 | if (nr_unshown) { | |
1e9e6365 HD |
254 | printk(KERN_ALERT |
255 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
256 | nr_unshown); |
257 | nr_unshown = 0; | |
258 | } | |
259 | nr_shown = 0; | |
260 | } | |
261 | if (nr_shown++ == 0) | |
262 | resume = jiffies + 60 * HZ; | |
263 | ||
1e9e6365 | 264 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 265 | current->comm, page_to_pfn(page)); |
1e9e6365 | 266 | printk(KERN_ALERT |
3dc14741 HD |
267 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
268 | page, (void *)page->flags, page_count(page), | |
269 | page_mapcount(page), page->mapping, page->index); | |
3dc14741 | 270 | |
1da177e4 | 271 | dump_stack(); |
d936cf9b | 272 | out: |
8cc3b392 HD |
273 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
274 | __ClearPageBuddy(page); | |
9f158333 | 275 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
276 | } |
277 | ||
1da177e4 LT |
278 | /* |
279 | * Higher-order pages are called "compound pages". They are structured thusly: | |
280 | * | |
281 | * The first PAGE_SIZE page is called the "head page". | |
282 | * | |
283 | * The remaining PAGE_SIZE pages are called "tail pages". | |
284 | * | |
285 | * All pages have PG_compound set. All pages have their ->private pointing at | |
286 | * the head page (even the head page has this). | |
287 | * | |
41d78ba5 HD |
288 | * The first tail page's ->lru.next holds the address of the compound page's |
289 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
290 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 291 | */ |
d98c7a09 HD |
292 | |
293 | static void free_compound_page(struct page *page) | |
294 | { | |
d85f3385 | 295 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
296 | } |
297 | ||
01ad1c08 | 298 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
299 | { |
300 | int i; | |
301 | int nr_pages = 1 << order; | |
302 | ||
303 | set_compound_page_dtor(page, free_compound_page); | |
304 | set_compound_order(page, order); | |
305 | __SetPageHead(page); | |
306 | for (i = 1; i < nr_pages; i++) { | |
307 | struct page *p = page + i; | |
308 | ||
309 | __SetPageTail(p); | |
310 | p->first_page = page; | |
311 | } | |
312 | } | |
313 | ||
8cc3b392 | 314 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
315 | { |
316 | int i; | |
317 | int nr_pages = 1 << order; | |
8cc3b392 | 318 | int bad = 0; |
1da177e4 | 319 | |
8cc3b392 HD |
320 | if (unlikely(compound_order(page) != order) || |
321 | unlikely(!PageHead(page))) { | |
224abf92 | 322 | bad_page(page); |
8cc3b392 HD |
323 | bad++; |
324 | } | |
1da177e4 | 325 | |
6d777953 | 326 | __ClearPageHead(page); |
8cc3b392 | 327 | |
18229df5 AW |
328 | for (i = 1; i < nr_pages; i++) { |
329 | struct page *p = page + i; | |
1da177e4 | 330 | |
e713a21d | 331 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 332 | bad_page(page); |
8cc3b392 HD |
333 | bad++; |
334 | } | |
d85f3385 | 335 | __ClearPageTail(p); |
1da177e4 | 336 | } |
8cc3b392 HD |
337 | |
338 | return bad; | |
1da177e4 | 339 | } |
1da177e4 | 340 | |
17cf4406 NP |
341 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
342 | { | |
343 | int i; | |
344 | ||
6626c5d5 AM |
345 | /* |
346 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
347 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
348 | */ | |
725d704e | 349 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
350 | for (i = 0; i < (1 << order); i++) |
351 | clear_highpage(page + i); | |
352 | } | |
353 | ||
6aa3001b AM |
354 | static inline void set_page_order(struct page *page, int order) |
355 | { | |
4c21e2f2 | 356 | set_page_private(page, order); |
676165a8 | 357 | __SetPageBuddy(page); |
1da177e4 LT |
358 | } |
359 | ||
360 | static inline void rmv_page_order(struct page *page) | |
361 | { | |
676165a8 | 362 | __ClearPageBuddy(page); |
4c21e2f2 | 363 | set_page_private(page, 0); |
1da177e4 LT |
364 | } |
365 | ||
366 | /* | |
367 | * Locate the struct page for both the matching buddy in our | |
368 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
369 | * | |
370 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
371 | * the following equation: | |
372 | * B2 = B1 ^ (1 << O) | |
373 | * For example, if the starting buddy (buddy2) is #8 its order | |
374 | * 1 buddy is #10: | |
375 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
376 | * | |
377 | * 2) Any buddy B will have an order O+1 parent P which | |
378 | * satisfies the following equation: | |
379 | * P = B & ~(1 << O) | |
380 | * | |
d6e05edc | 381 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
382 | */ |
383 | static inline struct page * | |
384 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
385 | { | |
386 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
387 | ||
388 | return page + (buddy_idx - page_idx); | |
389 | } | |
390 | ||
391 | static inline unsigned long | |
392 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
393 | { | |
394 | return (page_idx & ~(1 << order)); | |
395 | } | |
396 | ||
397 | /* | |
398 | * This function checks whether a page is free && is the buddy | |
399 | * we can do coalesce a page and its buddy if | |
13e7444b | 400 | * (a) the buddy is not in a hole && |
676165a8 | 401 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
402 | * (c) a page and its buddy have the same order && |
403 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
404 | * |
405 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
406 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 407 | * |
676165a8 | 408 | * For recording page's order, we use page_private(page). |
1da177e4 | 409 | */ |
cb2b95e1 AW |
410 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
411 | int order) | |
1da177e4 | 412 | { |
14e07298 | 413 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 414 | return 0; |
13e7444b | 415 | |
cb2b95e1 AW |
416 | if (page_zone_id(page) != page_zone_id(buddy)) |
417 | return 0; | |
418 | ||
419 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
a3af9c38 | 420 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 421 | return 1; |
676165a8 | 422 | } |
6aa3001b | 423 | return 0; |
1da177e4 LT |
424 | } |
425 | ||
426 | /* | |
427 | * Freeing function for a buddy system allocator. | |
428 | * | |
429 | * The concept of a buddy system is to maintain direct-mapped table | |
430 | * (containing bit values) for memory blocks of various "orders". | |
431 | * The bottom level table contains the map for the smallest allocatable | |
432 | * units of memory (here, pages), and each level above it describes | |
433 | * pairs of units from the levels below, hence, "buddies". | |
434 | * At a high level, all that happens here is marking the table entry | |
435 | * at the bottom level available, and propagating the changes upward | |
436 | * as necessary, plus some accounting needed to play nicely with other | |
437 | * parts of the VM system. | |
438 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 439 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 440 | * order is recorded in page_private(page) field. |
1da177e4 LT |
441 | * So when we are allocating or freeing one, we can derive the state of the |
442 | * other. That is, if we allocate a small block, and both were | |
443 | * free, the remainder of the region must be split into blocks. | |
444 | * If a block is freed, and its buddy is also free, then this | |
445 | * triggers coalescing into a block of larger size. | |
446 | * | |
447 | * -- wli | |
448 | */ | |
449 | ||
48db57f8 | 450 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
451 | struct zone *zone, unsigned int order, |
452 | int migratetype) | |
1da177e4 LT |
453 | { |
454 | unsigned long page_idx; | |
1da177e4 | 455 | |
224abf92 | 456 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
457 | if (unlikely(destroy_compound_page(page, order))) |
458 | return; | |
1da177e4 | 459 | |
ed0ae21d MG |
460 | VM_BUG_ON(migratetype == -1); |
461 | ||
1da177e4 LT |
462 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
463 | ||
f2260e6b | 464 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 465 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 466 | |
1da177e4 LT |
467 | while (order < MAX_ORDER-1) { |
468 | unsigned long combined_idx; | |
1da177e4 LT |
469 | struct page *buddy; |
470 | ||
1da177e4 | 471 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 472 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 473 | break; |
13e7444b | 474 | |
3c82d0ce | 475 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 476 | list_del(&buddy->lru); |
b2a0ac88 | 477 | zone->free_area[order].nr_free--; |
1da177e4 | 478 | rmv_page_order(buddy); |
13e7444b | 479 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
480 | page = page + (combined_idx - page_idx); |
481 | page_idx = combined_idx; | |
482 | order++; | |
483 | } | |
484 | set_page_order(page, order); | |
b2a0ac88 MG |
485 | list_add(&page->lru, |
486 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
487 | zone->free_area[order].nr_free++; |
488 | } | |
489 | ||
092cead6 KM |
490 | /* |
491 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
492 | * Page should not be on lru, so no need to fix that up. | |
493 | * free_pages_check() will verify... | |
494 | */ | |
495 | static inline void free_page_mlock(struct page *page) | |
496 | { | |
092cead6 KM |
497 | __dec_zone_page_state(page, NR_MLOCK); |
498 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
499 | } | |
092cead6 | 500 | |
224abf92 | 501 | static inline int free_pages_check(struct page *page) |
1da177e4 | 502 | { |
92be2e33 NP |
503 | if (unlikely(page_mapcount(page) | |
504 | (page->mapping != NULL) | | |
a3af9c38 | 505 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 506 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
224abf92 | 507 | bad_page(page); |
79f4b7bf | 508 | return 1; |
8cc3b392 | 509 | } |
79f4b7bf HD |
510 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
511 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
512 | return 0; | |
1da177e4 LT |
513 | } |
514 | ||
515 | /* | |
5f8dcc21 | 516 | * Frees a number of pages from the PCP lists |
1da177e4 | 517 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 518 | * count is the number of pages to free. |
1da177e4 LT |
519 | * |
520 | * If the zone was previously in an "all pages pinned" state then look to | |
521 | * see if this freeing clears that state. | |
522 | * | |
523 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
524 | * pinned" detection logic. | |
525 | */ | |
5f8dcc21 MG |
526 | static void free_pcppages_bulk(struct zone *zone, int count, |
527 | struct per_cpu_pages *pcp) | |
1da177e4 | 528 | { |
5f8dcc21 | 529 | int migratetype = 0; |
a6f9edd6 | 530 | int batch_free = 0; |
5f8dcc21 | 531 | |
c54ad30c | 532 | spin_lock(&zone->lock); |
e815af95 | 533 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 534 | zone->pages_scanned = 0; |
f2260e6b | 535 | |
5f8dcc21 | 536 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
a6f9edd6 | 537 | while (count) { |
48db57f8 | 538 | struct page *page; |
5f8dcc21 MG |
539 | struct list_head *list; |
540 | ||
541 | /* | |
a6f9edd6 MG |
542 | * Remove pages from lists in a round-robin fashion. A |
543 | * batch_free count is maintained that is incremented when an | |
544 | * empty list is encountered. This is so more pages are freed | |
545 | * off fuller lists instead of spinning excessively around empty | |
546 | * lists | |
5f8dcc21 MG |
547 | */ |
548 | do { | |
a6f9edd6 | 549 | batch_free++; |
5f8dcc21 MG |
550 | if (++migratetype == MIGRATE_PCPTYPES) |
551 | migratetype = 0; | |
552 | list = &pcp->lists[migratetype]; | |
553 | } while (list_empty(list)); | |
48db57f8 | 554 | |
a6f9edd6 MG |
555 | do { |
556 | page = list_entry(list->prev, struct page, lru); | |
557 | /* must delete as __free_one_page list manipulates */ | |
558 | list_del(&page->lru); | |
a7016235 HD |
559 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
560 | __free_one_page(page, zone, 0, page_private(page)); | |
561 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); | |
a6f9edd6 | 562 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 563 | } |
c54ad30c | 564 | spin_unlock(&zone->lock); |
1da177e4 LT |
565 | } |
566 | ||
ed0ae21d MG |
567 | static void free_one_page(struct zone *zone, struct page *page, int order, |
568 | int migratetype) | |
1da177e4 | 569 | { |
006d22d9 | 570 | spin_lock(&zone->lock); |
e815af95 | 571 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 572 | zone->pages_scanned = 0; |
f2260e6b MG |
573 | |
574 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); | |
ed0ae21d | 575 | __free_one_page(page, zone, order, migratetype); |
006d22d9 | 576 | spin_unlock(&zone->lock); |
48db57f8 NP |
577 | } |
578 | ||
579 | static void __free_pages_ok(struct page *page, unsigned int order) | |
580 | { | |
581 | unsigned long flags; | |
1da177e4 | 582 | int i; |
8cc3b392 | 583 | int bad = 0; |
451ea25d | 584 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 585 | |
b1eeab67 VN |
586 | kmemcheck_free_shadow(page, order); |
587 | ||
1da177e4 | 588 | for (i = 0 ; i < (1 << order) ; ++i) |
8cc3b392 HD |
589 | bad += free_pages_check(page + i); |
590 | if (bad) | |
689bcebf HD |
591 | return; |
592 | ||
3ac7fe5a | 593 | if (!PageHighMem(page)) { |
9858db50 | 594 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
595 | debug_check_no_obj_freed(page_address(page), |
596 | PAGE_SIZE << order); | |
597 | } | |
dafb1367 | 598 | arch_free_page(page, order); |
48db57f8 | 599 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 600 | |
c54ad30c | 601 | local_irq_save(flags); |
c277331d | 602 | if (unlikely(wasMlocked)) |
da456f14 | 603 | free_page_mlock(page); |
f8891e5e | 604 | __count_vm_events(PGFREE, 1 << order); |
ed0ae21d MG |
605 | free_one_page(page_zone(page), page, order, |
606 | get_pageblock_migratetype(page)); | |
c54ad30c | 607 | local_irq_restore(flags); |
1da177e4 LT |
608 | } |
609 | ||
a226f6c8 DH |
610 | /* |
611 | * permit the bootmem allocator to evade page validation on high-order frees | |
612 | */ | |
af370fb8 | 613 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
614 | { |
615 | if (order == 0) { | |
616 | __ClearPageReserved(page); | |
617 | set_page_count(page, 0); | |
7835e98b | 618 | set_page_refcounted(page); |
545b1ea9 | 619 | __free_page(page); |
a226f6c8 | 620 | } else { |
a226f6c8 DH |
621 | int loop; |
622 | ||
545b1ea9 | 623 | prefetchw(page); |
a226f6c8 DH |
624 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
625 | struct page *p = &page[loop]; | |
626 | ||
545b1ea9 NP |
627 | if (loop + 1 < BITS_PER_LONG) |
628 | prefetchw(p + 1); | |
a226f6c8 DH |
629 | __ClearPageReserved(p); |
630 | set_page_count(p, 0); | |
631 | } | |
632 | ||
7835e98b | 633 | set_page_refcounted(page); |
545b1ea9 | 634 | __free_pages(page, order); |
a226f6c8 DH |
635 | } |
636 | } | |
637 | ||
1da177e4 LT |
638 | |
639 | /* | |
640 | * The order of subdivision here is critical for the IO subsystem. | |
641 | * Please do not alter this order without good reasons and regression | |
642 | * testing. Specifically, as large blocks of memory are subdivided, | |
643 | * the order in which smaller blocks are delivered depends on the order | |
644 | * they're subdivided in this function. This is the primary factor | |
645 | * influencing the order in which pages are delivered to the IO | |
646 | * subsystem according to empirical testing, and this is also justified | |
647 | * by considering the behavior of a buddy system containing a single | |
648 | * large block of memory acted on by a series of small allocations. | |
649 | * This behavior is a critical factor in sglist merging's success. | |
650 | * | |
651 | * -- wli | |
652 | */ | |
085cc7d5 | 653 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
654 | int low, int high, struct free_area *area, |
655 | int migratetype) | |
1da177e4 LT |
656 | { |
657 | unsigned long size = 1 << high; | |
658 | ||
659 | while (high > low) { | |
660 | area--; | |
661 | high--; | |
662 | size >>= 1; | |
725d704e | 663 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 664 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
665 | area->nr_free++; |
666 | set_page_order(&page[size], high); | |
667 | } | |
1da177e4 LT |
668 | } |
669 | ||
1da177e4 LT |
670 | /* |
671 | * This page is about to be returned from the page allocator | |
672 | */ | |
2a7684a2 | 673 | static inline int check_new_page(struct page *page) |
1da177e4 | 674 | { |
92be2e33 NP |
675 | if (unlikely(page_mapcount(page) | |
676 | (page->mapping != NULL) | | |
a3af9c38 | 677 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 678 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
224abf92 | 679 | bad_page(page); |
689bcebf | 680 | return 1; |
8cc3b392 | 681 | } |
2a7684a2 WF |
682 | return 0; |
683 | } | |
684 | ||
685 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
686 | { | |
687 | int i; | |
688 | ||
689 | for (i = 0; i < (1 << order); i++) { | |
690 | struct page *p = page + i; | |
691 | if (unlikely(check_new_page(p))) | |
692 | return 1; | |
693 | } | |
689bcebf | 694 | |
4c21e2f2 | 695 | set_page_private(page, 0); |
7835e98b | 696 | set_page_refcounted(page); |
cc102509 NP |
697 | |
698 | arch_alloc_page(page, order); | |
1da177e4 | 699 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
700 | |
701 | if (gfp_flags & __GFP_ZERO) | |
702 | prep_zero_page(page, order, gfp_flags); | |
703 | ||
704 | if (order && (gfp_flags & __GFP_COMP)) | |
705 | prep_compound_page(page, order); | |
706 | ||
689bcebf | 707 | return 0; |
1da177e4 LT |
708 | } |
709 | ||
56fd56b8 MG |
710 | /* |
711 | * Go through the free lists for the given migratetype and remove | |
712 | * the smallest available page from the freelists | |
713 | */ | |
728ec980 MG |
714 | static inline |
715 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
716 | int migratetype) |
717 | { | |
718 | unsigned int current_order; | |
719 | struct free_area * area; | |
720 | struct page *page; | |
721 | ||
722 | /* Find a page of the appropriate size in the preferred list */ | |
723 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
724 | area = &(zone->free_area[current_order]); | |
725 | if (list_empty(&area->free_list[migratetype])) | |
726 | continue; | |
727 | ||
728 | page = list_entry(area->free_list[migratetype].next, | |
729 | struct page, lru); | |
730 | list_del(&page->lru); | |
731 | rmv_page_order(page); | |
732 | area->nr_free--; | |
56fd56b8 MG |
733 | expand(zone, page, order, current_order, area, migratetype); |
734 | return page; | |
735 | } | |
736 | ||
737 | return NULL; | |
738 | } | |
739 | ||
740 | ||
b2a0ac88 MG |
741 | /* |
742 | * This array describes the order lists are fallen back to when | |
743 | * the free lists for the desirable migrate type are depleted | |
744 | */ | |
745 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
746 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
747 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
748 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
749 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
750 | }; |
751 | ||
c361be55 MG |
752 | /* |
753 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 754 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
755 | * boundary. If alignment is required, use move_freepages_block() |
756 | */ | |
b69a7288 AB |
757 | static int move_freepages(struct zone *zone, |
758 | struct page *start_page, struct page *end_page, | |
759 | int migratetype) | |
c361be55 MG |
760 | { |
761 | struct page *page; | |
762 | unsigned long order; | |
d100313f | 763 | int pages_moved = 0; |
c361be55 MG |
764 | |
765 | #ifndef CONFIG_HOLES_IN_ZONE | |
766 | /* | |
767 | * page_zone is not safe to call in this context when | |
768 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
769 | * anyway as we check zone boundaries in move_freepages_block(). | |
770 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 771 | * grouping pages by mobility |
c361be55 MG |
772 | */ |
773 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
774 | #endif | |
775 | ||
776 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
777 | /* Make sure we are not inadvertently changing nodes */ |
778 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
779 | ||
c361be55 MG |
780 | if (!pfn_valid_within(page_to_pfn(page))) { |
781 | page++; | |
782 | continue; | |
783 | } | |
784 | ||
785 | if (!PageBuddy(page)) { | |
786 | page++; | |
787 | continue; | |
788 | } | |
789 | ||
790 | order = page_order(page); | |
791 | list_del(&page->lru); | |
792 | list_add(&page->lru, | |
793 | &zone->free_area[order].free_list[migratetype]); | |
794 | page += 1 << order; | |
d100313f | 795 | pages_moved += 1 << order; |
c361be55 MG |
796 | } |
797 | ||
d100313f | 798 | return pages_moved; |
c361be55 MG |
799 | } |
800 | ||
b69a7288 AB |
801 | static int move_freepages_block(struct zone *zone, struct page *page, |
802 | int migratetype) | |
c361be55 MG |
803 | { |
804 | unsigned long start_pfn, end_pfn; | |
805 | struct page *start_page, *end_page; | |
806 | ||
807 | start_pfn = page_to_pfn(page); | |
d9c23400 | 808 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 809 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
810 | end_page = start_page + pageblock_nr_pages - 1; |
811 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
812 | |
813 | /* Do not cross zone boundaries */ | |
814 | if (start_pfn < zone->zone_start_pfn) | |
815 | start_page = page; | |
816 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
817 | return 0; | |
818 | ||
819 | return move_freepages(zone, start_page, end_page, migratetype); | |
820 | } | |
821 | ||
2f66a68f MG |
822 | static void change_pageblock_range(struct page *pageblock_page, |
823 | int start_order, int migratetype) | |
824 | { | |
825 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
826 | ||
827 | while (nr_pageblocks--) { | |
828 | set_pageblock_migratetype(pageblock_page, migratetype); | |
829 | pageblock_page += pageblock_nr_pages; | |
830 | } | |
831 | } | |
832 | ||
b2a0ac88 | 833 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
834 | static inline struct page * |
835 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
836 | { |
837 | struct free_area * area; | |
838 | int current_order; | |
839 | struct page *page; | |
840 | int migratetype, i; | |
841 | ||
842 | /* Find the largest possible block of pages in the other list */ | |
843 | for (current_order = MAX_ORDER-1; current_order >= order; | |
844 | --current_order) { | |
845 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
846 | migratetype = fallbacks[start_migratetype][i]; | |
847 | ||
56fd56b8 MG |
848 | /* MIGRATE_RESERVE handled later if necessary */ |
849 | if (migratetype == MIGRATE_RESERVE) | |
850 | continue; | |
e010487d | 851 | |
b2a0ac88 MG |
852 | area = &(zone->free_area[current_order]); |
853 | if (list_empty(&area->free_list[migratetype])) | |
854 | continue; | |
855 | ||
856 | page = list_entry(area->free_list[migratetype].next, | |
857 | struct page, lru); | |
858 | area->nr_free--; | |
859 | ||
860 | /* | |
c361be55 | 861 | * If breaking a large block of pages, move all free |
46dafbca MG |
862 | * pages to the preferred allocation list. If falling |
863 | * back for a reclaimable kernel allocation, be more | |
864 | * agressive about taking ownership of free pages | |
b2a0ac88 | 865 | */ |
d9c23400 | 866 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
dd5d241e MG |
867 | start_migratetype == MIGRATE_RECLAIMABLE || |
868 | page_group_by_mobility_disabled) { | |
46dafbca MG |
869 | unsigned long pages; |
870 | pages = move_freepages_block(zone, page, | |
871 | start_migratetype); | |
872 | ||
873 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
874 | if (pages >= (1 << (pageblock_order-1)) || |
875 | page_group_by_mobility_disabled) | |
46dafbca MG |
876 | set_pageblock_migratetype(page, |
877 | start_migratetype); | |
878 | ||
b2a0ac88 | 879 | migratetype = start_migratetype; |
c361be55 | 880 | } |
b2a0ac88 MG |
881 | |
882 | /* Remove the page from the freelists */ | |
883 | list_del(&page->lru); | |
884 | rmv_page_order(page); | |
b2a0ac88 | 885 | |
2f66a68f MG |
886 | /* Take ownership for orders >= pageblock_order */ |
887 | if (current_order >= pageblock_order) | |
888 | change_pageblock_range(page, current_order, | |
b2a0ac88 MG |
889 | start_migratetype); |
890 | ||
891 | expand(zone, page, order, current_order, area, migratetype); | |
e0fff1bd MG |
892 | |
893 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
894 | start_migratetype, migratetype); | |
895 | ||
b2a0ac88 MG |
896 | return page; |
897 | } | |
898 | } | |
899 | ||
728ec980 | 900 | return NULL; |
b2a0ac88 MG |
901 | } |
902 | ||
56fd56b8 | 903 | /* |
1da177e4 LT |
904 | * Do the hard work of removing an element from the buddy allocator. |
905 | * Call me with the zone->lock already held. | |
906 | */ | |
b2a0ac88 MG |
907 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
908 | int migratetype) | |
1da177e4 | 909 | { |
1da177e4 LT |
910 | struct page *page; |
911 | ||
728ec980 | 912 | retry_reserve: |
56fd56b8 | 913 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 914 | |
728ec980 | 915 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 916 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 917 | |
728ec980 MG |
918 | /* |
919 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
920 | * is used because __rmqueue_smallest is an inline function | |
921 | * and we want just one call site | |
922 | */ | |
923 | if (!page) { | |
924 | migratetype = MIGRATE_RESERVE; | |
925 | goto retry_reserve; | |
926 | } | |
927 | } | |
928 | ||
0d3d062a | 929 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 930 | return page; |
1da177e4 LT |
931 | } |
932 | ||
933 | /* | |
934 | * Obtain a specified number of elements from the buddy allocator, all under | |
935 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
936 | * Returns the number of new pages which were placed at *list. | |
937 | */ | |
938 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 | 939 | unsigned long count, struct list_head *list, |
e084b2d9 | 940 | int migratetype, int cold) |
1da177e4 | 941 | { |
1da177e4 | 942 | int i; |
1da177e4 | 943 | |
c54ad30c | 944 | spin_lock(&zone->lock); |
1da177e4 | 945 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 946 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 947 | if (unlikely(page == NULL)) |
1da177e4 | 948 | break; |
81eabcbe MG |
949 | |
950 | /* | |
951 | * Split buddy pages returned by expand() are received here | |
952 | * in physical page order. The page is added to the callers and | |
953 | * list and the list head then moves forward. From the callers | |
954 | * perspective, the linked list is ordered by page number in | |
955 | * some conditions. This is useful for IO devices that can | |
956 | * merge IO requests if the physical pages are ordered | |
957 | * properly. | |
958 | */ | |
e084b2d9 MG |
959 | if (likely(cold == 0)) |
960 | list_add(&page->lru, list); | |
961 | else | |
962 | list_add_tail(&page->lru, list); | |
535131e6 | 963 | set_page_private(page, migratetype); |
81eabcbe | 964 | list = &page->lru; |
1da177e4 | 965 | } |
f2260e6b | 966 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 967 | spin_unlock(&zone->lock); |
085cc7d5 | 968 | return i; |
1da177e4 LT |
969 | } |
970 | ||
4ae7c039 | 971 | #ifdef CONFIG_NUMA |
8fce4d8e | 972 | /* |
4037d452 CL |
973 | * Called from the vmstat counter updater to drain pagesets of this |
974 | * currently executing processor on remote nodes after they have | |
975 | * expired. | |
976 | * | |
879336c3 CL |
977 | * Note that this function must be called with the thread pinned to |
978 | * a single processor. | |
8fce4d8e | 979 | */ |
4037d452 | 980 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 981 | { |
4ae7c039 | 982 | unsigned long flags; |
4037d452 | 983 | int to_drain; |
4ae7c039 | 984 | |
4037d452 CL |
985 | local_irq_save(flags); |
986 | if (pcp->count >= pcp->batch) | |
987 | to_drain = pcp->batch; | |
988 | else | |
989 | to_drain = pcp->count; | |
5f8dcc21 | 990 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 CL |
991 | pcp->count -= to_drain; |
992 | local_irq_restore(flags); | |
4ae7c039 CL |
993 | } |
994 | #endif | |
995 | ||
9f8f2172 CL |
996 | /* |
997 | * Drain pages of the indicated processor. | |
998 | * | |
999 | * The processor must either be the current processor and the | |
1000 | * thread pinned to the current processor or a processor that | |
1001 | * is not online. | |
1002 | */ | |
1003 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1004 | { |
c54ad30c | 1005 | unsigned long flags; |
1da177e4 | 1006 | struct zone *zone; |
1da177e4 | 1007 | |
ee99c71c | 1008 | for_each_populated_zone(zone) { |
1da177e4 | 1009 | struct per_cpu_pageset *pset; |
3dfa5721 | 1010 | struct per_cpu_pages *pcp; |
1da177e4 | 1011 | |
e7c8d5c9 | 1012 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
1013 | |
1014 | pcp = &pset->pcp; | |
1015 | local_irq_save(flags); | |
5f8dcc21 | 1016 | free_pcppages_bulk(zone, pcp->count, pcp); |
3dfa5721 CL |
1017 | pcp->count = 0; |
1018 | local_irq_restore(flags); | |
1da177e4 LT |
1019 | } |
1020 | } | |
1da177e4 | 1021 | |
9f8f2172 CL |
1022 | /* |
1023 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1024 | */ | |
1025 | void drain_local_pages(void *arg) | |
1026 | { | |
1027 | drain_pages(smp_processor_id()); | |
1028 | } | |
1029 | ||
1030 | /* | |
1031 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1032 | */ | |
1033 | void drain_all_pages(void) | |
1034 | { | |
15c8b6c1 | 1035 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
1036 | } |
1037 | ||
296699de | 1038 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1039 | |
1040 | void mark_free_pages(struct zone *zone) | |
1041 | { | |
f623f0db RW |
1042 | unsigned long pfn, max_zone_pfn; |
1043 | unsigned long flags; | |
b2a0ac88 | 1044 | int order, t; |
1da177e4 LT |
1045 | struct list_head *curr; |
1046 | ||
1047 | if (!zone->spanned_pages) | |
1048 | return; | |
1049 | ||
1050 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1051 | |
1052 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1053 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1054 | if (pfn_valid(pfn)) { | |
1055 | struct page *page = pfn_to_page(pfn); | |
1056 | ||
7be98234 RW |
1057 | if (!swsusp_page_is_forbidden(page)) |
1058 | swsusp_unset_page_free(page); | |
f623f0db | 1059 | } |
1da177e4 | 1060 | |
b2a0ac88 MG |
1061 | for_each_migratetype_order(order, t) { |
1062 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1063 | unsigned long i; |
1da177e4 | 1064 | |
f623f0db RW |
1065 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1066 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1067 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1068 | } |
b2a0ac88 | 1069 | } |
1da177e4 LT |
1070 | spin_unlock_irqrestore(&zone->lock, flags); |
1071 | } | |
e2c55dc8 | 1072 | #endif /* CONFIG_PM */ |
1da177e4 | 1073 | |
1da177e4 LT |
1074 | /* |
1075 | * Free a 0-order page | |
1076 | */ | |
920c7a5d | 1077 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1078 | { |
1079 | struct zone *zone = page_zone(page); | |
1080 | struct per_cpu_pages *pcp; | |
1081 | unsigned long flags; | |
5f8dcc21 | 1082 | int migratetype; |
451ea25d | 1083 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 1084 | |
b1eeab67 VN |
1085 | kmemcheck_free_shadow(page, 0); |
1086 | ||
1da177e4 LT |
1087 | if (PageAnon(page)) |
1088 | page->mapping = NULL; | |
224abf92 | 1089 | if (free_pages_check(page)) |
689bcebf HD |
1090 | return; |
1091 | ||
3ac7fe5a | 1092 | if (!PageHighMem(page)) { |
9858db50 | 1093 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); |
3ac7fe5a TG |
1094 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); |
1095 | } | |
dafb1367 | 1096 | arch_free_page(page, 0); |
689bcebf HD |
1097 | kernel_map_pages(page, 1, 0); |
1098 | ||
3dfa5721 | 1099 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
5f8dcc21 MG |
1100 | migratetype = get_pageblock_migratetype(page); |
1101 | set_page_private(page, migratetype); | |
1da177e4 | 1102 | local_irq_save(flags); |
c277331d | 1103 | if (unlikely(wasMlocked)) |
da456f14 | 1104 | free_page_mlock(page); |
f8891e5e | 1105 | __count_vm_event(PGFREE); |
da456f14 | 1106 | |
5f8dcc21 MG |
1107 | /* |
1108 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1109 | * Free ISOLATE pages back to the allocator because they are being | |
1110 | * offlined but treat RESERVE as movable pages so we can get those | |
1111 | * areas back if necessary. Otherwise, we may have to free | |
1112 | * excessively into the page allocator | |
1113 | */ | |
1114 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1115 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1116 | free_one_page(zone, page, 0, migratetype); | |
1117 | goto out; | |
1118 | } | |
1119 | migratetype = MIGRATE_MOVABLE; | |
1120 | } | |
1121 | ||
3dfa5721 | 1122 | if (cold) |
5f8dcc21 | 1123 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1124 | else |
5f8dcc21 | 1125 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1126 | pcp->count++; |
48db57f8 | 1127 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1128 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1129 | pcp->count -= pcp->batch; |
1130 | } | |
5f8dcc21 MG |
1131 | |
1132 | out: | |
1da177e4 LT |
1133 | local_irq_restore(flags); |
1134 | put_cpu(); | |
1135 | } | |
1136 | ||
920c7a5d | 1137 | void free_hot_page(struct page *page) |
1da177e4 | 1138 | { |
4b4f278c | 1139 | trace_mm_page_free_direct(page, 0); |
1da177e4 LT |
1140 | free_hot_cold_page(page, 0); |
1141 | } | |
1142 | ||
8dfcc9ba NP |
1143 | /* |
1144 | * split_page takes a non-compound higher-order page, and splits it into | |
1145 | * n (1<<order) sub-pages: page[0..n] | |
1146 | * Each sub-page must be freed individually. | |
1147 | * | |
1148 | * Note: this is probably too low level an operation for use in drivers. | |
1149 | * Please consult with lkml before using this in your driver. | |
1150 | */ | |
1151 | void split_page(struct page *page, unsigned int order) | |
1152 | { | |
1153 | int i; | |
1154 | ||
725d704e NP |
1155 | VM_BUG_ON(PageCompound(page)); |
1156 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1157 | |
1158 | #ifdef CONFIG_KMEMCHECK | |
1159 | /* | |
1160 | * Split shadow pages too, because free(page[0]) would | |
1161 | * otherwise free the whole shadow. | |
1162 | */ | |
1163 | if (kmemcheck_page_is_tracked(page)) | |
1164 | split_page(virt_to_page(page[0].shadow), order); | |
1165 | #endif | |
1166 | ||
7835e98b NP |
1167 | for (i = 1; i < (1 << order); i++) |
1168 | set_page_refcounted(page + i); | |
8dfcc9ba | 1169 | } |
8dfcc9ba | 1170 | |
1da177e4 LT |
1171 | /* |
1172 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1173 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1174 | * or two. | |
1175 | */ | |
0a15c3e9 MG |
1176 | static inline |
1177 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1178 | struct zone *zone, int order, gfp_t gfp_flags, |
1179 | int migratetype) | |
1da177e4 LT |
1180 | { |
1181 | unsigned long flags; | |
689bcebf | 1182 | struct page *page; |
1da177e4 | 1183 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1184 | int cpu; |
1da177e4 | 1185 | |
689bcebf | 1186 | again: |
a74609fa | 1187 | cpu = get_cpu(); |
48db57f8 | 1188 | if (likely(order == 0)) { |
1da177e4 | 1189 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1190 | struct list_head *list; |
1da177e4 | 1191 | |
3dfa5721 | 1192 | pcp = &zone_pcp(zone, cpu)->pcp; |
5f8dcc21 | 1193 | list = &pcp->lists[migratetype]; |
1da177e4 | 1194 | local_irq_save(flags); |
5f8dcc21 | 1195 | if (list_empty(list)) { |
535131e6 | 1196 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1197 | pcp->batch, list, |
e084b2d9 | 1198 | migratetype, cold); |
5f8dcc21 | 1199 | if (unlikely(list_empty(list))) |
6fb332fa | 1200 | goto failed; |
535131e6 | 1201 | } |
b92a6edd | 1202 | |
5f8dcc21 MG |
1203 | if (cold) |
1204 | page = list_entry(list->prev, struct page, lru); | |
1205 | else | |
1206 | page = list_entry(list->next, struct page, lru); | |
1207 | ||
b92a6edd MG |
1208 | list_del(&page->lru); |
1209 | pcp->count--; | |
7fb1d9fc | 1210 | } else { |
dab48dab AM |
1211 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1212 | /* | |
1213 | * __GFP_NOFAIL is not to be used in new code. | |
1214 | * | |
1215 | * All __GFP_NOFAIL callers should be fixed so that they | |
1216 | * properly detect and handle allocation failures. | |
1217 | * | |
1218 | * We most definitely don't want callers attempting to | |
4923abf9 | 1219 | * allocate greater than order-1 page units with |
dab48dab AM |
1220 | * __GFP_NOFAIL. |
1221 | */ | |
4923abf9 | 1222 | WARN_ON_ONCE(order > 1); |
dab48dab | 1223 | } |
1da177e4 | 1224 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1225 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1226 | spin_unlock(&zone->lock); |
1227 | if (!page) | |
1228 | goto failed; | |
6ccf80eb | 1229 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
1da177e4 LT |
1230 | } |
1231 | ||
f8891e5e | 1232 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1233 | zone_statistics(preferred_zone, zone); |
a74609fa NP |
1234 | local_irq_restore(flags); |
1235 | put_cpu(); | |
1da177e4 | 1236 | |
725d704e | 1237 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1238 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1239 | goto again; |
1da177e4 | 1240 | return page; |
a74609fa NP |
1241 | |
1242 | failed: | |
1243 | local_irq_restore(flags); | |
1244 | put_cpu(); | |
1245 | return NULL; | |
1da177e4 LT |
1246 | } |
1247 | ||
41858966 MG |
1248 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
1249 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1250 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1251 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1252 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1253 | ||
1254 | /* Mask to get the watermark bits */ | |
1255 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1256 | ||
3148890b NP |
1257 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
1258 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1259 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1260 | |
933e312e AM |
1261 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1262 | ||
1263 | static struct fail_page_alloc_attr { | |
1264 | struct fault_attr attr; | |
1265 | ||
1266 | u32 ignore_gfp_highmem; | |
1267 | u32 ignore_gfp_wait; | |
54114994 | 1268 | u32 min_order; |
933e312e AM |
1269 | |
1270 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1271 | ||
1272 | struct dentry *ignore_gfp_highmem_file; | |
1273 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1274 | struct dentry *min_order_file; |
933e312e AM |
1275 | |
1276 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1277 | ||
1278 | } fail_page_alloc = { | |
1279 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1280 | .ignore_gfp_wait = 1, |
1281 | .ignore_gfp_highmem = 1, | |
54114994 | 1282 | .min_order = 1, |
933e312e AM |
1283 | }; |
1284 | ||
1285 | static int __init setup_fail_page_alloc(char *str) | |
1286 | { | |
1287 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1288 | } | |
1289 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1290 | ||
1291 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1292 | { | |
54114994 AM |
1293 | if (order < fail_page_alloc.min_order) |
1294 | return 0; | |
933e312e AM |
1295 | if (gfp_mask & __GFP_NOFAIL) |
1296 | return 0; | |
1297 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1298 | return 0; | |
1299 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1300 | return 0; | |
1301 | ||
1302 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1303 | } | |
1304 | ||
1305 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1306 | ||
1307 | static int __init fail_page_alloc_debugfs(void) | |
1308 | { | |
1309 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1310 | struct dentry *dir; | |
1311 | int err; | |
1312 | ||
1313 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1314 | "fail_page_alloc"); | |
1315 | if (err) | |
1316 | return err; | |
1317 | dir = fail_page_alloc.attr.dentries.dir; | |
1318 | ||
1319 | fail_page_alloc.ignore_gfp_wait_file = | |
1320 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1321 | &fail_page_alloc.ignore_gfp_wait); | |
1322 | ||
1323 | fail_page_alloc.ignore_gfp_highmem_file = | |
1324 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1325 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1326 | fail_page_alloc.min_order_file = |
1327 | debugfs_create_u32("min-order", mode, dir, | |
1328 | &fail_page_alloc.min_order); | |
933e312e AM |
1329 | |
1330 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1331 | !fail_page_alloc.ignore_gfp_highmem_file || |
1332 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1333 | err = -ENOMEM; |
1334 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1335 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1336 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1337 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1338 | } | |
1339 | ||
1340 | return err; | |
1341 | } | |
1342 | ||
1343 | late_initcall(fail_page_alloc_debugfs); | |
1344 | ||
1345 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1346 | ||
1347 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1348 | ||
1349 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1350 | { | |
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1355 | ||
1da177e4 LT |
1356 | /* |
1357 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1358 | * of the allocation. | |
1359 | */ | |
1360 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1361 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1362 | { |
1363 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1364 | long min = mark; |
1365 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1366 | int o; |
1367 | ||
7fb1d9fc | 1368 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1369 | min -= min / 2; |
7fb1d9fc | 1370 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1371 | min -= min / 4; |
1372 | ||
1373 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1374 | return 0; | |
1375 | for (o = 0; o < order; o++) { | |
1376 | /* At the next order, this order's pages become unavailable */ | |
1377 | free_pages -= z->free_area[o].nr_free << o; | |
1378 | ||
1379 | /* Require fewer higher order pages to be free */ | |
1380 | min >>= 1; | |
1381 | ||
1382 | if (free_pages <= min) | |
1383 | return 0; | |
1384 | } | |
1385 | return 1; | |
1386 | } | |
1387 | ||
9276b1bc PJ |
1388 | #ifdef CONFIG_NUMA |
1389 | /* | |
1390 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1391 | * skip over zones that are not allowed by the cpuset, or that have | |
1392 | * been recently (in last second) found to be nearly full. See further | |
1393 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1394 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1395 | * |
1396 | * If the zonelist cache is present in the passed in zonelist, then | |
1397 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1398 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1399 | * |
1400 | * If the zonelist cache is not available for this zonelist, does | |
1401 | * nothing and returns NULL. | |
1402 | * | |
1403 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1404 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1405 | * | |
1406 | * We hold off even calling zlc_setup, until after we've checked the | |
1407 | * first zone in the zonelist, on the theory that most allocations will | |
1408 | * be satisfied from that first zone, so best to examine that zone as | |
1409 | * quickly as we can. | |
1410 | */ | |
1411 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1412 | { | |
1413 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1414 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1415 | ||
1416 | zlc = zonelist->zlcache_ptr; | |
1417 | if (!zlc) | |
1418 | return NULL; | |
1419 | ||
f05111f5 | 1420 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1421 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1422 | zlc->last_full_zap = jiffies; | |
1423 | } | |
1424 | ||
1425 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1426 | &cpuset_current_mems_allowed : | |
37b07e41 | 1427 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1428 | return allowednodes; |
1429 | } | |
1430 | ||
1431 | /* | |
1432 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1433 | * if it is worth looking at further for free memory: | |
1434 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1435 | * bit set in the zonelist_cache fullzones BITMAP). | |
1436 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1437 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1438 | * Return true (non-zero) if zone is worth looking at further, or | |
1439 | * else return false (zero) if it is not. | |
1440 | * | |
1441 | * This check -ignores- the distinction between various watermarks, | |
1442 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1443 | * found to be full for any variation of these watermarks, it will | |
1444 | * be considered full for up to one second by all requests, unless | |
1445 | * we are so low on memory on all allowed nodes that we are forced | |
1446 | * into the second scan of the zonelist. | |
1447 | * | |
1448 | * In the second scan we ignore this zonelist cache and exactly | |
1449 | * apply the watermarks to all zones, even it is slower to do so. | |
1450 | * We are low on memory in the second scan, and should leave no stone | |
1451 | * unturned looking for a free page. | |
1452 | */ | |
dd1a239f | 1453 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1454 | nodemask_t *allowednodes) |
1455 | { | |
1456 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1457 | int i; /* index of *z in zonelist zones */ | |
1458 | int n; /* node that zone *z is on */ | |
1459 | ||
1460 | zlc = zonelist->zlcache_ptr; | |
1461 | if (!zlc) | |
1462 | return 1; | |
1463 | ||
dd1a239f | 1464 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1465 | n = zlc->z_to_n[i]; |
1466 | ||
1467 | /* This zone is worth trying if it is allowed but not full */ | |
1468 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1469 | } | |
1470 | ||
1471 | /* | |
1472 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1473 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1474 | * from that zone don't waste time re-examining it. | |
1475 | */ | |
dd1a239f | 1476 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1477 | { |
1478 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1479 | int i; /* index of *z in zonelist zones */ | |
1480 | ||
1481 | zlc = zonelist->zlcache_ptr; | |
1482 | if (!zlc) | |
1483 | return; | |
1484 | ||
dd1a239f | 1485 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1486 | |
1487 | set_bit(i, zlc->fullzones); | |
1488 | } | |
1489 | ||
1490 | #else /* CONFIG_NUMA */ | |
1491 | ||
1492 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1493 | { | |
1494 | return NULL; | |
1495 | } | |
1496 | ||
dd1a239f | 1497 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1498 | nodemask_t *allowednodes) |
1499 | { | |
1500 | return 1; | |
1501 | } | |
1502 | ||
dd1a239f | 1503 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1504 | { |
1505 | } | |
1506 | #endif /* CONFIG_NUMA */ | |
1507 | ||
7fb1d9fc | 1508 | /* |
0798e519 | 1509 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1510 | * a page. |
1511 | */ | |
1512 | static struct page * | |
19770b32 | 1513 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1514 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1515 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1516 | { |
dd1a239f | 1517 | struct zoneref *z; |
7fb1d9fc | 1518 | struct page *page = NULL; |
54a6eb5c | 1519 | int classzone_idx; |
5117f45d | 1520 | struct zone *zone; |
9276b1bc PJ |
1521 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1522 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1523 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1524 | |
19770b32 | 1525 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1526 | zonelist_scan: |
7fb1d9fc | 1527 | /* |
9276b1bc | 1528 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1529 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1530 | */ | |
19770b32 MG |
1531 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1532 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1533 | if (NUMA_BUILD && zlc_active && |
1534 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1535 | continue; | |
7fb1d9fc | 1536 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1537 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1538 | goto try_next_zone; |
7fb1d9fc | 1539 | |
41858966 | 1540 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1541 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1542 | unsigned long mark; |
fa5e084e MG |
1543 | int ret; |
1544 | ||
41858966 | 1545 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1546 | if (zone_watermark_ok(zone, order, mark, |
1547 | classzone_idx, alloc_flags)) | |
1548 | goto try_this_zone; | |
1549 | ||
1550 | if (zone_reclaim_mode == 0) | |
1551 | goto this_zone_full; | |
1552 | ||
1553 | ret = zone_reclaim(zone, gfp_mask, order); | |
1554 | switch (ret) { | |
1555 | case ZONE_RECLAIM_NOSCAN: | |
1556 | /* did not scan */ | |
1557 | goto try_next_zone; | |
1558 | case ZONE_RECLAIM_FULL: | |
1559 | /* scanned but unreclaimable */ | |
1560 | goto this_zone_full; | |
1561 | default: | |
1562 | /* did we reclaim enough */ | |
1563 | if (!zone_watermark_ok(zone, order, mark, | |
1564 | classzone_idx, alloc_flags)) | |
9276b1bc | 1565 | goto this_zone_full; |
0798e519 | 1566 | } |
7fb1d9fc RS |
1567 | } |
1568 | ||
fa5e084e | 1569 | try_this_zone: |
3dd28266 MG |
1570 | page = buffered_rmqueue(preferred_zone, zone, order, |
1571 | gfp_mask, migratetype); | |
0798e519 | 1572 | if (page) |
7fb1d9fc | 1573 | break; |
9276b1bc PJ |
1574 | this_zone_full: |
1575 | if (NUMA_BUILD) | |
1576 | zlc_mark_zone_full(zonelist, z); | |
1577 | try_next_zone: | |
62bc62a8 | 1578 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
d395b734 MG |
1579 | /* |
1580 | * we do zlc_setup after the first zone is tried but only | |
1581 | * if there are multiple nodes make it worthwhile | |
1582 | */ | |
9276b1bc PJ |
1583 | allowednodes = zlc_setup(zonelist, alloc_flags); |
1584 | zlc_active = 1; | |
1585 | did_zlc_setup = 1; | |
1586 | } | |
54a6eb5c | 1587 | } |
9276b1bc PJ |
1588 | |
1589 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1590 | /* Disable zlc cache for second zonelist scan */ | |
1591 | zlc_active = 0; | |
1592 | goto zonelist_scan; | |
1593 | } | |
7fb1d9fc | 1594 | return page; |
753ee728 MH |
1595 | } |
1596 | ||
11e33f6a MG |
1597 | static inline int |
1598 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1599 | unsigned long pages_reclaimed) | |
1da177e4 | 1600 | { |
11e33f6a MG |
1601 | /* Do not loop if specifically requested */ |
1602 | if (gfp_mask & __GFP_NORETRY) | |
1603 | return 0; | |
1da177e4 | 1604 | |
11e33f6a MG |
1605 | /* |
1606 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1607 | * means __GFP_NOFAIL, but that may not be true in other | |
1608 | * implementations. | |
1609 | */ | |
1610 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1611 | return 1; | |
1612 | ||
1613 | /* | |
1614 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1615 | * specified, then we retry until we no longer reclaim any pages | |
1616 | * (above), or we've reclaimed an order of pages at least as | |
1617 | * large as the allocation's order. In both cases, if the | |
1618 | * allocation still fails, we stop retrying. | |
1619 | */ | |
1620 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1621 | return 1; | |
cf40bd16 | 1622 | |
11e33f6a MG |
1623 | /* |
1624 | * Don't let big-order allocations loop unless the caller | |
1625 | * explicitly requests that. | |
1626 | */ | |
1627 | if (gfp_mask & __GFP_NOFAIL) | |
1628 | return 1; | |
1da177e4 | 1629 | |
11e33f6a MG |
1630 | return 0; |
1631 | } | |
933e312e | 1632 | |
11e33f6a MG |
1633 | static inline struct page * |
1634 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1635 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1636 | nodemask_t *nodemask, struct zone *preferred_zone, |
1637 | int migratetype) | |
11e33f6a MG |
1638 | { |
1639 | struct page *page; | |
1640 | ||
1641 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
1642 | if (!try_set_zone_oom(zonelist, gfp_mask)) { | |
1643 | schedule_timeout_uninterruptible(1); | |
1da177e4 LT |
1644 | return NULL; |
1645 | } | |
6b1de916 | 1646 | |
11e33f6a MG |
1647 | /* |
1648 | * Go through the zonelist yet one more time, keep very high watermark | |
1649 | * here, this is only to catch a parallel oom killing, we must fail if | |
1650 | * we're still under heavy pressure. | |
1651 | */ | |
1652 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1653 | order, zonelist, high_zoneidx, | |
5117f45d | 1654 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 1655 | preferred_zone, migratetype); |
7fb1d9fc | 1656 | if (page) |
11e33f6a MG |
1657 | goto out; |
1658 | ||
4365a567 KH |
1659 | if (!(gfp_mask & __GFP_NOFAIL)) { |
1660 | /* The OOM killer will not help higher order allocs */ | |
1661 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1662 | goto out; | |
1663 | /* | |
1664 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
1665 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
1666 | * The caller should handle page allocation failure by itself if | |
1667 | * it specifies __GFP_THISNODE. | |
1668 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
1669 | */ | |
1670 | if (gfp_mask & __GFP_THISNODE) | |
1671 | goto out; | |
1672 | } | |
11e33f6a | 1673 | /* Exhausted what can be done so it's blamo time */ |
4365a567 | 1674 | out_of_memory(zonelist, gfp_mask, order, nodemask); |
11e33f6a MG |
1675 | |
1676 | out: | |
1677 | clear_zonelist_oom(zonelist, gfp_mask); | |
1678 | return page; | |
1679 | } | |
1680 | ||
1681 | /* The really slow allocator path where we enter direct reclaim */ | |
1682 | static inline struct page * | |
1683 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1684 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
5117f45d | 1685 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
3dd28266 | 1686 | int migratetype, unsigned long *did_some_progress) |
11e33f6a MG |
1687 | { |
1688 | struct page *page = NULL; | |
1689 | struct reclaim_state reclaim_state; | |
1690 | struct task_struct *p = current; | |
1691 | ||
1692 | cond_resched(); | |
1693 | ||
1694 | /* We now go into synchronous reclaim */ | |
1695 | cpuset_memory_pressure_bump(); | |
11e33f6a MG |
1696 | p->flags |= PF_MEMALLOC; |
1697 | lockdep_set_current_reclaim_state(gfp_mask); | |
1698 | reclaim_state.reclaimed_slab = 0; | |
1699 | p->reclaim_state = &reclaim_state; | |
1700 | ||
1701 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1702 | ||
1703 | p->reclaim_state = NULL; | |
1704 | lockdep_clear_current_reclaim_state(); | |
1705 | p->flags &= ~PF_MEMALLOC; | |
1706 | ||
1707 | cond_resched(); | |
1708 | ||
1709 | if (order != 0) | |
1710 | drain_all_pages(); | |
1711 | ||
1712 | if (likely(*did_some_progress)) | |
1713 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1714 | zonelist, high_zoneidx, |
3dd28266 MG |
1715 | alloc_flags, preferred_zone, |
1716 | migratetype); | |
11e33f6a MG |
1717 | return page; |
1718 | } | |
1719 | ||
1da177e4 | 1720 | /* |
11e33f6a MG |
1721 | * This is called in the allocator slow-path if the allocation request is of |
1722 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 1723 | */ |
11e33f6a MG |
1724 | static inline struct page * |
1725 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1726 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1727 | nodemask_t *nodemask, struct zone *preferred_zone, |
1728 | int migratetype) | |
11e33f6a MG |
1729 | { |
1730 | struct page *page; | |
1731 | ||
1732 | do { | |
1733 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1734 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 1735 | preferred_zone, migratetype); |
11e33f6a MG |
1736 | |
1737 | if (!page && gfp_mask & __GFP_NOFAIL) | |
8aa7e847 | 1738 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
1739 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
1740 | ||
1741 | return page; | |
1742 | } | |
1743 | ||
1744 | static inline | |
1745 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
1746 | enum zone_type high_zoneidx) | |
1da177e4 | 1747 | { |
dd1a239f MG |
1748 | struct zoneref *z; |
1749 | struct zone *zone; | |
1da177e4 | 1750 | |
11e33f6a MG |
1751 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
1752 | wakeup_kswapd(zone, order); | |
1753 | } | |
cf40bd16 | 1754 | |
341ce06f PZ |
1755 | static inline int |
1756 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
1757 | { | |
1758 | struct task_struct *p = current; | |
1759 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; | |
1760 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 1761 | |
a56f57ff MG |
1762 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
1763 | BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH); | |
933e312e | 1764 | |
341ce06f PZ |
1765 | /* |
1766 | * The caller may dip into page reserves a bit more if the caller | |
1767 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1768 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1769 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1770 | */ | |
a56f57ff | 1771 | alloc_flags |= (gfp_mask & __GFP_HIGH); |
1da177e4 | 1772 | |
341ce06f PZ |
1773 | if (!wait) { |
1774 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 1775 | /* |
341ce06f PZ |
1776 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
1777 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 1778 | */ |
341ce06f | 1779 | alloc_flags &= ~ALLOC_CPUSET; |
9d0ed60f | 1780 | } else if (unlikely(rt_task(p)) && !in_interrupt()) |
341ce06f PZ |
1781 | alloc_flags |= ALLOC_HARDER; |
1782 | ||
1783 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
1784 | if (!in_interrupt() && | |
1785 | ((p->flags & PF_MEMALLOC) || | |
1786 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
1787 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1da177e4 | 1788 | } |
6b1de916 | 1789 | |
341ce06f PZ |
1790 | return alloc_flags; |
1791 | } | |
1792 | ||
11e33f6a MG |
1793 | static inline struct page * |
1794 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
1795 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1796 | nodemask_t *nodemask, struct zone *preferred_zone, |
1797 | int migratetype) | |
11e33f6a MG |
1798 | { |
1799 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1800 | struct page *page = NULL; | |
1801 | int alloc_flags; | |
1802 | unsigned long pages_reclaimed = 0; | |
1803 | unsigned long did_some_progress; | |
1804 | struct task_struct *p = current; | |
1da177e4 | 1805 | |
72807a74 MG |
1806 | /* |
1807 | * In the slowpath, we sanity check order to avoid ever trying to | |
1808 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
1809 | * be using allocators in order of preference for an area that is | |
1810 | * too large. | |
1811 | */ | |
1fc28b70 MG |
1812 | if (order >= MAX_ORDER) { |
1813 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 1814 | return NULL; |
1fc28b70 | 1815 | } |
1da177e4 | 1816 | |
952f3b51 CL |
1817 | /* |
1818 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1819 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1820 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1821 | * using a larger set of nodes after it has established that the | |
1822 | * allowed per node queues are empty and that nodes are | |
1823 | * over allocated. | |
1824 | */ | |
1825 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1826 | goto nopage; | |
1827 | ||
cc4a6851 | 1828 | restart: |
11e33f6a | 1829 | wake_all_kswapd(order, zonelist, high_zoneidx); |
1da177e4 | 1830 | |
9bf2229f | 1831 | /* |
7fb1d9fc RS |
1832 | * OK, we're below the kswapd watermark and have kicked background |
1833 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1834 | * to how we want to proceed. | |
9bf2229f | 1835 | */ |
341ce06f | 1836 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 1837 | |
341ce06f | 1838 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 1839 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
1840 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
1841 | preferred_zone, migratetype); | |
7fb1d9fc RS |
1842 | if (page) |
1843 | goto got_pg; | |
1da177e4 | 1844 | |
b43a57bb | 1845 | rebalance: |
11e33f6a | 1846 | /* Allocate without watermarks if the context allows */ |
341ce06f PZ |
1847 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
1848 | page = __alloc_pages_high_priority(gfp_mask, order, | |
1849 | zonelist, high_zoneidx, nodemask, | |
1850 | preferred_zone, migratetype); | |
1851 | if (page) | |
1852 | goto got_pg; | |
1da177e4 LT |
1853 | } |
1854 | ||
1855 | /* Atomic allocations - we can't balance anything */ | |
1856 | if (!wait) | |
1857 | goto nopage; | |
1858 | ||
341ce06f PZ |
1859 | /* Avoid recursion of direct reclaim */ |
1860 | if (p->flags & PF_MEMALLOC) | |
1861 | goto nopage; | |
1862 | ||
6583bb64 DR |
1863 | /* Avoid allocations with no watermarks from looping endlessly */ |
1864 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
1865 | goto nopage; | |
1866 | ||
11e33f6a MG |
1867 | /* Try direct reclaim and then allocating */ |
1868 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
1869 | zonelist, high_zoneidx, | |
1870 | nodemask, | |
5117f45d | 1871 | alloc_flags, preferred_zone, |
3dd28266 | 1872 | migratetype, &did_some_progress); |
11e33f6a MG |
1873 | if (page) |
1874 | goto got_pg; | |
1da177e4 | 1875 | |
e33c3b5e | 1876 | /* |
11e33f6a MG |
1877 | * If we failed to make any progress reclaiming, then we are |
1878 | * running out of options and have to consider going OOM | |
e33c3b5e | 1879 | */ |
11e33f6a MG |
1880 | if (!did_some_progress) { |
1881 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
1882 | if (oom_killer_disabled) |
1883 | goto nopage; | |
11e33f6a MG |
1884 | page = __alloc_pages_may_oom(gfp_mask, order, |
1885 | zonelist, high_zoneidx, | |
3dd28266 MG |
1886 | nodemask, preferred_zone, |
1887 | migratetype); | |
11e33f6a MG |
1888 | if (page) |
1889 | goto got_pg; | |
1da177e4 | 1890 | |
11e33f6a | 1891 | /* |
82553a93 DR |
1892 | * The OOM killer does not trigger for high-order |
1893 | * ~__GFP_NOFAIL allocations so if no progress is being | |
1894 | * made, there are no other options and retrying is | |
1895 | * unlikely to help. | |
11e33f6a | 1896 | */ |
82553a93 DR |
1897 | if (order > PAGE_ALLOC_COSTLY_ORDER && |
1898 | !(gfp_mask & __GFP_NOFAIL)) | |
11e33f6a | 1899 | goto nopage; |
e2c55dc8 | 1900 | |
ff0ceb9d DR |
1901 | goto restart; |
1902 | } | |
1da177e4 LT |
1903 | } |
1904 | ||
11e33f6a | 1905 | /* Check if we should retry the allocation */ |
a41f24ea | 1906 | pages_reclaimed += did_some_progress; |
11e33f6a MG |
1907 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
1908 | /* Wait for some write requests to complete then retry */ | |
8aa7e847 | 1909 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
1910 | goto rebalance; |
1911 | } | |
1912 | ||
1913 | nopage: | |
1914 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1915 | printk(KERN_WARNING "%s: page allocation failure." | |
1916 | " order:%d, mode:0x%x\n", | |
1917 | p->comm, order, gfp_mask); | |
1918 | dump_stack(); | |
578c2fd6 | 1919 | show_mem(); |
1da177e4 | 1920 | } |
b1eeab67 | 1921 | return page; |
1da177e4 | 1922 | got_pg: |
b1eeab67 VN |
1923 | if (kmemcheck_enabled) |
1924 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1da177e4 | 1925 | return page; |
11e33f6a | 1926 | |
1da177e4 | 1927 | } |
11e33f6a MG |
1928 | |
1929 | /* | |
1930 | * This is the 'heart' of the zoned buddy allocator. | |
1931 | */ | |
1932 | struct page * | |
1933 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
1934 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1935 | { | |
1936 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 1937 | struct zone *preferred_zone; |
11e33f6a | 1938 | struct page *page; |
3dd28266 | 1939 | int migratetype = allocflags_to_migratetype(gfp_mask); |
11e33f6a | 1940 | |
dcce284a BH |
1941 | gfp_mask &= gfp_allowed_mask; |
1942 | ||
11e33f6a MG |
1943 | lockdep_trace_alloc(gfp_mask); |
1944 | ||
1945 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1946 | ||
1947 | if (should_fail_alloc_page(gfp_mask, order)) | |
1948 | return NULL; | |
1949 | ||
1950 | /* | |
1951 | * Check the zones suitable for the gfp_mask contain at least one | |
1952 | * valid zone. It's possible to have an empty zonelist as a result | |
1953 | * of GFP_THISNODE and a memoryless node | |
1954 | */ | |
1955 | if (unlikely(!zonelist->_zonerefs->zone)) | |
1956 | return NULL; | |
1957 | ||
5117f45d MG |
1958 | /* The preferred zone is used for statistics later */ |
1959 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); | |
1960 | if (!preferred_zone) | |
1961 | return NULL; | |
1962 | ||
1963 | /* First allocation attempt */ | |
11e33f6a | 1964 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
5117f45d | 1965 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
3dd28266 | 1966 | preferred_zone, migratetype); |
11e33f6a MG |
1967 | if (unlikely(!page)) |
1968 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 1969 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 1970 | preferred_zone, migratetype); |
11e33f6a | 1971 | |
4b4f278c | 1972 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
11e33f6a | 1973 | return page; |
1da177e4 | 1974 | } |
d239171e | 1975 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
1976 | |
1977 | /* | |
1978 | * Common helper functions. | |
1979 | */ | |
920c7a5d | 1980 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 1981 | { |
945a1113 AM |
1982 | struct page *page; |
1983 | ||
1984 | /* | |
1985 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
1986 | * a highmem page | |
1987 | */ | |
1988 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
1989 | ||
1da177e4 LT |
1990 | page = alloc_pages(gfp_mask, order); |
1991 | if (!page) | |
1992 | return 0; | |
1993 | return (unsigned long) page_address(page); | |
1994 | } | |
1da177e4 LT |
1995 | EXPORT_SYMBOL(__get_free_pages); |
1996 | ||
920c7a5d | 1997 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 1998 | { |
945a1113 | 1999 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2000 | } |
1da177e4 LT |
2001 | EXPORT_SYMBOL(get_zeroed_page); |
2002 | ||
2003 | void __pagevec_free(struct pagevec *pvec) | |
2004 | { | |
2005 | int i = pagevec_count(pvec); | |
2006 | ||
4b4f278c MG |
2007 | while (--i >= 0) { |
2008 | trace_mm_pagevec_free(pvec->pages[i], pvec->cold); | |
1da177e4 | 2009 | free_hot_cold_page(pvec->pages[i], pvec->cold); |
4b4f278c | 2010 | } |
1da177e4 LT |
2011 | } |
2012 | ||
920c7a5d | 2013 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2014 | { |
b5810039 | 2015 | if (put_page_testzero(page)) { |
4b4f278c | 2016 | trace_mm_page_free_direct(page, order); |
1da177e4 LT |
2017 | if (order == 0) |
2018 | free_hot_page(page); | |
2019 | else | |
2020 | __free_pages_ok(page, order); | |
2021 | } | |
2022 | } | |
2023 | ||
2024 | EXPORT_SYMBOL(__free_pages); | |
2025 | ||
920c7a5d | 2026 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2027 | { |
2028 | if (addr != 0) { | |
725d704e | 2029 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2030 | __free_pages(virt_to_page((void *)addr), order); |
2031 | } | |
2032 | } | |
2033 | ||
2034 | EXPORT_SYMBOL(free_pages); | |
2035 | ||
2be0ffe2 TT |
2036 | /** |
2037 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2038 | * @size: the number of bytes to allocate | |
2039 | * @gfp_mask: GFP flags for the allocation | |
2040 | * | |
2041 | * This function is similar to alloc_pages(), except that it allocates the | |
2042 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2043 | * allocate memory in power-of-two pages. | |
2044 | * | |
2045 | * This function is also limited by MAX_ORDER. | |
2046 | * | |
2047 | * Memory allocated by this function must be released by free_pages_exact(). | |
2048 | */ | |
2049 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2050 | { | |
2051 | unsigned int order = get_order(size); | |
2052 | unsigned long addr; | |
2053 | ||
2054 | addr = __get_free_pages(gfp_mask, order); | |
2055 | if (addr) { | |
2056 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2057 | unsigned long used = addr + PAGE_ALIGN(size); | |
2058 | ||
5bfd7560 | 2059 | split_page(virt_to_page((void *)addr), order); |
2be0ffe2 TT |
2060 | while (used < alloc_end) { |
2061 | free_page(used); | |
2062 | used += PAGE_SIZE; | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | return (void *)addr; | |
2067 | } | |
2068 | EXPORT_SYMBOL(alloc_pages_exact); | |
2069 | ||
2070 | /** | |
2071 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2072 | * @virt: the value returned by alloc_pages_exact. | |
2073 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2074 | * | |
2075 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2076 | */ | |
2077 | void free_pages_exact(void *virt, size_t size) | |
2078 | { | |
2079 | unsigned long addr = (unsigned long)virt; | |
2080 | unsigned long end = addr + PAGE_ALIGN(size); | |
2081 | ||
2082 | while (addr < end) { | |
2083 | free_page(addr); | |
2084 | addr += PAGE_SIZE; | |
2085 | } | |
2086 | } | |
2087 | EXPORT_SYMBOL(free_pages_exact); | |
2088 | ||
1da177e4 LT |
2089 | static unsigned int nr_free_zone_pages(int offset) |
2090 | { | |
dd1a239f | 2091 | struct zoneref *z; |
54a6eb5c MG |
2092 | struct zone *zone; |
2093 | ||
e310fd43 | 2094 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2095 | unsigned int sum = 0; |
2096 | ||
0e88460d | 2097 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2098 | |
54a6eb5c | 2099 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2100 | unsigned long size = zone->present_pages; |
41858966 | 2101 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2102 | if (size > high) |
2103 | sum += size - high; | |
1da177e4 LT |
2104 | } |
2105 | ||
2106 | return sum; | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2111 | */ | |
2112 | unsigned int nr_free_buffer_pages(void) | |
2113 | { | |
af4ca457 | 2114 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2115 | } |
c2f1a551 | 2116 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2117 | |
2118 | /* | |
2119 | * Amount of free RAM allocatable within all zones | |
2120 | */ | |
2121 | unsigned int nr_free_pagecache_pages(void) | |
2122 | { | |
2a1e274a | 2123 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2124 | } |
08e0f6a9 CL |
2125 | |
2126 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2127 | { |
08e0f6a9 | 2128 | if (NUMA_BUILD) |
25ba77c1 | 2129 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2130 | } |
1da177e4 | 2131 | |
1da177e4 LT |
2132 | void si_meminfo(struct sysinfo *val) |
2133 | { | |
2134 | val->totalram = totalram_pages; | |
2135 | val->sharedram = 0; | |
d23ad423 | 2136 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2137 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2138 | val->totalhigh = totalhigh_pages; |
2139 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2140 | val->mem_unit = PAGE_SIZE; |
2141 | } | |
2142 | ||
2143 | EXPORT_SYMBOL(si_meminfo); | |
2144 | ||
2145 | #ifdef CONFIG_NUMA | |
2146 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2147 | { | |
2148 | pg_data_t *pgdat = NODE_DATA(nid); | |
2149 | ||
2150 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2151 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2152 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2153 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2154 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2155 | NR_FREE_PAGES); | |
98d2b0eb CL |
2156 | #else |
2157 | val->totalhigh = 0; | |
2158 | val->freehigh = 0; | |
2159 | #endif | |
1da177e4 LT |
2160 | val->mem_unit = PAGE_SIZE; |
2161 | } | |
2162 | #endif | |
2163 | ||
2164 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
2165 | ||
2166 | /* | |
2167 | * Show free area list (used inside shift_scroll-lock stuff) | |
2168 | * We also calculate the percentage fragmentation. We do this by counting the | |
2169 | * memory on each free list with the exception of the first item on the list. | |
2170 | */ | |
2171 | void show_free_areas(void) | |
2172 | { | |
c7241913 | 2173 | int cpu; |
1da177e4 LT |
2174 | struct zone *zone; |
2175 | ||
ee99c71c | 2176 | for_each_populated_zone(zone) { |
c7241913 JS |
2177 | show_node(zone); |
2178 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2179 | |
6b482c67 | 2180 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2181 | struct per_cpu_pageset *pageset; |
2182 | ||
e7c8d5c9 | 2183 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 2184 | |
3dfa5721 CL |
2185 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2186 | cpu, pageset->pcp.high, | |
2187 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2188 | } |
2189 | } | |
2190 | ||
a731286d KM |
2191 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2192 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2193 | " unevictable:%lu" |
b76146ed | 2194 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2195 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
4b02108a | 2196 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe | 2197 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2198 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2199 | global_page_state(NR_ISOLATED_ANON), |
2200 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2201 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2202 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2203 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2204 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2205 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2206 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2207 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2208 | global_page_state(NR_SLAB_RECLAIMABLE), |
2209 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2210 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2211 | global_page_state(NR_SHMEM), |
a25700a5 AM |
2212 | global_page_state(NR_PAGETABLE), |
2213 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2214 | |
ee99c71c | 2215 | for_each_populated_zone(zone) { |
1da177e4 LT |
2216 | int i; |
2217 | ||
2218 | show_node(zone); | |
2219 | printk("%s" | |
2220 | " free:%lukB" | |
2221 | " min:%lukB" | |
2222 | " low:%lukB" | |
2223 | " high:%lukB" | |
4f98a2fe RR |
2224 | " active_anon:%lukB" |
2225 | " inactive_anon:%lukB" | |
2226 | " active_file:%lukB" | |
2227 | " inactive_file:%lukB" | |
7b854121 | 2228 | " unevictable:%lukB" |
a731286d KM |
2229 | " isolated(anon):%lukB" |
2230 | " isolated(file):%lukB" | |
1da177e4 | 2231 | " present:%lukB" |
4a0aa73f KM |
2232 | " mlocked:%lukB" |
2233 | " dirty:%lukB" | |
2234 | " writeback:%lukB" | |
2235 | " mapped:%lukB" | |
4b02108a | 2236 | " shmem:%lukB" |
4a0aa73f KM |
2237 | " slab_reclaimable:%lukB" |
2238 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 2239 | " kernel_stack:%lukB" |
4a0aa73f KM |
2240 | " pagetables:%lukB" |
2241 | " unstable:%lukB" | |
2242 | " bounce:%lukB" | |
2243 | " writeback_tmp:%lukB" | |
1da177e4 LT |
2244 | " pages_scanned:%lu" |
2245 | " all_unreclaimable? %s" | |
2246 | "\n", | |
2247 | zone->name, | |
d23ad423 | 2248 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
2249 | K(min_wmark_pages(zone)), |
2250 | K(low_wmark_pages(zone)), | |
2251 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
2252 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2253 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2254 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2255 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 2256 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
2257 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
2258 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 2259 | K(zone->present_pages), |
4a0aa73f KM |
2260 | K(zone_page_state(zone, NR_MLOCK)), |
2261 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
2262 | K(zone_page_state(zone, NR_WRITEBACK)), | |
2263 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 2264 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
2265 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
2266 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
2267 | zone_page_state(zone, NR_KERNEL_STACK) * |
2268 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
2269 | K(zone_page_state(zone, NR_PAGETABLE)), |
2270 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
2271 | K(zone_page_state(zone, NR_BOUNCE)), | |
2272 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), | |
1da177e4 | 2273 | zone->pages_scanned, |
e815af95 | 2274 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
2275 | ); |
2276 | printk("lowmem_reserve[]:"); | |
2277 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2278 | printk(" %lu", zone->lowmem_reserve[i]); | |
2279 | printk("\n"); | |
2280 | } | |
2281 | ||
ee99c71c | 2282 | for_each_populated_zone(zone) { |
8f9de51a | 2283 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
2284 | |
2285 | show_node(zone); | |
2286 | printk("%s: ", zone->name); | |
1da177e4 LT |
2287 | |
2288 | spin_lock_irqsave(&zone->lock, flags); | |
2289 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2290 | nr[order] = zone->free_area[order].nr_free; |
2291 | total += nr[order] << order; | |
1da177e4 LT |
2292 | } |
2293 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2294 | for (order = 0; order < MAX_ORDER; order++) |
2295 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2296 | printk("= %lukB\n", K(total)); |
2297 | } | |
2298 | ||
e6f3602d LW |
2299 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2300 | ||
1da177e4 LT |
2301 | show_swap_cache_info(); |
2302 | } | |
2303 | ||
19770b32 MG |
2304 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2305 | { | |
2306 | zoneref->zone = zone; | |
2307 | zoneref->zone_idx = zone_idx(zone); | |
2308 | } | |
2309 | ||
1da177e4 LT |
2310 | /* |
2311 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2312 | * |
2313 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2314 | */ |
f0c0b2b8 KH |
2315 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2316 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2317 | { |
1a93205b CL |
2318 | struct zone *zone; |
2319 | ||
98d2b0eb | 2320 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2321 | zone_type++; |
02a68a5e CL |
2322 | |
2323 | do { | |
2f6726e5 | 2324 | zone_type--; |
070f8032 | 2325 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2326 | if (populated_zone(zone)) { |
dd1a239f MG |
2327 | zoneref_set_zone(zone, |
2328 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2329 | check_highest_zone(zone_type); |
1da177e4 | 2330 | } |
02a68a5e | 2331 | |
2f6726e5 | 2332 | } while (zone_type); |
070f8032 | 2333 | return nr_zones; |
1da177e4 LT |
2334 | } |
2335 | ||
f0c0b2b8 KH |
2336 | |
2337 | /* | |
2338 | * zonelist_order: | |
2339 | * 0 = automatic detection of better ordering. | |
2340 | * 1 = order by ([node] distance, -zonetype) | |
2341 | * 2 = order by (-zonetype, [node] distance) | |
2342 | * | |
2343 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2344 | * the same zonelist. So only NUMA can configure this param. | |
2345 | */ | |
2346 | #define ZONELIST_ORDER_DEFAULT 0 | |
2347 | #define ZONELIST_ORDER_NODE 1 | |
2348 | #define ZONELIST_ORDER_ZONE 2 | |
2349 | ||
2350 | /* zonelist order in the kernel. | |
2351 | * set_zonelist_order() will set this to NODE or ZONE. | |
2352 | */ | |
2353 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2354 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2355 | ||
2356 | ||
1da177e4 | 2357 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2358 | /* The value user specified ....changed by config */ |
2359 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2360 | /* string for sysctl */ | |
2361 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2362 | char numa_zonelist_order[16] = "default"; | |
2363 | ||
2364 | /* | |
2365 | * interface for configure zonelist ordering. | |
2366 | * command line option "numa_zonelist_order" | |
2367 | * = "[dD]efault - default, automatic configuration. | |
2368 | * = "[nN]ode - order by node locality, then by zone within node | |
2369 | * = "[zZ]one - order by zone, then by locality within zone | |
2370 | */ | |
2371 | ||
2372 | static int __parse_numa_zonelist_order(char *s) | |
2373 | { | |
2374 | if (*s == 'd' || *s == 'D') { | |
2375 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2376 | } else if (*s == 'n' || *s == 'N') { | |
2377 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2378 | } else if (*s == 'z' || *s == 'Z') { | |
2379 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2380 | } else { | |
2381 | printk(KERN_WARNING | |
2382 | "Ignoring invalid numa_zonelist_order value: " | |
2383 | "%s\n", s); | |
2384 | return -EINVAL; | |
2385 | } | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | static __init int setup_numa_zonelist_order(char *s) | |
2390 | { | |
2391 | if (s) | |
2392 | return __parse_numa_zonelist_order(s); | |
2393 | return 0; | |
2394 | } | |
2395 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2396 | ||
2397 | /* | |
2398 | * sysctl handler for numa_zonelist_order | |
2399 | */ | |
2400 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 2401 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
2402 | loff_t *ppos) |
2403 | { | |
2404 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2405 | int ret; | |
443c6f14 | 2406 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 2407 | |
443c6f14 | 2408 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 2409 | if (write) |
443c6f14 | 2410 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 2411 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 2412 | if (ret) |
443c6f14 | 2413 | goto out; |
f0c0b2b8 KH |
2414 | if (write) { |
2415 | int oldval = user_zonelist_order; | |
2416 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2417 | /* | |
2418 | * bogus value. restore saved string | |
2419 | */ | |
2420 | strncpy((char*)table->data, saved_string, | |
2421 | NUMA_ZONELIST_ORDER_LEN); | |
2422 | user_zonelist_order = oldval; | |
2423 | } else if (oldval != user_zonelist_order) | |
2424 | build_all_zonelists(); | |
2425 | } | |
443c6f14 AK |
2426 | out: |
2427 | mutex_unlock(&zl_order_mutex); | |
2428 | return ret; | |
f0c0b2b8 KH |
2429 | } |
2430 | ||
2431 | ||
62bc62a8 | 2432 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
2433 | static int node_load[MAX_NUMNODES]; |
2434 | ||
1da177e4 | 2435 | /** |
4dc3b16b | 2436 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2437 | * @node: node whose fallback list we're appending |
2438 | * @used_node_mask: nodemask_t of already used nodes | |
2439 | * | |
2440 | * We use a number of factors to determine which is the next node that should | |
2441 | * appear on a given node's fallback list. The node should not have appeared | |
2442 | * already in @node's fallback list, and it should be the next closest node | |
2443 | * according to the distance array (which contains arbitrary distance values | |
2444 | * from each node to each node in the system), and should also prefer nodes | |
2445 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2446 | * on them otherwise. | |
2447 | * It returns -1 if no node is found. | |
2448 | */ | |
f0c0b2b8 | 2449 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2450 | { |
4cf808eb | 2451 | int n, val; |
1da177e4 LT |
2452 | int min_val = INT_MAX; |
2453 | int best_node = -1; | |
a70f7302 | 2454 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2455 | |
4cf808eb LT |
2456 | /* Use the local node if we haven't already */ |
2457 | if (!node_isset(node, *used_node_mask)) { | |
2458 | node_set(node, *used_node_mask); | |
2459 | return node; | |
2460 | } | |
1da177e4 | 2461 | |
37b07e41 | 2462 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2463 | |
2464 | /* Don't want a node to appear more than once */ | |
2465 | if (node_isset(n, *used_node_mask)) | |
2466 | continue; | |
2467 | ||
1da177e4 LT |
2468 | /* Use the distance array to find the distance */ |
2469 | val = node_distance(node, n); | |
2470 | ||
4cf808eb LT |
2471 | /* Penalize nodes under us ("prefer the next node") */ |
2472 | val += (n < node); | |
2473 | ||
1da177e4 | 2474 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2475 | tmp = cpumask_of_node(n); |
2476 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2477 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2478 | ||
2479 | /* Slight preference for less loaded node */ | |
2480 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2481 | val += node_load[n]; | |
2482 | ||
2483 | if (val < min_val) { | |
2484 | min_val = val; | |
2485 | best_node = n; | |
2486 | } | |
2487 | } | |
2488 | ||
2489 | if (best_node >= 0) | |
2490 | node_set(best_node, *used_node_mask); | |
2491 | ||
2492 | return best_node; | |
2493 | } | |
2494 | ||
f0c0b2b8 KH |
2495 | |
2496 | /* | |
2497 | * Build zonelists ordered by node and zones within node. | |
2498 | * This results in maximum locality--normal zone overflows into local | |
2499 | * DMA zone, if any--but risks exhausting DMA zone. | |
2500 | */ | |
2501 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2502 | { |
f0c0b2b8 | 2503 | int j; |
1da177e4 | 2504 | struct zonelist *zonelist; |
f0c0b2b8 | 2505 | |
54a6eb5c | 2506 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2507 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2508 | ; |
2509 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2510 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2511 | zonelist->_zonerefs[j].zone = NULL; |
2512 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2513 | } |
2514 | ||
523b9458 CL |
2515 | /* |
2516 | * Build gfp_thisnode zonelists | |
2517 | */ | |
2518 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2519 | { | |
523b9458 CL |
2520 | int j; |
2521 | struct zonelist *zonelist; | |
2522 | ||
54a6eb5c MG |
2523 | zonelist = &pgdat->node_zonelists[1]; |
2524 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2525 | zonelist->_zonerefs[j].zone = NULL; |
2526 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2527 | } |
2528 | ||
f0c0b2b8 KH |
2529 | /* |
2530 | * Build zonelists ordered by zone and nodes within zones. | |
2531 | * This results in conserving DMA zone[s] until all Normal memory is | |
2532 | * exhausted, but results in overflowing to remote node while memory | |
2533 | * may still exist in local DMA zone. | |
2534 | */ | |
2535 | static int node_order[MAX_NUMNODES]; | |
2536 | ||
2537 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2538 | { | |
f0c0b2b8 KH |
2539 | int pos, j, node; |
2540 | int zone_type; /* needs to be signed */ | |
2541 | struct zone *z; | |
2542 | struct zonelist *zonelist; | |
2543 | ||
54a6eb5c MG |
2544 | zonelist = &pgdat->node_zonelists[0]; |
2545 | pos = 0; | |
2546 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2547 | for (j = 0; j < nr_nodes; j++) { | |
2548 | node = node_order[j]; | |
2549 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2550 | if (populated_zone(z)) { | |
dd1a239f MG |
2551 | zoneref_set_zone(z, |
2552 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2553 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2554 | } |
2555 | } | |
f0c0b2b8 | 2556 | } |
dd1a239f MG |
2557 | zonelist->_zonerefs[pos].zone = NULL; |
2558 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2559 | } |
2560 | ||
2561 | static int default_zonelist_order(void) | |
2562 | { | |
2563 | int nid, zone_type; | |
2564 | unsigned long low_kmem_size,total_size; | |
2565 | struct zone *z; | |
2566 | int average_size; | |
2567 | /* | |
2568 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2569 | * If they are really small and used heavily, the system can fall | |
2570 | * into OOM very easily. | |
2571 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2572 | */ | |
2573 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2574 | low_kmem_size = 0; | |
2575 | total_size = 0; | |
2576 | for_each_online_node(nid) { | |
2577 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2578 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2579 | if (populated_zone(z)) { | |
2580 | if (zone_type < ZONE_NORMAL) | |
2581 | low_kmem_size += z->present_pages; | |
2582 | total_size += z->present_pages; | |
2583 | } | |
2584 | } | |
2585 | } | |
2586 | if (!low_kmem_size || /* there are no DMA area. */ | |
2587 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2588 | return ZONELIST_ORDER_NODE; | |
2589 | /* | |
2590 | * look into each node's config. | |
2591 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2592 | * local memory, NODE_ORDER may be suitable. | |
2593 | */ | |
37b07e41 LS |
2594 | average_size = total_size / |
2595 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2596 | for_each_online_node(nid) { |
2597 | low_kmem_size = 0; | |
2598 | total_size = 0; | |
2599 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2600 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2601 | if (populated_zone(z)) { | |
2602 | if (zone_type < ZONE_NORMAL) | |
2603 | low_kmem_size += z->present_pages; | |
2604 | total_size += z->present_pages; | |
2605 | } | |
2606 | } | |
2607 | if (low_kmem_size && | |
2608 | total_size > average_size && /* ignore small node */ | |
2609 | low_kmem_size > total_size * 70/100) | |
2610 | return ZONELIST_ORDER_NODE; | |
2611 | } | |
2612 | return ZONELIST_ORDER_ZONE; | |
2613 | } | |
2614 | ||
2615 | static void set_zonelist_order(void) | |
2616 | { | |
2617 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2618 | current_zonelist_order = default_zonelist_order(); | |
2619 | else | |
2620 | current_zonelist_order = user_zonelist_order; | |
2621 | } | |
2622 | ||
2623 | static void build_zonelists(pg_data_t *pgdat) | |
2624 | { | |
2625 | int j, node, load; | |
2626 | enum zone_type i; | |
1da177e4 | 2627 | nodemask_t used_mask; |
f0c0b2b8 KH |
2628 | int local_node, prev_node; |
2629 | struct zonelist *zonelist; | |
2630 | int order = current_zonelist_order; | |
1da177e4 LT |
2631 | |
2632 | /* initialize zonelists */ | |
523b9458 | 2633 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2634 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2635 | zonelist->_zonerefs[0].zone = NULL; |
2636 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2637 | } |
2638 | ||
2639 | /* NUMA-aware ordering of nodes */ | |
2640 | local_node = pgdat->node_id; | |
62bc62a8 | 2641 | load = nr_online_nodes; |
1da177e4 LT |
2642 | prev_node = local_node; |
2643 | nodes_clear(used_mask); | |
f0c0b2b8 | 2644 | |
f0c0b2b8 KH |
2645 | memset(node_order, 0, sizeof(node_order)); |
2646 | j = 0; | |
2647 | ||
1da177e4 | 2648 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2649 | int distance = node_distance(local_node, node); |
2650 | ||
2651 | /* | |
2652 | * If another node is sufficiently far away then it is better | |
2653 | * to reclaim pages in a zone before going off node. | |
2654 | */ | |
2655 | if (distance > RECLAIM_DISTANCE) | |
2656 | zone_reclaim_mode = 1; | |
2657 | ||
1da177e4 LT |
2658 | /* |
2659 | * We don't want to pressure a particular node. | |
2660 | * So adding penalty to the first node in same | |
2661 | * distance group to make it round-robin. | |
2662 | */ | |
9eeff239 | 2663 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2664 | node_load[node] = load; |
2665 | ||
1da177e4 LT |
2666 | prev_node = node; |
2667 | load--; | |
f0c0b2b8 KH |
2668 | if (order == ZONELIST_ORDER_NODE) |
2669 | build_zonelists_in_node_order(pgdat, node); | |
2670 | else | |
2671 | node_order[j++] = node; /* remember order */ | |
2672 | } | |
1da177e4 | 2673 | |
f0c0b2b8 KH |
2674 | if (order == ZONELIST_ORDER_ZONE) { |
2675 | /* calculate node order -- i.e., DMA last! */ | |
2676 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2677 | } |
523b9458 CL |
2678 | |
2679 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2680 | } |
2681 | ||
9276b1bc | 2682 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2683 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2684 | { |
54a6eb5c MG |
2685 | struct zonelist *zonelist; |
2686 | struct zonelist_cache *zlc; | |
dd1a239f | 2687 | struct zoneref *z; |
9276b1bc | 2688 | |
54a6eb5c MG |
2689 | zonelist = &pgdat->node_zonelists[0]; |
2690 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2691 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2692 | for (z = zonelist->_zonerefs; z->zone; z++) |
2693 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2694 | } |
2695 | ||
f0c0b2b8 | 2696 | |
1da177e4 LT |
2697 | #else /* CONFIG_NUMA */ |
2698 | ||
f0c0b2b8 KH |
2699 | static void set_zonelist_order(void) |
2700 | { | |
2701 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2702 | } | |
2703 | ||
2704 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2705 | { |
19655d34 | 2706 | int node, local_node; |
54a6eb5c MG |
2707 | enum zone_type j; |
2708 | struct zonelist *zonelist; | |
1da177e4 LT |
2709 | |
2710 | local_node = pgdat->node_id; | |
1da177e4 | 2711 | |
54a6eb5c MG |
2712 | zonelist = &pgdat->node_zonelists[0]; |
2713 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2714 | |
54a6eb5c MG |
2715 | /* |
2716 | * Now we build the zonelist so that it contains the zones | |
2717 | * of all the other nodes. | |
2718 | * We don't want to pressure a particular node, so when | |
2719 | * building the zones for node N, we make sure that the | |
2720 | * zones coming right after the local ones are those from | |
2721 | * node N+1 (modulo N) | |
2722 | */ | |
2723 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2724 | if (!node_online(node)) | |
2725 | continue; | |
2726 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2727 | MAX_NR_ZONES - 1); | |
1da177e4 | 2728 | } |
54a6eb5c MG |
2729 | for (node = 0; node < local_node; node++) { |
2730 | if (!node_online(node)) | |
2731 | continue; | |
2732 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2733 | MAX_NR_ZONES - 1); | |
2734 | } | |
2735 | ||
dd1a239f MG |
2736 | zonelist->_zonerefs[j].zone = NULL; |
2737 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
2738 | } |
2739 | ||
9276b1bc | 2740 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2741 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2742 | { |
54a6eb5c | 2743 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
2744 | } |
2745 | ||
1da177e4 LT |
2746 | #endif /* CONFIG_NUMA */ |
2747 | ||
9b1a4d38 | 2748 | /* return values int ....just for stop_machine() */ |
f0c0b2b8 | 2749 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2750 | { |
6811378e | 2751 | int nid; |
9276b1bc | 2752 | |
7f9cfb31 BL |
2753 | #ifdef CONFIG_NUMA |
2754 | memset(node_load, 0, sizeof(node_load)); | |
2755 | #endif | |
9276b1bc | 2756 | for_each_online_node(nid) { |
7ea1530a CL |
2757 | pg_data_t *pgdat = NODE_DATA(nid); |
2758 | ||
2759 | build_zonelists(pgdat); | |
2760 | build_zonelist_cache(pgdat); | |
9276b1bc | 2761 | } |
6811378e YG |
2762 | return 0; |
2763 | } | |
2764 | ||
f0c0b2b8 | 2765 | void build_all_zonelists(void) |
6811378e | 2766 | { |
f0c0b2b8 KH |
2767 | set_zonelist_order(); |
2768 | ||
6811378e | 2769 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2770 | __build_all_zonelists(NULL); |
68ad8df4 | 2771 | mminit_verify_zonelist(); |
6811378e YG |
2772 | cpuset_init_current_mems_allowed(); |
2773 | } else { | |
183ff22b | 2774 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 2775 | of zonelist */ |
9b1a4d38 | 2776 | stop_machine(__build_all_zonelists, NULL, NULL); |
6811378e YG |
2777 | /* cpuset refresh routine should be here */ |
2778 | } | |
bd1e22b8 | 2779 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2780 | /* |
2781 | * Disable grouping by mobility if the number of pages in the | |
2782 | * system is too low to allow the mechanism to work. It would be | |
2783 | * more accurate, but expensive to check per-zone. This check is | |
2784 | * made on memory-hotadd so a system can start with mobility | |
2785 | * disabled and enable it later | |
2786 | */ | |
d9c23400 | 2787 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2788 | page_group_by_mobility_disabled = 1; |
2789 | else | |
2790 | page_group_by_mobility_disabled = 0; | |
2791 | ||
2792 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2793 | "Total pages: %ld\n", | |
62bc62a8 | 2794 | nr_online_nodes, |
f0c0b2b8 | 2795 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 2796 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2797 | vm_total_pages); |
2798 | #ifdef CONFIG_NUMA | |
2799 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2800 | #endif | |
1da177e4 LT |
2801 | } |
2802 | ||
2803 | /* | |
2804 | * Helper functions to size the waitqueue hash table. | |
2805 | * Essentially these want to choose hash table sizes sufficiently | |
2806 | * large so that collisions trying to wait on pages are rare. | |
2807 | * But in fact, the number of active page waitqueues on typical | |
2808 | * systems is ridiculously low, less than 200. So this is even | |
2809 | * conservative, even though it seems large. | |
2810 | * | |
2811 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2812 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2813 | */ | |
2814 | #define PAGES_PER_WAITQUEUE 256 | |
2815 | ||
cca448fe | 2816 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2817 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2818 | { |
2819 | unsigned long size = 1; | |
2820 | ||
2821 | pages /= PAGES_PER_WAITQUEUE; | |
2822 | ||
2823 | while (size < pages) | |
2824 | size <<= 1; | |
2825 | ||
2826 | /* | |
2827 | * Once we have dozens or even hundreds of threads sleeping | |
2828 | * on IO we've got bigger problems than wait queue collision. | |
2829 | * Limit the size of the wait table to a reasonable size. | |
2830 | */ | |
2831 | size = min(size, 4096UL); | |
2832 | ||
2833 | return max(size, 4UL); | |
2834 | } | |
cca448fe YG |
2835 | #else |
2836 | /* | |
2837 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2838 | * a suitable size for its wait_table. So we use the maximum size now. | |
2839 | * | |
2840 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2841 | * | |
2842 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2843 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2844 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2845 | * | |
2846 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2847 | * or more by the traditional way. (See above). It equals: | |
2848 | * | |
2849 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2850 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2851 | * powerpc (64K page size) : = (32G +16M)byte. | |
2852 | */ | |
2853 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2854 | { | |
2855 | return 4096UL; | |
2856 | } | |
2857 | #endif | |
1da177e4 LT |
2858 | |
2859 | /* | |
2860 | * This is an integer logarithm so that shifts can be used later | |
2861 | * to extract the more random high bits from the multiplicative | |
2862 | * hash function before the remainder is taken. | |
2863 | */ | |
2864 | static inline unsigned long wait_table_bits(unsigned long size) | |
2865 | { | |
2866 | return ffz(~size); | |
2867 | } | |
2868 | ||
2869 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2870 | ||
56fd56b8 | 2871 | /* |
d9c23400 | 2872 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
2873 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
2874 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
2875 | * higher will lead to a bigger reserve which will get freed as contiguous |
2876 | * blocks as reclaim kicks in | |
2877 | */ | |
2878 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2879 | { | |
2880 | unsigned long start_pfn, pfn, end_pfn; | |
2881 | struct page *page; | |
78986a67 MG |
2882 | unsigned long block_migratetype; |
2883 | int reserve; | |
56fd56b8 MG |
2884 | |
2885 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2886 | start_pfn = zone->zone_start_pfn; | |
2887 | end_pfn = start_pfn + zone->spanned_pages; | |
41858966 | 2888 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 2889 | pageblock_order; |
56fd56b8 | 2890 | |
78986a67 MG |
2891 | /* |
2892 | * Reserve blocks are generally in place to help high-order atomic | |
2893 | * allocations that are short-lived. A min_free_kbytes value that | |
2894 | * would result in more than 2 reserve blocks for atomic allocations | |
2895 | * is assumed to be in place to help anti-fragmentation for the | |
2896 | * future allocation of hugepages at runtime. | |
2897 | */ | |
2898 | reserve = min(2, reserve); | |
2899 | ||
d9c23400 | 2900 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2901 | if (!pfn_valid(pfn)) |
2902 | continue; | |
2903 | page = pfn_to_page(pfn); | |
2904 | ||
344c790e AL |
2905 | /* Watch out for overlapping nodes */ |
2906 | if (page_to_nid(page) != zone_to_nid(zone)) | |
2907 | continue; | |
2908 | ||
56fd56b8 MG |
2909 | /* Blocks with reserved pages will never free, skip them. */ |
2910 | if (PageReserved(page)) | |
2911 | continue; | |
2912 | ||
2913 | block_migratetype = get_pageblock_migratetype(page); | |
2914 | ||
2915 | /* If this block is reserved, account for it */ | |
2916 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2917 | reserve--; | |
2918 | continue; | |
2919 | } | |
2920 | ||
2921 | /* Suitable for reserving if this block is movable */ | |
2922 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2923 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2924 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2925 | reserve--; | |
2926 | continue; | |
2927 | } | |
2928 | ||
2929 | /* | |
2930 | * If the reserve is met and this is a previous reserved block, | |
2931 | * take it back | |
2932 | */ | |
2933 | if (block_migratetype == MIGRATE_RESERVE) { | |
2934 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2935 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2936 | } | |
2937 | } | |
2938 | } | |
ac0e5b7a | 2939 | |
1da177e4 LT |
2940 | /* |
2941 | * Initially all pages are reserved - free ones are freed | |
2942 | * up by free_all_bootmem() once the early boot process is | |
2943 | * done. Non-atomic initialization, single-pass. | |
2944 | */ | |
c09b4240 | 2945 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2946 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2947 | { |
1da177e4 | 2948 | struct page *page; |
29751f69 AW |
2949 | unsigned long end_pfn = start_pfn + size; |
2950 | unsigned long pfn; | |
86051ca5 | 2951 | struct zone *z; |
1da177e4 | 2952 | |
22b31eec HD |
2953 | if (highest_memmap_pfn < end_pfn - 1) |
2954 | highest_memmap_pfn = end_pfn - 1; | |
2955 | ||
86051ca5 | 2956 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 2957 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2958 | /* |
2959 | * There can be holes in boot-time mem_map[]s | |
2960 | * handed to this function. They do not | |
2961 | * exist on hotplugged memory. | |
2962 | */ | |
2963 | if (context == MEMMAP_EARLY) { | |
2964 | if (!early_pfn_valid(pfn)) | |
2965 | continue; | |
2966 | if (!early_pfn_in_nid(pfn, nid)) | |
2967 | continue; | |
2968 | } | |
d41dee36 AW |
2969 | page = pfn_to_page(pfn); |
2970 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 2971 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 2972 | init_page_count(page); |
1da177e4 LT |
2973 | reset_page_mapcount(page); |
2974 | SetPageReserved(page); | |
b2a0ac88 MG |
2975 | /* |
2976 | * Mark the block movable so that blocks are reserved for | |
2977 | * movable at startup. This will force kernel allocations | |
2978 | * to reserve their blocks rather than leaking throughout | |
2979 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2980 | * kernel allocations are made. Later some blocks near |
2981 | * the start are marked MIGRATE_RESERVE by | |
2982 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
2983 | * |
2984 | * bitmap is created for zone's valid pfn range. but memmap | |
2985 | * can be created for invalid pages (for alignment) | |
2986 | * check here not to call set_pageblock_migratetype() against | |
2987 | * pfn out of zone. | |
b2a0ac88 | 2988 | */ |
86051ca5 KH |
2989 | if ((z->zone_start_pfn <= pfn) |
2990 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
2991 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 2992 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2993 | |
1da177e4 LT |
2994 | INIT_LIST_HEAD(&page->lru); |
2995 | #ifdef WANT_PAGE_VIRTUAL | |
2996 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2997 | if (!is_highmem_idx(zone)) | |
3212c6be | 2998 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2999 | #endif |
1da177e4 LT |
3000 | } |
3001 | } | |
3002 | ||
1e548deb | 3003 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3004 | { |
b2a0ac88 MG |
3005 | int order, t; |
3006 | for_each_migratetype_order(order, t) { | |
3007 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3008 | zone->free_area[order].nr_free = 0; |
3009 | } | |
3010 | } | |
3011 | ||
3012 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3013 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3014 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3015 | #endif |
3016 | ||
1d6f4e60 | 3017 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3018 | { |
3a6be87f | 3019 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3020 | int batch; |
3021 | ||
3022 | /* | |
3023 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3024 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3025 | * |
3026 | * OK, so we don't know how big the cache is. So guess. | |
3027 | */ | |
3028 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3029 | if (batch * PAGE_SIZE > 512 * 1024) |
3030 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3031 | batch /= 4; /* We effectively *= 4 below */ |
3032 | if (batch < 1) | |
3033 | batch = 1; | |
3034 | ||
3035 | /* | |
0ceaacc9 NP |
3036 | * Clamp the batch to a 2^n - 1 value. Having a power |
3037 | * of 2 value was found to be more likely to have | |
3038 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3039 | * |
0ceaacc9 NP |
3040 | * For example if 2 tasks are alternately allocating |
3041 | * batches of pages, one task can end up with a lot | |
3042 | * of pages of one half of the possible page colors | |
3043 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3044 | */ |
9155203a | 3045 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3046 | |
e7c8d5c9 | 3047 | return batch; |
3a6be87f DH |
3048 | |
3049 | #else | |
3050 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3051 | * conditions. | |
3052 | * | |
3053 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3054 | * of contiguous memory as there's no hardware page translation to | |
3055 | * assemble apparent contiguous memory from discontiguous pages. | |
3056 | * | |
3057 | * Queueing large contiguous runs of pages for batching, however, | |
3058 | * causes the pages to actually be freed in smaller chunks. As there | |
3059 | * can be a significant delay between the individual batches being | |
3060 | * recycled, this leads to the once large chunks of space being | |
3061 | * fragmented and becoming unavailable for high-order allocations. | |
3062 | */ | |
3063 | return 0; | |
3064 | #endif | |
e7c8d5c9 CL |
3065 | } |
3066 | ||
b69a7288 | 3067 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3068 | { |
3069 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3070 | int migratetype; |
2caaad41 | 3071 | |
1c6fe946 MD |
3072 | memset(p, 0, sizeof(*p)); |
3073 | ||
3dfa5721 | 3074 | pcp = &p->pcp; |
2caaad41 | 3075 | pcp->count = 0; |
2caaad41 CL |
3076 | pcp->high = 6 * batch; |
3077 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3078 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3079 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3080 | } |
3081 | ||
8ad4b1fb RS |
3082 | /* |
3083 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3084 | * to the value high for the pageset p. | |
3085 | */ | |
3086 | ||
3087 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3088 | unsigned long high) | |
3089 | { | |
3090 | struct per_cpu_pages *pcp; | |
3091 | ||
3dfa5721 | 3092 | pcp = &p->pcp; |
8ad4b1fb RS |
3093 | pcp->high = high; |
3094 | pcp->batch = max(1UL, high/4); | |
3095 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3096 | pcp->batch = PAGE_SHIFT * 8; | |
3097 | } | |
3098 | ||
3099 | ||
e7c8d5c9 CL |
3100 | #ifdef CONFIG_NUMA |
3101 | /* | |
2caaad41 CL |
3102 | * Boot pageset table. One per cpu which is going to be used for all |
3103 | * zones and all nodes. The parameters will be set in such a way | |
3104 | * that an item put on a list will immediately be handed over to | |
3105 | * the buddy list. This is safe since pageset manipulation is done | |
3106 | * with interrupts disabled. | |
3107 | * | |
3108 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
3109 | * |
3110 | * The boot_pagesets must be kept even after bootup is complete for | |
3111 | * unused processors and/or zones. They do play a role for bootstrapping | |
3112 | * hotplugged processors. | |
3113 | * | |
3114 | * zoneinfo_show() and maybe other functions do | |
3115 | * not check if the processor is online before following the pageset pointer. | |
3116 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 3117 | */ |
88a2a4ac | 3118 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
3119 | |
3120 | /* | |
3121 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
3122 | * per cpu pageset array in struct zone. |
3123 | */ | |
6292d9aa | 3124 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
3125 | { |
3126 | struct zone *zone, *dzone; | |
37c0708d CL |
3127 | int node = cpu_to_node(cpu); |
3128 | ||
3129 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 | 3130 | |
ee99c71c | 3131 | for_each_populated_zone(zone) { |
23316bc8 | 3132 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 3133 | GFP_KERNEL, node); |
23316bc8 | 3134 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 3135 | goto bad; |
e7c8d5c9 | 3136 | |
23316bc8 | 3137 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
3138 | |
3139 | if (percpu_pagelist_fraction) | |
3140 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
4365a567 | 3141 | (zone->present_pages / percpu_pagelist_fraction)); |
e7c8d5c9 CL |
3142 | } |
3143 | ||
3144 | return 0; | |
3145 | bad: | |
3146 | for_each_zone(dzone) { | |
64191688 AM |
3147 | if (!populated_zone(dzone)) |
3148 | continue; | |
e7c8d5c9 CL |
3149 | if (dzone == zone) |
3150 | break; | |
23316bc8 | 3151 | kfree(zone_pcp(dzone, cpu)); |
364df0eb | 3152 | zone_pcp(dzone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 CL |
3153 | } |
3154 | return -ENOMEM; | |
3155 | } | |
3156 | ||
3157 | static inline void free_zone_pagesets(int cpu) | |
3158 | { | |
e7c8d5c9 CL |
3159 | struct zone *zone; |
3160 | ||
3161 | for_each_zone(zone) { | |
3162 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
3163 | ||
f3ef9ead DR |
3164 | /* Free per_cpu_pageset if it is slab allocated */ |
3165 | if (pset != &boot_pageset[cpu]) | |
3166 | kfree(pset); | |
364df0eb | 3167 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 | 3168 | } |
e7c8d5c9 CL |
3169 | } |
3170 | ||
9c7b216d | 3171 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
3172 | unsigned long action, |
3173 | void *hcpu) | |
3174 | { | |
3175 | int cpu = (long)hcpu; | |
3176 | int ret = NOTIFY_OK; | |
3177 | ||
3178 | switch (action) { | |
ce421c79 | 3179 | case CPU_UP_PREPARE: |
8bb78442 | 3180 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
3181 | if (process_zones(cpu)) |
3182 | ret = NOTIFY_BAD; | |
3183 | break; | |
3184 | case CPU_UP_CANCELED: | |
8bb78442 | 3185 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 3186 | case CPU_DEAD: |
8bb78442 | 3187 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
3188 | free_zone_pagesets(cpu); |
3189 | break; | |
3190 | default: | |
3191 | break; | |
e7c8d5c9 CL |
3192 | } |
3193 | return ret; | |
3194 | } | |
3195 | ||
74b85f37 | 3196 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
3197 | { &pageset_cpuup_callback, NULL, 0 }; |
3198 | ||
78d9955b | 3199 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
3200 | { |
3201 | int err; | |
3202 | ||
3203 | /* Initialize per_cpu_pageset for cpu 0. | |
3204 | * A cpuup callback will do this for every cpu | |
3205 | * as it comes online | |
3206 | */ | |
3207 | err = process_zones(smp_processor_id()); | |
3208 | BUG_ON(err); | |
3209 | register_cpu_notifier(&pageset_notifier); | |
3210 | } | |
3211 | ||
3212 | #endif | |
3213 | ||
577a32f6 | 3214 | static noinline __init_refok |
cca448fe | 3215 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
3216 | { |
3217 | int i; | |
3218 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 3219 | size_t alloc_size; |
ed8ece2e DH |
3220 | |
3221 | /* | |
3222 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3223 | * per zone. | |
3224 | */ | |
02b694de YG |
3225 | zone->wait_table_hash_nr_entries = |
3226 | wait_table_hash_nr_entries(zone_size_pages); | |
3227 | zone->wait_table_bits = | |
3228 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
3229 | alloc_size = zone->wait_table_hash_nr_entries |
3230 | * sizeof(wait_queue_head_t); | |
3231 | ||
cd94b9db | 3232 | if (!slab_is_available()) { |
cca448fe YG |
3233 | zone->wait_table = (wait_queue_head_t *) |
3234 | alloc_bootmem_node(pgdat, alloc_size); | |
3235 | } else { | |
3236 | /* | |
3237 | * This case means that a zone whose size was 0 gets new memory | |
3238 | * via memory hot-add. | |
3239 | * But it may be the case that a new node was hot-added. In | |
3240 | * this case vmalloc() will not be able to use this new node's | |
3241 | * memory - this wait_table must be initialized to use this new | |
3242 | * node itself as well. | |
3243 | * To use this new node's memory, further consideration will be | |
3244 | * necessary. | |
3245 | */ | |
8691f3a7 | 3246 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3247 | } |
3248 | if (!zone->wait_table) | |
3249 | return -ENOMEM; | |
ed8ece2e | 3250 | |
02b694de | 3251 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3252 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3253 | |
3254 | return 0; | |
ed8ece2e DH |
3255 | } |
3256 | ||
112067f0 SL |
3257 | static int __zone_pcp_update(void *data) |
3258 | { | |
3259 | struct zone *zone = data; | |
3260 | int cpu; | |
3261 | unsigned long batch = zone_batchsize(zone), flags; | |
3262 | ||
3263 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3264 | struct per_cpu_pageset *pset; | |
3265 | struct per_cpu_pages *pcp; | |
3266 | ||
3267 | pset = zone_pcp(zone, cpu); | |
3268 | pcp = &pset->pcp; | |
3269 | ||
3270 | local_irq_save(flags); | |
5f8dcc21 | 3271 | free_pcppages_bulk(zone, pcp->count, pcp); |
112067f0 SL |
3272 | setup_pageset(pset, batch); |
3273 | local_irq_restore(flags); | |
3274 | } | |
3275 | return 0; | |
3276 | } | |
3277 | ||
3278 | void zone_pcp_update(struct zone *zone) | |
3279 | { | |
3280 | stop_machine(__zone_pcp_update, zone, NULL); | |
3281 | } | |
3282 | ||
c09b4240 | 3283 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
3284 | { |
3285 | int cpu; | |
3286 | unsigned long batch = zone_batchsize(zone); | |
3287 | ||
3288 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3289 | #ifdef CONFIG_NUMA | |
3290 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 3291 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
3292 | setup_pageset(&boot_pageset[cpu],0); |
3293 | #else | |
3294 | setup_pageset(zone_pcp(zone,cpu), batch); | |
3295 | #endif | |
3296 | } | |
f5335c0f AB |
3297 | if (zone->present_pages) |
3298 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
3299 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
3300 | } |
3301 | ||
718127cc YG |
3302 | __meminit int init_currently_empty_zone(struct zone *zone, |
3303 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3304 | unsigned long size, |
3305 | enum memmap_context context) | |
ed8ece2e DH |
3306 | { |
3307 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3308 | int ret; |
3309 | ret = zone_wait_table_init(zone, size); | |
3310 | if (ret) | |
3311 | return ret; | |
ed8ece2e DH |
3312 | pgdat->nr_zones = zone_idx(zone) + 1; |
3313 | ||
ed8ece2e DH |
3314 | zone->zone_start_pfn = zone_start_pfn; |
3315 | ||
708614e6 MG |
3316 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3317 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3318 | pgdat->node_id, | |
3319 | (unsigned long)zone_idx(zone), | |
3320 | zone_start_pfn, (zone_start_pfn + size)); | |
3321 | ||
1e548deb | 3322 | zone_init_free_lists(zone); |
718127cc YG |
3323 | |
3324 | return 0; | |
ed8ece2e DH |
3325 | } |
3326 | ||
c713216d MG |
3327 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3328 | /* | |
3329 | * Basic iterator support. Return the first range of PFNs for a node | |
3330 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3331 | */ | |
a3142c8e | 3332 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
3333 | { |
3334 | int i; | |
3335 | ||
3336 | for (i = 0; i < nr_nodemap_entries; i++) | |
3337 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3338 | return i; | |
3339 | ||
3340 | return -1; | |
3341 | } | |
3342 | ||
3343 | /* | |
3344 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 3345 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 3346 | */ |
a3142c8e | 3347 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
3348 | { |
3349 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3350 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3351 | return index; | |
3352 | ||
3353 | return -1; | |
3354 | } | |
3355 | ||
3356 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3357 | /* | |
3358 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3359 | * Architectures may implement their own version but if add_active_range() | |
3360 | * was used and there are no special requirements, this is a convenient | |
3361 | * alternative | |
3362 | */ | |
f2dbcfa7 | 3363 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
3364 | { |
3365 | int i; | |
3366 | ||
3367 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3368 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3369 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3370 | ||
3371 | if (start_pfn <= pfn && pfn < end_pfn) | |
3372 | return early_node_map[i].nid; | |
3373 | } | |
cc2559bc KH |
3374 | /* This is a memory hole */ |
3375 | return -1; | |
c713216d MG |
3376 | } |
3377 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3378 | ||
f2dbcfa7 KH |
3379 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3380 | { | |
cc2559bc KH |
3381 | int nid; |
3382 | ||
3383 | nid = __early_pfn_to_nid(pfn); | |
3384 | if (nid >= 0) | |
3385 | return nid; | |
3386 | /* just returns 0 */ | |
3387 | return 0; | |
f2dbcfa7 KH |
3388 | } |
3389 | ||
cc2559bc KH |
3390 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3391 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3392 | { | |
3393 | int nid; | |
3394 | ||
3395 | nid = __early_pfn_to_nid(pfn); | |
3396 | if (nid >= 0 && nid != node) | |
3397 | return false; | |
3398 | return true; | |
3399 | } | |
3400 | #endif | |
f2dbcfa7 | 3401 | |
c713216d MG |
3402 | /* Basic iterator support to walk early_node_map[] */ |
3403 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3404 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3405 | i = next_active_region_index_in_nid(i, nid)) | |
3406 | ||
3407 | /** | |
3408 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3409 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3410 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3411 | * |
3412 | * If an architecture guarantees that all ranges registered with | |
3413 | * add_active_ranges() contain no holes and may be freed, this | |
3414 | * this function may be used instead of calling free_bootmem() manually. | |
3415 | */ | |
3416 | void __init free_bootmem_with_active_regions(int nid, | |
3417 | unsigned long max_low_pfn) | |
3418 | { | |
3419 | int i; | |
3420 | ||
3421 | for_each_active_range_index_in_nid(i, nid) { | |
3422 | unsigned long size_pages = 0; | |
3423 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3424 | ||
3425 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3426 | continue; | |
3427 | ||
3428 | if (end_pfn > max_low_pfn) | |
3429 | end_pfn = max_low_pfn; | |
3430 | ||
3431 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3432 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3433 | PFN_PHYS(early_node_map[i].start_pfn), | |
3434 | size_pages << PAGE_SHIFT); | |
3435 | } | |
3436 | } | |
3437 | ||
b5bc6c0e YL |
3438 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
3439 | { | |
3440 | int i; | |
d52d53b8 | 3441 | int ret; |
b5bc6c0e | 3442 | |
d52d53b8 YL |
3443 | for_each_active_range_index_in_nid(i, nid) { |
3444 | ret = work_fn(early_node_map[i].start_pfn, | |
3445 | early_node_map[i].end_pfn, data); | |
3446 | if (ret) | |
3447 | break; | |
3448 | } | |
b5bc6c0e | 3449 | } |
c713216d MG |
3450 | /** |
3451 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3452 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3453 | * |
3454 | * If an architecture guarantees that all ranges registered with | |
3455 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3456 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3457 | */ |
3458 | void __init sparse_memory_present_with_active_regions(int nid) | |
3459 | { | |
3460 | int i; | |
3461 | ||
3462 | for_each_active_range_index_in_nid(i, nid) | |
3463 | memory_present(early_node_map[i].nid, | |
3464 | early_node_map[i].start_pfn, | |
3465 | early_node_map[i].end_pfn); | |
3466 | } | |
3467 | ||
3468 | /** | |
3469 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3470 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3471 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3472 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3473 | * |
3474 | * It returns the start and end page frame of a node based on information | |
3475 | * provided by an arch calling add_active_range(). If called for a node | |
3476 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3477 | * PFNs will be 0. |
c713216d | 3478 | */ |
a3142c8e | 3479 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3480 | unsigned long *start_pfn, unsigned long *end_pfn) |
3481 | { | |
3482 | int i; | |
3483 | *start_pfn = -1UL; | |
3484 | *end_pfn = 0; | |
3485 | ||
3486 | for_each_active_range_index_in_nid(i, nid) { | |
3487 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3488 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3489 | } | |
3490 | ||
633c0666 | 3491 | if (*start_pfn == -1UL) |
c713216d | 3492 | *start_pfn = 0; |
c713216d MG |
3493 | } |
3494 | ||
2a1e274a MG |
3495 | /* |
3496 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3497 | * assumption is made that zones within a node are ordered in monotonic | |
3498 | * increasing memory addresses so that the "highest" populated zone is used | |
3499 | */ | |
b69a7288 | 3500 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3501 | { |
3502 | int zone_index; | |
3503 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3504 | if (zone_index == ZONE_MOVABLE) | |
3505 | continue; | |
3506 | ||
3507 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3508 | arch_zone_lowest_possible_pfn[zone_index]) | |
3509 | break; | |
3510 | } | |
3511 | ||
3512 | VM_BUG_ON(zone_index == -1); | |
3513 | movable_zone = zone_index; | |
3514 | } | |
3515 | ||
3516 | /* | |
3517 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3518 | * because it is sized independant of architecture. Unlike the other zones, | |
3519 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3520 | * in each node depending on the size of each node and how evenly kernelcore | |
3521 | * is distributed. This helper function adjusts the zone ranges | |
3522 | * provided by the architecture for a given node by using the end of the | |
3523 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3524 | * zones within a node are in order of monotonic increases memory addresses | |
3525 | */ | |
b69a7288 | 3526 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
3527 | unsigned long zone_type, |
3528 | unsigned long node_start_pfn, | |
3529 | unsigned long node_end_pfn, | |
3530 | unsigned long *zone_start_pfn, | |
3531 | unsigned long *zone_end_pfn) | |
3532 | { | |
3533 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3534 | if (zone_movable_pfn[nid]) { | |
3535 | /* Size ZONE_MOVABLE */ | |
3536 | if (zone_type == ZONE_MOVABLE) { | |
3537 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3538 | *zone_end_pfn = min(node_end_pfn, | |
3539 | arch_zone_highest_possible_pfn[movable_zone]); | |
3540 | ||
3541 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3542 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3543 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3544 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3545 | ||
3546 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3547 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3548 | *zone_start_pfn = *zone_end_pfn; | |
3549 | } | |
3550 | } | |
3551 | ||
c713216d MG |
3552 | /* |
3553 | * Return the number of pages a zone spans in a node, including holes | |
3554 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3555 | */ | |
6ea6e688 | 3556 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3557 | unsigned long zone_type, |
3558 | unsigned long *ignored) | |
3559 | { | |
3560 | unsigned long node_start_pfn, node_end_pfn; | |
3561 | unsigned long zone_start_pfn, zone_end_pfn; | |
3562 | ||
3563 | /* Get the start and end of the node and zone */ | |
3564 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3565 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3566 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3567 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3568 | node_start_pfn, node_end_pfn, | |
3569 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3570 | |
3571 | /* Check that this node has pages within the zone's required range */ | |
3572 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3573 | return 0; | |
3574 | ||
3575 | /* Move the zone boundaries inside the node if necessary */ | |
3576 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3577 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3578 | ||
3579 | /* Return the spanned pages */ | |
3580 | return zone_end_pfn - zone_start_pfn; | |
3581 | } | |
3582 | ||
3583 | /* | |
3584 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3585 | * then all holes in the requested range will be accounted for. |
c713216d | 3586 | */ |
32996250 | 3587 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3588 | unsigned long range_start_pfn, |
3589 | unsigned long range_end_pfn) | |
3590 | { | |
3591 | int i = 0; | |
3592 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3593 | unsigned long start_pfn; | |
3594 | ||
3595 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3596 | i = first_active_region_index_in_nid(nid); | |
3597 | if (i == -1) | |
3598 | return 0; | |
3599 | ||
b5445f95 MG |
3600 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3601 | ||
9c7cd687 MG |
3602 | /* Account for ranges before physical memory on this node */ |
3603 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3604 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3605 | |
3606 | /* Find all holes for the zone within the node */ | |
3607 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3608 | ||
3609 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3610 | if (prev_end_pfn >= range_end_pfn) | |
3611 | break; | |
3612 | ||
3613 | /* Make sure the end of the zone is not within the hole */ | |
3614 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3615 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3616 | ||
3617 | /* Update the hole size cound and move on */ | |
3618 | if (start_pfn > range_start_pfn) { | |
3619 | BUG_ON(prev_end_pfn > start_pfn); | |
3620 | hole_pages += start_pfn - prev_end_pfn; | |
3621 | } | |
3622 | prev_end_pfn = early_node_map[i].end_pfn; | |
3623 | } | |
3624 | ||
9c7cd687 MG |
3625 | /* Account for ranges past physical memory on this node */ |
3626 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3627 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3628 | max(range_start_pfn, prev_end_pfn); |
3629 | ||
c713216d MG |
3630 | return hole_pages; |
3631 | } | |
3632 | ||
3633 | /** | |
3634 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3635 | * @start_pfn: The start PFN to start searching for holes | |
3636 | * @end_pfn: The end PFN to stop searching for holes | |
3637 | * | |
88ca3b94 | 3638 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3639 | */ |
3640 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3641 | unsigned long end_pfn) | |
3642 | { | |
3643 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3644 | } | |
3645 | ||
3646 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3647 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3648 | unsigned long zone_type, |
3649 | unsigned long *ignored) | |
3650 | { | |
9c7cd687 MG |
3651 | unsigned long node_start_pfn, node_end_pfn; |
3652 | unsigned long zone_start_pfn, zone_end_pfn; | |
3653 | ||
3654 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3655 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3656 | node_start_pfn); | |
3657 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3658 | node_end_pfn); | |
3659 | ||
2a1e274a MG |
3660 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3661 | node_start_pfn, node_end_pfn, | |
3662 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3663 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3664 | } |
0e0b864e | 3665 | |
c713216d | 3666 | #else |
6ea6e688 | 3667 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3668 | unsigned long zone_type, |
3669 | unsigned long *zones_size) | |
3670 | { | |
3671 | return zones_size[zone_type]; | |
3672 | } | |
3673 | ||
6ea6e688 | 3674 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3675 | unsigned long zone_type, |
3676 | unsigned long *zholes_size) | |
3677 | { | |
3678 | if (!zholes_size) | |
3679 | return 0; | |
3680 | ||
3681 | return zholes_size[zone_type]; | |
3682 | } | |
0e0b864e | 3683 | |
c713216d MG |
3684 | #endif |
3685 | ||
a3142c8e | 3686 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3687 | unsigned long *zones_size, unsigned long *zholes_size) |
3688 | { | |
3689 | unsigned long realtotalpages, totalpages = 0; | |
3690 | enum zone_type i; | |
3691 | ||
3692 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3693 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3694 | zones_size); | |
3695 | pgdat->node_spanned_pages = totalpages; | |
3696 | ||
3697 | realtotalpages = totalpages; | |
3698 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3699 | realtotalpages -= | |
3700 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3701 | zholes_size); | |
3702 | pgdat->node_present_pages = realtotalpages; | |
3703 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3704 | realtotalpages); | |
3705 | } | |
3706 | ||
835c134e MG |
3707 | #ifndef CONFIG_SPARSEMEM |
3708 | /* | |
3709 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3710 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3711 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3712 | * round what is now in bits to nearest long in bits, then return it in |
3713 | * bytes. | |
3714 | */ | |
3715 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3716 | { | |
3717 | unsigned long usemapsize; | |
3718 | ||
d9c23400 MG |
3719 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3720 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3721 | usemapsize *= NR_PAGEBLOCK_BITS; |
3722 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3723 | ||
3724 | return usemapsize / 8; | |
3725 | } | |
3726 | ||
3727 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3728 | struct zone *zone, unsigned long zonesize) | |
3729 | { | |
3730 | unsigned long usemapsize = usemap_size(zonesize); | |
3731 | zone->pageblock_flags = NULL; | |
58a01a45 | 3732 | if (usemapsize) |
835c134e | 3733 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
835c134e MG |
3734 | } |
3735 | #else | |
3736 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3737 | struct zone *zone, unsigned long zonesize) {} | |
3738 | #endif /* CONFIG_SPARSEMEM */ | |
3739 | ||
d9c23400 | 3740 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3741 | |
3742 | /* Return a sensible default order for the pageblock size. */ | |
3743 | static inline int pageblock_default_order(void) | |
3744 | { | |
3745 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3746 | return HUGETLB_PAGE_ORDER; | |
3747 | ||
3748 | return MAX_ORDER-1; | |
3749 | } | |
3750 | ||
d9c23400 MG |
3751 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3752 | static inline void __init set_pageblock_order(unsigned int order) | |
3753 | { | |
3754 | /* Check that pageblock_nr_pages has not already been setup */ | |
3755 | if (pageblock_order) | |
3756 | return; | |
3757 | ||
3758 | /* | |
3759 | * Assume the largest contiguous order of interest is a huge page. | |
3760 | * This value may be variable depending on boot parameters on IA64 | |
3761 | */ | |
3762 | pageblock_order = order; | |
3763 | } | |
3764 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3765 | ||
ba72cb8c MG |
3766 | /* |
3767 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3768 | * and pageblock_default_order() are unused as pageblock_order is set | |
3769 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3770 | * pageblock_order based on the kernel config | |
3771 | */ | |
3772 | static inline int pageblock_default_order(unsigned int order) | |
3773 | { | |
3774 | return MAX_ORDER-1; | |
3775 | } | |
d9c23400 MG |
3776 | #define set_pageblock_order(x) do {} while (0) |
3777 | ||
3778 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3779 | ||
1da177e4 LT |
3780 | /* |
3781 | * Set up the zone data structures: | |
3782 | * - mark all pages reserved | |
3783 | * - mark all memory queues empty | |
3784 | * - clear the memory bitmaps | |
3785 | */ | |
b5a0e011 | 3786 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3787 | unsigned long *zones_size, unsigned long *zholes_size) |
3788 | { | |
2f1b6248 | 3789 | enum zone_type j; |
ed8ece2e | 3790 | int nid = pgdat->node_id; |
1da177e4 | 3791 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3792 | int ret; |
1da177e4 | 3793 | |
208d54e5 | 3794 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3795 | pgdat->nr_zones = 0; |
3796 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3797 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 3798 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
3799 | |
3800 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3801 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3802 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 3803 | enum lru_list l; |
1da177e4 | 3804 | |
c713216d MG |
3805 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3806 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3807 | zholes_size); | |
1da177e4 | 3808 | |
0e0b864e MG |
3809 | /* |
3810 | * Adjust realsize so that it accounts for how much memory | |
3811 | * is used by this zone for memmap. This affects the watermark | |
3812 | * and per-cpu initialisations | |
3813 | */ | |
f7232154 JW |
3814 | memmap_pages = |
3815 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
3816 | if (realsize >= memmap_pages) { |
3817 | realsize -= memmap_pages; | |
5594c8c8 YL |
3818 | if (memmap_pages) |
3819 | printk(KERN_DEBUG | |
3820 | " %s zone: %lu pages used for memmap\n", | |
3821 | zone_names[j], memmap_pages); | |
0e0b864e MG |
3822 | } else |
3823 | printk(KERN_WARNING | |
3824 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3825 | zone_names[j], memmap_pages, realsize); | |
3826 | ||
6267276f CL |
3827 | /* Account for reserved pages */ |
3828 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3829 | realsize -= dma_reserve; |
d903ef9f | 3830 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 3831 | zone_names[0], dma_reserve); |
0e0b864e MG |
3832 | } |
3833 | ||
98d2b0eb | 3834 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3835 | nr_kernel_pages += realsize; |
3836 | nr_all_pages += realsize; | |
3837 | ||
3838 | zone->spanned_pages = size; | |
3839 | zone->present_pages = realsize; | |
9614634f | 3840 | #ifdef CONFIG_NUMA |
d5f541ed | 3841 | zone->node = nid; |
8417bba4 | 3842 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3843 | / 100; |
0ff38490 | 3844 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3845 | #endif |
1da177e4 LT |
3846 | zone->name = zone_names[j]; |
3847 | spin_lock_init(&zone->lock); | |
3848 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3849 | zone_seqlock_init(zone); |
1da177e4 | 3850 | zone->zone_pgdat = pgdat; |
1da177e4 | 3851 | |
3bb1a852 | 3852 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3853 | |
ed8ece2e | 3854 | zone_pcp_init(zone); |
b69408e8 CL |
3855 | for_each_lru(l) { |
3856 | INIT_LIST_HEAD(&zone->lru[l].list); | |
f8629631 | 3857 | zone->reclaim_stat.nr_saved_scan[l] = 0; |
b69408e8 | 3858 | } |
6e901571 KM |
3859 | zone->reclaim_stat.recent_rotated[0] = 0; |
3860 | zone->reclaim_stat.recent_rotated[1] = 0; | |
3861 | zone->reclaim_stat.recent_scanned[0] = 0; | |
3862 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 3863 | zap_zone_vm_stats(zone); |
e815af95 | 3864 | zone->flags = 0; |
1da177e4 LT |
3865 | if (!size) |
3866 | continue; | |
3867 | ||
ba72cb8c | 3868 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3869 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3870 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3871 | size, MEMMAP_EARLY); | |
718127cc | 3872 | BUG_ON(ret); |
76cdd58e | 3873 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 3874 | zone_start_pfn += size; |
1da177e4 LT |
3875 | } |
3876 | } | |
3877 | ||
577a32f6 | 3878 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3879 | { |
1da177e4 LT |
3880 | /* Skip empty nodes */ |
3881 | if (!pgdat->node_spanned_pages) | |
3882 | return; | |
3883 | ||
d41dee36 | 3884 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3885 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3886 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3887 | unsigned long size, start, end; |
d41dee36 AW |
3888 | struct page *map; |
3889 | ||
e984bb43 BP |
3890 | /* |
3891 | * The zone's endpoints aren't required to be MAX_ORDER | |
3892 | * aligned but the node_mem_map endpoints must be in order | |
3893 | * for the buddy allocator to function correctly. | |
3894 | */ | |
3895 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3896 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3897 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3898 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3899 | map = alloc_remap(pgdat->node_id, size); |
3900 | if (!map) | |
3901 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3902 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3903 | } |
12d810c1 | 3904 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3905 | /* |
3906 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3907 | */ | |
c713216d | 3908 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3909 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3910 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3911 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3912 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3913 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3914 | } | |
1da177e4 | 3915 | #endif |
d41dee36 | 3916 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3917 | } |
3918 | ||
9109fb7b JW |
3919 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
3920 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 3921 | { |
9109fb7b JW |
3922 | pg_data_t *pgdat = NODE_DATA(nid); |
3923 | ||
1da177e4 LT |
3924 | pgdat->node_id = nid; |
3925 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3926 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3927 | |
3928 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
3929 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
3930 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3931 | nid, (unsigned long)pgdat, | |
3932 | (unsigned long)pgdat->node_mem_map); | |
3933 | #endif | |
1da177e4 LT |
3934 | |
3935 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3936 | } | |
3937 | ||
c713216d | 3938 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3939 | |
3940 | #if MAX_NUMNODES > 1 | |
3941 | /* | |
3942 | * Figure out the number of possible node ids. | |
3943 | */ | |
3944 | static void __init setup_nr_node_ids(void) | |
3945 | { | |
3946 | unsigned int node; | |
3947 | unsigned int highest = 0; | |
3948 | ||
3949 | for_each_node_mask(node, node_possible_map) | |
3950 | highest = node; | |
3951 | nr_node_ids = highest + 1; | |
3952 | } | |
3953 | #else | |
3954 | static inline void setup_nr_node_ids(void) | |
3955 | { | |
3956 | } | |
3957 | #endif | |
3958 | ||
c713216d MG |
3959 | /** |
3960 | * add_active_range - Register a range of PFNs backed by physical memory | |
3961 | * @nid: The node ID the range resides on | |
3962 | * @start_pfn: The start PFN of the available physical memory | |
3963 | * @end_pfn: The end PFN of the available physical memory | |
3964 | * | |
3965 | * These ranges are stored in an early_node_map[] and later used by | |
3966 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3967 | * range spans a memory hole, it is up to the architecture to ensure | |
3968 | * the memory is not freed by the bootmem allocator. If possible | |
3969 | * the range being registered will be merged with existing ranges. | |
3970 | */ | |
3971 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3972 | unsigned long end_pfn) | |
3973 | { | |
3974 | int i; | |
3975 | ||
6b74ab97 MG |
3976 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
3977 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3978 | "%d entries of %d used\n", | |
3979 | nid, start_pfn, end_pfn, | |
3980 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 3981 | |
2dbb51c4 MG |
3982 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
3983 | ||
c713216d MG |
3984 | /* Merge with existing active regions if possible */ |
3985 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3986 | if (early_node_map[i].nid != nid) | |
3987 | continue; | |
3988 | ||
3989 | /* Skip if an existing region covers this new one */ | |
3990 | if (start_pfn >= early_node_map[i].start_pfn && | |
3991 | end_pfn <= early_node_map[i].end_pfn) | |
3992 | return; | |
3993 | ||
3994 | /* Merge forward if suitable */ | |
3995 | if (start_pfn <= early_node_map[i].end_pfn && | |
3996 | end_pfn > early_node_map[i].end_pfn) { | |
3997 | early_node_map[i].end_pfn = end_pfn; | |
3998 | return; | |
3999 | } | |
4000 | ||
4001 | /* Merge backward if suitable */ | |
d2dbe08d | 4002 | if (start_pfn < early_node_map[i].start_pfn && |
c713216d MG |
4003 | end_pfn >= early_node_map[i].start_pfn) { |
4004 | early_node_map[i].start_pfn = start_pfn; | |
4005 | return; | |
4006 | } | |
4007 | } | |
4008 | ||
4009 | /* Check that early_node_map is large enough */ | |
4010 | if (i >= MAX_ACTIVE_REGIONS) { | |
4011 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
4012 | MAX_ACTIVE_REGIONS); | |
4013 | return; | |
4014 | } | |
4015 | ||
4016 | early_node_map[i].nid = nid; | |
4017 | early_node_map[i].start_pfn = start_pfn; | |
4018 | early_node_map[i].end_pfn = end_pfn; | |
4019 | nr_nodemap_entries = i + 1; | |
4020 | } | |
4021 | ||
4022 | /** | |
cc1050ba | 4023 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 4024 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
4025 | * @start_pfn: The new PFN of the range |
4026 | * @end_pfn: The new PFN of the range | |
c713216d MG |
4027 | * |
4028 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
4029 | * The map is kept near the end physical page range that has already been |
4030 | * registered. This function allows an arch to shrink an existing registered | |
4031 | * range. | |
c713216d | 4032 | */ |
cc1050ba YL |
4033 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
4034 | unsigned long end_pfn) | |
c713216d | 4035 | { |
cc1a9d86 YL |
4036 | int i, j; |
4037 | int removed = 0; | |
c713216d | 4038 | |
cc1050ba YL |
4039 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
4040 | nid, start_pfn, end_pfn); | |
4041 | ||
c713216d | 4042 | /* Find the old active region end and shrink */ |
cc1a9d86 | 4043 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
4044 | if (early_node_map[i].start_pfn >= start_pfn && |
4045 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 4046 | /* clear it */ |
cc1050ba | 4047 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
4048 | early_node_map[i].end_pfn = 0; |
4049 | removed = 1; | |
4050 | continue; | |
4051 | } | |
cc1050ba YL |
4052 | if (early_node_map[i].start_pfn < start_pfn && |
4053 | early_node_map[i].end_pfn > start_pfn) { | |
4054 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
4055 | early_node_map[i].end_pfn = start_pfn; | |
4056 | if (temp_end_pfn > end_pfn) | |
4057 | add_active_range(nid, end_pfn, temp_end_pfn); | |
4058 | continue; | |
4059 | } | |
4060 | if (early_node_map[i].start_pfn >= start_pfn && | |
4061 | early_node_map[i].end_pfn > end_pfn && | |
4062 | early_node_map[i].start_pfn < end_pfn) { | |
4063 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 4064 | continue; |
c713216d | 4065 | } |
cc1a9d86 YL |
4066 | } |
4067 | ||
4068 | if (!removed) | |
4069 | return; | |
4070 | ||
4071 | /* remove the blank ones */ | |
4072 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
4073 | if (early_node_map[i].nid != nid) | |
4074 | continue; | |
4075 | if (early_node_map[i].end_pfn) | |
4076 | continue; | |
4077 | /* we found it, get rid of it */ | |
4078 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
4079 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
4080 | sizeof(early_node_map[j])); | |
4081 | j = nr_nodemap_entries - 1; | |
4082 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
4083 | nr_nodemap_entries--; | |
4084 | } | |
c713216d MG |
4085 | } |
4086 | ||
4087 | /** | |
4088 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 4089 | * |
c713216d MG |
4090 | * During discovery, it may be found that a table like SRAT is invalid |
4091 | * and an alternative discovery method must be used. This function removes | |
4092 | * all currently registered regions. | |
4093 | */ | |
88ca3b94 | 4094 | void __init remove_all_active_ranges(void) |
c713216d MG |
4095 | { |
4096 | memset(early_node_map, 0, sizeof(early_node_map)); | |
4097 | nr_nodemap_entries = 0; | |
4098 | } | |
4099 | ||
4100 | /* Compare two active node_active_regions */ | |
4101 | static int __init cmp_node_active_region(const void *a, const void *b) | |
4102 | { | |
4103 | struct node_active_region *arange = (struct node_active_region *)a; | |
4104 | struct node_active_region *brange = (struct node_active_region *)b; | |
4105 | ||
4106 | /* Done this way to avoid overflows */ | |
4107 | if (arange->start_pfn > brange->start_pfn) | |
4108 | return 1; | |
4109 | if (arange->start_pfn < brange->start_pfn) | |
4110 | return -1; | |
4111 | ||
4112 | return 0; | |
4113 | } | |
4114 | ||
4115 | /* sort the node_map by start_pfn */ | |
32996250 | 4116 | void __init sort_node_map(void) |
c713216d MG |
4117 | { |
4118 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
4119 | sizeof(struct node_active_region), | |
4120 | cmp_node_active_region, NULL); | |
4121 | } | |
4122 | ||
a6af2bc3 | 4123 | /* Find the lowest pfn for a node */ |
b69a7288 | 4124 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
4125 | { |
4126 | int i; | |
a6af2bc3 | 4127 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 4128 | |
c713216d MG |
4129 | /* Assuming a sorted map, the first range found has the starting pfn */ |
4130 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 4131 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 4132 | |
a6af2bc3 MG |
4133 | if (min_pfn == ULONG_MAX) { |
4134 | printk(KERN_WARNING | |
2bc0d261 | 4135 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4136 | return 0; |
4137 | } | |
4138 | ||
4139 | return min_pfn; | |
c713216d MG |
4140 | } |
4141 | ||
4142 | /** | |
4143 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4144 | * | |
4145 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4146 | * add_active_range(). |
c713216d MG |
4147 | */ |
4148 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4149 | { | |
4150 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4151 | } | |
4152 | ||
37b07e41 LS |
4153 | /* |
4154 | * early_calculate_totalpages() | |
4155 | * Sum pages in active regions for movable zone. | |
4156 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4157 | */ | |
484f51f8 | 4158 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
4159 | { |
4160 | int i; | |
4161 | unsigned long totalpages = 0; | |
4162 | ||
37b07e41 LS |
4163 | for (i = 0; i < nr_nodemap_entries; i++) { |
4164 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 4165 | early_node_map[i].start_pfn; |
37b07e41 LS |
4166 | totalpages += pages; |
4167 | if (pages) | |
4168 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
4169 | } | |
4170 | return totalpages; | |
7e63efef MG |
4171 | } |
4172 | ||
2a1e274a MG |
4173 | /* |
4174 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4175 | * is spread evenly between nodes as long as the nodes have enough | |
4176 | * memory. When they don't, some nodes will have more kernelcore than | |
4177 | * others | |
4178 | */ | |
b69a7288 | 4179 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
4180 | { |
4181 | int i, nid; | |
4182 | unsigned long usable_startpfn; | |
4183 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4184 | /* save the state before borrow the nodemask */ |
4185 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4186 | unsigned long totalpages = early_calculate_totalpages(); |
4187 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4188 | |
7e63efef MG |
4189 | /* |
4190 | * If movablecore was specified, calculate what size of | |
4191 | * kernelcore that corresponds so that memory usable for | |
4192 | * any allocation type is evenly spread. If both kernelcore | |
4193 | * and movablecore are specified, then the value of kernelcore | |
4194 | * will be used for required_kernelcore if it's greater than | |
4195 | * what movablecore would have allowed. | |
4196 | */ | |
4197 | if (required_movablecore) { | |
7e63efef MG |
4198 | unsigned long corepages; |
4199 | ||
4200 | /* | |
4201 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4202 | * was requested by the user | |
4203 | */ | |
4204 | required_movablecore = | |
4205 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4206 | corepages = totalpages - required_movablecore; | |
4207 | ||
4208 | required_kernelcore = max(required_kernelcore, corepages); | |
4209 | } | |
4210 | ||
2a1e274a MG |
4211 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4212 | if (!required_kernelcore) | |
66918dcd | 4213 | goto out; |
2a1e274a MG |
4214 | |
4215 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4216 | find_usable_zone_for_movable(); | |
4217 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4218 | ||
4219 | restart: | |
4220 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4221 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4222 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
4223 | /* |
4224 | * Recalculate kernelcore_node if the division per node | |
4225 | * now exceeds what is necessary to satisfy the requested | |
4226 | * amount of memory for the kernel | |
4227 | */ | |
4228 | if (required_kernelcore < kernelcore_node) | |
4229 | kernelcore_node = required_kernelcore / usable_nodes; | |
4230 | ||
4231 | /* | |
4232 | * As the map is walked, we track how much memory is usable | |
4233 | * by the kernel using kernelcore_remaining. When it is | |
4234 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4235 | */ | |
4236 | kernelcore_remaining = kernelcore_node; | |
4237 | ||
4238 | /* Go through each range of PFNs within this node */ | |
4239 | for_each_active_range_index_in_nid(i, nid) { | |
4240 | unsigned long start_pfn, end_pfn; | |
4241 | unsigned long size_pages; | |
4242 | ||
4243 | start_pfn = max(early_node_map[i].start_pfn, | |
4244 | zone_movable_pfn[nid]); | |
4245 | end_pfn = early_node_map[i].end_pfn; | |
4246 | if (start_pfn >= end_pfn) | |
4247 | continue; | |
4248 | ||
4249 | /* Account for what is only usable for kernelcore */ | |
4250 | if (start_pfn < usable_startpfn) { | |
4251 | unsigned long kernel_pages; | |
4252 | kernel_pages = min(end_pfn, usable_startpfn) | |
4253 | - start_pfn; | |
4254 | ||
4255 | kernelcore_remaining -= min(kernel_pages, | |
4256 | kernelcore_remaining); | |
4257 | required_kernelcore -= min(kernel_pages, | |
4258 | required_kernelcore); | |
4259 | ||
4260 | /* Continue if range is now fully accounted */ | |
4261 | if (end_pfn <= usable_startpfn) { | |
4262 | ||
4263 | /* | |
4264 | * Push zone_movable_pfn to the end so | |
4265 | * that if we have to rebalance | |
4266 | * kernelcore across nodes, we will | |
4267 | * not double account here | |
4268 | */ | |
4269 | zone_movable_pfn[nid] = end_pfn; | |
4270 | continue; | |
4271 | } | |
4272 | start_pfn = usable_startpfn; | |
4273 | } | |
4274 | ||
4275 | /* | |
4276 | * The usable PFN range for ZONE_MOVABLE is from | |
4277 | * start_pfn->end_pfn. Calculate size_pages as the | |
4278 | * number of pages used as kernelcore | |
4279 | */ | |
4280 | size_pages = end_pfn - start_pfn; | |
4281 | if (size_pages > kernelcore_remaining) | |
4282 | size_pages = kernelcore_remaining; | |
4283 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4284 | ||
4285 | /* | |
4286 | * Some kernelcore has been met, update counts and | |
4287 | * break if the kernelcore for this node has been | |
4288 | * satisified | |
4289 | */ | |
4290 | required_kernelcore -= min(required_kernelcore, | |
4291 | size_pages); | |
4292 | kernelcore_remaining -= size_pages; | |
4293 | if (!kernelcore_remaining) | |
4294 | break; | |
4295 | } | |
4296 | } | |
4297 | ||
4298 | /* | |
4299 | * If there is still required_kernelcore, we do another pass with one | |
4300 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4301 | * along on the nodes that still have memory until kernelcore is | |
4302 | * satisified | |
4303 | */ | |
4304 | usable_nodes--; | |
4305 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4306 | goto restart; | |
4307 | ||
4308 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4309 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4310 | zone_movable_pfn[nid] = | |
4311 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4312 | |
4313 | out: | |
4314 | /* restore the node_state */ | |
4315 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4316 | } |
4317 | ||
37b07e41 LS |
4318 | /* Any regular memory on that node ? */ |
4319 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4320 | { | |
4321 | #ifdef CONFIG_HIGHMEM | |
4322 | enum zone_type zone_type; | |
4323 | ||
4324 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4325 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4326 | if (zone->present_pages) | |
4327 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4328 | } | |
4329 | #endif | |
4330 | } | |
4331 | ||
c713216d MG |
4332 | /** |
4333 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4334 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4335 | * |
4336 | * This will call free_area_init_node() for each active node in the system. | |
4337 | * Using the page ranges provided by add_active_range(), the size of each | |
4338 | * zone in each node and their holes is calculated. If the maximum PFN | |
4339 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4340 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4341 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4342 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4343 | * at arch_max_dma_pfn. | |
4344 | */ | |
4345 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4346 | { | |
4347 | unsigned long nid; | |
db99100d | 4348 | int i; |
c713216d | 4349 | |
a6af2bc3 MG |
4350 | /* Sort early_node_map as initialisation assumes it is sorted */ |
4351 | sort_node_map(); | |
4352 | ||
c713216d MG |
4353 | /* Record where the zone boundaries are */ |
4354 | memset(arch_zone_lowest_possible_pfn, 0, | |
4355 | sizeof(arch_zone_lowest_possible_pfn)); | |
4356 | memset(arch_zone_highest_possible_pfn, 0, | |
4357 | sizeof(arch_zone_highest_possible_pfn)); | |
4358 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4359 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4360 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4361 | if (i == ZONE_MOVABLE) |
4362 | continue; | |
c713216d MG |
4363 | arch_zone_lowest_possible_pfn[i] = |
4364 | arch_zone_highest_possible_pfn[i-1]; | |
4365 | arch_zone_highest_possible_pfn[i] = | |
4366 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4367 | } | |
2a1e274a MG |
4368 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4369 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4370 | ||
4371 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4372 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4373 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4374 | |
c713216d MG |
4375 | /* Print out the zone ranges */ |
4376 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4377 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4378 | if (i == ZONE_MOVABLE) | |
4379 | continue; | |
5dab8ec1 | 4380 | printk(" %-8s %0#10lx -> %0#10lx\n", |
c713216d MG |
4381 | zone_names[i], |
4382 | arch_zone_lowest_possible_pfn[i], | |
4383 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4384 | } |
4385 | ||
4386 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4387 | printk("Movable zone start PFN for each node\n"); | |
4388 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4389 | if (zone_movable_pfn[i]) | |
4390 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4391 | } | |
c713216d MG |
4392 | |
4393 | /* Print out the early_node_map[] */ | |
4394 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4395 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 4396 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
4397 | early_node_map[i].start_pfn, |
4398 | early_node_map[i].end_pfn); | |
4399 | ||
4400 | /* Initialise every node */ | |
708614e6 | 4401 | mminit_verify_pageflags_layout(); |
8ef82866 | 4402 | setup_nr_node_ids(); |
c713216d MG |
4403 | for_each_online_node(nid) { |
4404 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4405 | free_area_init_node(nid, NULL, |
c713216d | 4406 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4407 | |
4408 | /* Any memory on that node */ | |
4409 | if (pgdat->node_present_pages) | |
4410 | node_set_state(nid, N_HIGH_MEMORY); | |
4411 | check_for_regular_memory(pgdat); | |
c713216d MG |
4412 | } |
4413 | } | |
2a1e274a | 4414 | |
7e63efef | 4415 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4416 | { |
4417 | unsigned long long coremem; | |
4418 | if (!p) | |
4419 | return -EINVAL; | |
4420 | ||
4421 | coremem = memparse(p, &p); | |
7e63efef | 4422 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4423 | |
7e63efef | 4424 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4425 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4426 | ||
4427 | return 0; | |
4428 | } | |
ed7ed365 | 4429 | |
7e63efef MG |
4430 | /* |
4431 | * kernelcore=size sets the amount of memory for use for allocations that | |
4432 | * cannot be reclaimed or migrated. | |
4433 | */ | |
4434 | static int __init cmdline_parse_kernelcore(char *p) | |
4435 | { | |
4436 | return cmdline_parse_core(p, &required_kernelcore); | |
4437 | } | |
4438 | ||
4439 | /* | |
4440 | * movablecore=size sets the amount of memory for use for allocations that | |
4441 | * can be reclaimed or migrated. | |
4442 | */ | |
4443 | static int __init cmdline_parse_movablecore(char *p) | |
4444 | { | |
4445 | return cmdline_parse_core(p, &required_movablecore); | |
4446 | } | |
4447 | ||
ed7ed365 | 4448 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4449 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4450 | |
c713216d MG |
4451 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4452 | ||
0e0b864e | 4453 | /** |
88ca3b94 RD |
4454 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4455 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4456 | * |
4457 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4458 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4459 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4460 | * function may optionally be used to account for unfreeable pages in the |
4461 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4462 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4463 | */ |
4464 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4465 | { | |
4466 | dma_reserve = new_dma_reserve; | |
4467 | } | |
4468 | ||
93b7504e | 4469 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
52765583 | 4470 | struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] }; |
1da177e4 | 4471 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4472 | #endif |
1da177e4 LT |
4473 | |
4474 | void __init free_area_init(unsigned long *zones_size) | |
4475 | { | |
9109fb7b | 4476 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4477 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4478 | } | |
1da177e4 | 4479 | |
1da177e4 LT |
4480 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4481 | unsigned long action, void *hcpu) | |
4482 | { | |
4483 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4484 | |
8bb78442 | 4485 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4486 | drain_pages(cpu); |
4487 | ||
4488 | /* | |
4489 | * Spill the event counters of the dead processor | |
4490 | * into the current processors event counters. | |
4491 | * This artificially elevates the count of the current | |
4492 | * processor. | |
4493 | */ | |
f8891e5e | 4494 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4495 | |
4496 | /* | |
4497 | * Zero the differential counters of the dead processor | |
4498 | * so that the vm statistics are consistent. | |
4499 | * | |
4500 | * This is only okay since the processor is dead and cannot | |
4501 | * race with what we are doing. | |
4502 | */ | |
2244b95a | 4503 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4504 | } |
4505 | return NOTIFY_OK; | |
4506 | } | |
1da177e4 LT |
4507 | |
4508 | void __init page_alloc_init(void) | |
4509 | { | |
4510 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4511 | } | |
4512 | ||
cb45b0e9 HA |
4513 | /* |
4514 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4515 | * or min_free_kbytes changes. | |
4516 | */ | |
4517 | static void calculate_totalreserve_pages(void) | |
4518 | { | |
4519 | struct pglist_data *pgdat; | |
4520 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4521 | enum zone_type i, j; |
cb45b0e9 HA |
4522 | |
4523 | for_each_online_pgdat(pgdat) { | |
4524 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4525 | struct zone *zone = pgdat->node_zones + i; | |
4526 | unsigned long max = 0; | |
4527 | ||
4528 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4529 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4530 | if (zone->lowmem_reserve[j] > max) | |
4531 | max = zone->lowmem_reserve[j]; | |
4532 | } | |
4533 | ||
41858966 MG |
4534 | /* we treat the high watermark as reserved pages. */ |
4535 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
4536 | |
4537 | if (max > zone->present_pages) | |
4538 | max = zone->present_pages; | |
4539 | reserve_pages += max; | |
4540 | } | |
4541 | } | |
4542 | totalreserve_pages = reserve_pages; | |
4543 | } | |
4544 | ||
1da177e4 LT |
4545 | /* |
4546 | * setup_per_zone_lowmem_reserve - called whenever | |
4547 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4548 | * has a correct pages reserved value, so an adequate number of | |
4549 | * pages are left in the zone after a successful __alloc_pages(). | |
4550 | */ | |
4551 | static void setup_per_zone_lowmem_reserve(void) | |
4552 | { | |
4553 | struct pglist_data *pgdat; | |
2f6726e5 | 4554 | enum zone_type j, idx; |
1da177e4 | 4555 | |
ec936fc5 | 4556 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4557 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4558 | struct zone *zone = pgdat->node_zones + j; | |
4559 | unsigned long present_pages = zone->present_pages; | |
4560 | ||
4561 | zone->lowmem_reserve[j] = 0; | |
4562 | ||
2f6726e5 CL |
4563 | idx = j; |
4564 | while (idx) { | |
1da177e4 LT |
4565 | struct zone *lower_zone; |
4566 | ||
2f6726e5 CL |
4567 | idx--; |
4568 | ||
1da177e4 LT |
4569 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4570 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4571 | ||
4572 | lower_zone = pgdat->node_zones + idx; | |
4573 | lower_zone->lowmem_reserve[j] = present_pages / | |
4574 | sysctl_lowmem_reserve_ratio[idx]; | |
4575 | present_pages += lower_zone->present_pages; | |
4576 | } | |
4577 | } | |
4578 | } | |
cb45b0e9 HA |
4579 | |
4580 | /* update totalreserve_pages */ | |
4581 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4582 | } |
4583 | ||
88ca3b94 | 4584 | /** |
bc75d33f | 4585 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
bce7394a | 4586 | * or when memory is hot-{added|removed} |
88ca3b94 | 4587 | * |
bc75d33f MK |
4588 | * Ensures that the watermark[min,low,high] values for each zone are set |
4589 | * correctly with respect to min_free_kbytes. | |
1da177e4 | 4590 | */ |
bc75d33f | 4591 | void setup_per_zone_wmarks(void) |
1da177e4 LT |
4592 | { |
4593 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4594 | unsigned long lowmem_pages = 0; | |
4595 | struct zone *zone; | |
4596 | unsigned long flags; | |
4597 | ||
4598 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4599 | for_each_zone(zone) { | |
4600 | if (!is_highmem(zone)) | |
4601 | lowmem_pages += zone->present_pages; | |
4602 | } | |
4603 | ||
4604 | for_each_zone(zone) { | |
ac924c60 AM |
4605 | u64 tmp; |
4606 | ||
1125b4e3 | 4607 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4608 | tmp = (u64)pages_min * zone->present_pages; |
4609 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4610 | if (is_highmem(zone)) { |
4611 | /* | |
669ed175 NP |
4612 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4613 | * need highmem pages, so cap pages_min to a small | |
4614 | * value here. | |
4615 | * | |
41858966 | 4616 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
4617 | * deltas controls asynch page reclaim, and so should |
4618 | * not be capped for highmem. | |
1da177e4 LT |
4619 | */ |
4620 | int min_pages; | |
4621 | ||
4622 | min_pages = zone->present_pages / 1024; | |
4623 | if (min_pages < SWAP_CLUSTER_MAX) | |
4624 | min_pages = SWAP_CLUSTER_MAX; | |
4625 | if (min_pages > 128) | |
4626 | min_pages = 128; | |
41858966 | 4627 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 4628 | } else { |
669ed175 NP |
4629 | /* |
4630 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4631 | * proportionate to the zone's size. |
4632 | */ | |
41858966 | 4633 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
4634 | } |
4635 | ||
41858966 MG |
4636 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
4637 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
56fd56b8 | 4638 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4639 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4640 | } |
cb45b0e9 HA |
4641 | |
4642 | /* update totalreserve_pages */ | |
4643 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4644 | } |
4645 | ||
55a4462a | 4646 | /* |
556adecb RR |
4647 | * The inactive anon list should be small enough that the VM never has to |
4648 | * do too much work, but large enough that each inactive page has a chance | |
4649 | * to be referenced again before it is swapped out. | |
4650 | * | |
4651 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4652 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4653 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4654 | * the anonymous pages are kept on the inactive list. | |
4655 | * | |
4656 | * total target max | |
4657 | * memory ratio inactive anon | |
4658 | * ------------------------------------- | |
4659 | * 10MB 1 5MB | |
4660 | * 100MB 1 50MB | |
4661 | * 1GB 3 250MB | |
4662 | * 10GB 10 0.9GB | |
4663 | * 100GB 31 3GB | |
4664 | * 1TB 101 10GB | |
4665 | * 10TB 320 32GB | |
4666 | */ | |
96cb4df5 | 4667 | void calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 4668 | { |
96cb4df5 | 4669 | unsigned int gb, ratio; |
556adecb | 4670 | |
96cb4df5 MK |
4671 | /* Zone size in gigabytes */ |
4672 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
4673 | if (gb) | |
556adecb | 4674 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
4675 | else |
4676 | ratio = 1; | |
556adecb | 4677 | |
96cb4df5 MK |
4678 | zone->inactive_ratio = ratio; |
4679 | } | |
556adecb | 4680 | |
96cb4df5 MK |
4681 | static void __init setup_per_zone_inactive_ratio(void) |
4682 | { | |
4683 | struct zone *zone; | |
4684 | ||
4685 | for_each_zone(zone) | |
4686 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
4687 | } |
4688 | ||
1da177e4 LT |
4689 | /* |
4690 | * Initialise min_free_kbytes. | |
4691 | * | |
4692 | * For small machines we want it small (128k min). For large machines | |
4693 | * we want it large (64MB max). But it is not linear, because network | |
4694 | * bandwidth does not increase linearly with machine size. We use | |
4695 | * | |
4696 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4697 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4698 | * | |
4699 | * which yields | |
4700 | * | |
4701 | * 16MB: 512k | |
4702 | * 32MB: 724k | |
4703 | * 64MB: 1024k | |
4704 | * 128MB: 1448k | |
4705 | * 256MB: 2048k | |
4706 | * 512MB: 2896k | |
4707 | * 1024MB: 4096k | |
4708 | * 2048MB: 5792k | |
4709 | * 4096MB: 8192k | |
4710 | * 8192MB: 11584k | |
4711 | * 16384MB: 16384k | |
4712 | */ | |
bc75d33f | 4713 | static int __init init_per_zone_wmark_min(void) |
1da177e4 LT |
4714 | { |
4715 | unsigned long lowmem_kbytes; | |
4716 | ||
4717 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4718 | ||
4719 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4720 | if (min_free_kbytes < 128) | |
4721 | min_free_kbytes = 128; | |
4722 | if (min_free_kbytes > 65536) | |
4723 | min_free_kbytes = 65536; | |
bc75d33f | 4724 | setup_per_zone_wmarks(); |
1da177e4 | 4725 | setup_per_zone_lowmem_reserve(); |
556adecb | 4726 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
4727 | return 0; |
4728 | } | |
bc75d33f | 4729 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
4730 | |
4731 | /* | |
4732 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4733 | * that we can call two helper functions whenever min_free_kbytes | |
4734 | * changes. | |
4735 | */ | |
4736 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4737 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 4738 | { |
8d65af78 | 4739 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 4740 | if (write) |
bc75d33f | 4741 | setup_per_zone_wmarks(); |
1da177e4 LT |
4742 | return 0; |
4743 | } | |
4744 | ||
9614634f CL |
4745 | #ifdef CONFIG_NUMA |
4746 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4747 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
4748 | { |
4749 | struct zone *zone; | |
4750 | int rc; | |
4751 | ||
8d65af78 | 4752 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
4753 | if (rc) |
4754 | return rc; | |
4755 | ||
4756 | for_each_zone(zone) | |
8417bba4 | 4757 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4758 | sysctl_min_unmapped_ratio) / 100; |
4759 | return 0; | |
4760 | } | |
0ff38490 CL |
4761 | |
4762 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4763 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
4764 | { |
4765 | struct zone *zone; | |
4766 | int rc; | |
4767 | ||
8d65af78 | 4768 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
4769 | if (rc) |
4770 | return rc; | |
4771 | ||
4772 | for_each_zone(zone) | |
4773 | zone->min_slab_pages = (zone->present_pages * | |
4774 | sysctl_min_slab_ratio) / 100; | |
4775 | return 0; | |
4776 | } | |
9614634f CL |
4777 | #endif |
4778 | ||
1da177e4 LT |
4779 | /* |
4780 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4781 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4782 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4783 | * | |
4784 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 4785 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
4786 | * if in function of the boot time zone sizes. |
4787 | */ | |
4788 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4789 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 4790 | { |
8d65af78 | 4791 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
4792 | setup_per_zone_lowmem_reserve(); |
4793 | return 0; | |
4794 | } | |
4795 | ||
8ad4b1fb RS |
4796 | /* |
4797 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4798 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4799 | * can have before it gets flushed back to buddy allocator. | |
4800 | */ | |
4801 | ||
4802 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4803 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
4804 | { |
4805 | struct zone *zone; | |
4806 | unsigned int cpu; | |
4807 | int ret; | |
4808 | ||
8d65af78 | 4809 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
8ad4b1fb RS |
4810 | if (!write || (ret == -EINVAL)) |
4811 | return ret; | |
364df0eb | 4812 | for_each_populated_zone(zone) { |
8ad4b1fb RS |
4813 | for_each_online_cpu(cpu) { |
4814 | unsigned long high; | |
4815 | high = zone->present_pages / percpu_pagelist_fraction; | |
4816 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4817 | } | |
4818 | } | |
4819 | return 0; | |
4820 | } | |
4821 | ||
f034b5d4 | 4822 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4823 | |
4824 | #ifdef CONFIG_NUMA | |
4825 | static int __init set_hashdist(char *str) | |
4826 | { | |
4827 | if (!str) | |
4828 | return 0; | |
4829 | hashdist = simple_strtoul(str, &str, 0); | |
4830 | return 1; | |
4831 | } | |
4832 | __setup("hashdist=", set_hashdist); | |
4833 | #endif | |
4834 | ||
4835 | /* | |
4836 | * allocate a large system hash table from bootmem | |
4837 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4838 | * quantity of entries | |
4839 | * - limit is the number of hash buckets, not the total allocation size | |
4840 | */ | |
4841 | void *__init alloc_large_system_hash(const char *tablename, | |
4842 | unsigned long bucketsize, | |
4843 | unsigned long numentries, | |
4844 | int scale, | |
4845 | int flags, | |
4846 | unsigned int *_hash_shift, | |
4847 | unsigned int *_hash_mask, | |
4848 | unsigned long limit) | |
4849 | { | |
4850 | unsigned long long max = limit; | |
4851 | unsigned long log2qty, size; | |
4852 | void *table = NULL; | |
4853 | ||
4854 | /* allow the kernel cmdline to have a say */ | |
4855 | if (!numentries) { | |
4856 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4857 | numentries = nr_kernel_pages; |
1da177e4 LT |
4858 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4859 | numentries >>= 20 - PAGE_SHIFT; | |
4860 | numentries <<= 20 - PAGE_SHIFT; | |
4861 | ||
4862 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4863 | if (scale > PAGE_SHIFT) | |
4864 | numentries >>= (scale - PAGE_SHIFT); | |
4865 | else | |
4866 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4867 | |
4868 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
4869 | if (unlikely(flags & HASH_SMALL)) { |
4870 | /* Makes no sense without HASH_EARLY */ | |
4871 | WARN_ON(!(flags & HASH_EARLY)); | |
4872 | if (!(numentries >> *_hash_shift)) { | |
4873 | numentries = 1UL << *_hash_shift; | |
4874 | BUG_ON(!numentries); | |
4875 | } | |
4876 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 4877 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 4878 | } |
6e692ed3 | 4879 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4880 | |
4881 | /* limit allocation size to 1/16 total memory by default */ | |
4882 | if (max == 0) { | |
4883 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4884 | do_div(max, bucketsize); | |
4885 | } | |
4886 | ||
4887 | if (numentries > max) | |
4888 | numentries = max; | |
4889 | ||
f0d1b0b3 | 4890 | log2qty = ilog2(numentries); |
1da177e4 LT |
4891 | |
4892 | do { | |
4893 | size = bucketsize << log2qty; | |
4894 | if (flags & HASH_EARLY) | |
74768ed8 | 4895 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
4896 | else if (hashdist) |
4897 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4898 | else { | |
1037b83b ED |
4899 | /* |
4900 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
4901 | * some pages at the end of hash table which |
4902 | * alloc_pages_exact() automatically does | |
1037b83b | 4903 | */ |
264ef8a9 | 4904 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 4905 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
4906 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
4907 | } | |
1da177e4 LT |
4908 | } |
4909 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4910 | ||
4911 | if (!table) | |
4912 | panic("Failed to allocate %s hash table\n", tablename); | |
4913 | ||
b49ad484 | 4914 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4915 | tablename, |
4916 | (1U << log2qty), | |
f0d1b0b3 | 4917 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4918 | size); |
4919 | ||
4920 | if (_hash_shift) | |
4921 | *_hash_shift = log2qty; | |
4922 | if (_hash_mask) | |
4923 | *_hash_mask = (1 << log2qty) - 1; | |
4924 | ||
4925 | return table; | |
4926 | } | |
a117e66e | 4927 | |
835c134e MG |
4928 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4929 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4930 | unsigned long pfn) | |
4931 | { | |
4932 | #ifdef CONFIG_SPARSEMEM | |
4933 | return __pfn_to_section(pfn)->pageblock_flags; | |
4934 | #else | |
4935 | return zone->pageblock_flags; | |
4936 | #endif /* CONFIG_SPARSEMEM */ | |
4937 | } | |
4938 | ||
4939 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4940 | { | |
4941 | #ifdef CONFIG_SPARSEMEM | |
4942 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4943 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4944 | #else |
4945 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4946 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4947 | #endif /* CONFIG_SPARSEMEM */ |
4948 | } | |
4949 | ||
4950 | /** | |
d9c23400 | 4951 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4952 | * @page: The page within the block of interest |
4953 | * @start_bitidx: The first bit of interest to retrieve | |
4954 | * @end_bitidx: The last bit of interest | |
4955 | * returns pageblock_bits flags | |
4956 | */ | |
4957 | unsigned long get_pageblock_flags_group(struct page *page, | |
4958 | int start_bitidx, int end_bitidx) | |
4959 | { | |
4960 | struct zone *zone; | |
4961 | unsigned long *bitmap; | |
4962 | unsigned long pfn, bitidx; | |
4963 | unsigned long flags = 0; | |
4964 | unsigned long value = 1; | |
4965 | ||
4966 | zone = page_zone(page); | |
4967 | pfn = page_to_pfn(page); | |
4968 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4969 | bitidx = pfn_to_bitidx(zone, pfn); | |
4970 | ||
4971 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4972 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4973 | flags |= value; | |
6220ec78 | 4974 | |
835c134e MG |
4975 | return flags; |
4976 | } | |
4977 | ||
4978 | /** | |
d9c23400 | 4979 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4980 | * @page: The page within the block of interest |
4981 | * @start_bitidx: The first bit of interest | |
4982 | * @end_bitidx: The last bit of interest | |
4983 | * @flags: The flags to set | |
4984 | */ | |
4985 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4986 | int start_bitidx, int end_bitidx) | |
4987 | { | |
4988 | struct zone *zone; | |
4989 | unsigned long *bitmap; | |
4990 | unsigned long pfn, bitidx; | |
4991 | unsigned long value = 1; | |
4992 | ||
4993 | zone = page_zone(page); | |
4994 | pfn = page_to_pfn(page); | |
4995 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4996 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
4997 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
4998 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
4999 | |
5000 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5001 | if (flags & value) | |
5002 | __set_bit(bitidx + start_bitidx, bitmap); | |
5003 | else | |
5004 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5005 | } | |
a5d76b54 KH |
5006 | |
5007 | /* | |
5008 | * This is designed as sub function...plz see page_isolation.c also. | |
5009 | * set/clear page block's type to be ISOLATE. | |
5010 | * page allocater never alloc memory from ISOLATE block. | |
5011 | */ | |
5012 | ||
5013 | int set_migratetype_isolate(struct page *page) | |
5014 | { | |
5015 | struct zone *zone; | |
925cc71e RJ |
5016 | struct page *curr_page; |
5017 | unsigned long flags, pfn, iter; | |
5018 | unsigned long immobile = 0; | |
5019 | struct memory_isolate_notify arg; | |
5020 | int notifier_ret; | |
a5d76b54 | 5021 | int ret = -EBUSY; |
8e7e40d9 | 5022 | int zone_idx; |
a5d76b54 KH |
5023 | |
5024 | zone = page_zone(page); | |
8e7e40d9 | 5025 | zone_idx = zone_idx(zone); |
925cc71e | 5026 | |
a5d76b54 | 5027 | spin_lock_irqsave(&zone->lock, flags); |
925cc71e RJ |
5028 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE || |
5029 | zone_idx == ZONE_MOVABLE) { | |
5030 | ret = 0; | |
5031 | goto out; | |
5032 | } | |
5033 | ||
5034 | pfn = page_to_pfn(page); | |
5035 | arg.start_pfn = pfn; | |
5036 | arg.nr_pages = pageblock_nr_pages; | |
5037 | arg.pages_found = 0; | |
5038 | ||
a5d76b54 | 5039 | /* |
925cc71e RJ |
5040 | * It may be possible to isolate a pageblock even if the |
5041 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
5042 | * notifier chain is used by balloon drivers to return the | |
5043 | * number of pages in a range that are held by the balloon | |
5044 | * driver to shrink memory. If all the pages are accounted for | |
5045 | * by balloons, are free, or on the LRU, isolation can continue. | |
5046 | * Later, for example, when memory hotplug notifier runs, these | |
5047 | * pages reported as "can be isolated" should be isolated(freed) | |
5048 | * by the balloon driver through the memory notifier chain. | |
a5d76b54 | 5049 | */ |
925cc71e RJ |
5050 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); |
5051 | notifier_ret = notifier_to_errno(notifier_ret); | |
5052 | if (notifier_ret || !arg.pages_found) | |
a5d76b54 | 5053 | goto out; |
925cc71e RJ |
5054 | |
5055 | for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) { | |
5056 | if (!pfn_valid_within(pfn)) | |
5057 | continue; | |
5058 | ||
5059 | curr_page = pfn_to_page(iter); | |
5060 | if (!page_count(curr_page) || PageLRU(curr_page)) | |
5061 | continue; | |
5062 | ||
5063 | immobile++; | |
5064 | } | |
5065 | ||
5066 | if (arg.pages_found == immobile) | |
5067 | ret = 0; | |
5068 | ||
a5d76b54 | 5069 | out: |
925cc71e RJ |
5070 | if (!ret) { |
5071 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
5072 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
5073 | } | |
5074 | ||
a5d76b54 KH |
5075 | spin_unlock_irqrestore(&zone->lock, flags); |
5076 | if (!ret) | |
9f8f2172 | 5077 | drain_all_pages(); |
a5d76b54 KH |
5078 | return ret; |
5079 | } | |
5080 | ||
5081 | void unset_migratetype_isolate(struct page *page) | |
5082 | { | |
5083 | struct zone *zone; | |
5084 | unsigned long flags; | |
5085 | zone = page_zone(page); | |
5086 | spin_lock_irqsave(&zone->lock, flags); | |
5087 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
5088 | goto out; | |
5089 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5090 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
5091 | out: | |
5092 | spin_unlock_irqrestore(&zone->lock, flags); | |
5093 | } | |
0c0e6195 KH |
5094 | |
5095 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5096 | /* | |
5097 | * All pages in the range must be isolated before calling this. | |
5098 | */ | |
5099 | void | |
5100 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5101 | { | |
5102 | struct page *page; | |
5103 | struct zone *zone; | |
5104 | int order, i; | |
5105 | unsigned long pfn; | |
5106 | unsigned long flags; | |
5107 | /* find the first valid pfn */ | |
5108 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5109 | if (pfn_valid(pfn)) | |
5110 | break; | |
5111 | if (pfn == end_pfn) | |
5112 | return; | |
5113 | zone = page_zone(pfn_to_page(pfn)); | |
5114 | spin_lock_irqsave(&zone->lock, flags); | |
5115 | pfn = start_pfn; | |
5116 | while (pfn < end_pfn) { | |
5117 | if (!pfn_valid(pfn)) { | |
5118 | pfn++; | |
5119 | continue; | |
5120 | } | |
5121 | page = pfn_to_page(pfn); | |
5122 | BUG_ON(page_count(page)); | |
5123 | BUG_ON(!PageBuddy(page)); | |
5124 | order = page_order(page); | |
5125 | #ifdef CONFIG_DEBUG_VM | |
5126 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5127 | pfn, 1 << order, end_pfn); | |
5128 | #endif | |
5129 | list_del(&page->lru); | |
5130 | rmv_page_order(page); | |
5131 | zone->free_area[order].nr_free--; | |
5132 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
5133 | - (1UL << order)); | |
5134 | for (i = 0; i < (1 << order); i++) | |
5135 | SetPageReserved((page+i)); | |
5136 | pfn += (1 << order); | |
5137 | } | |
5138 | spin_unlock_irqrestore(&zone->lock, flags); | |
5139 | } | |
5140 | #endif | |
8d22ba1b WF |
5141 | |
5142 | #ifdef CONFIG_MEMORY_FAILURE | |
5143 | bool is_free_buddy_page(struct page *page) | |
5144 | { | |
5145 | struct zone *zone = page_zone(page); | |
5146 | unsigned long pfn = page_to_pfn(page); | |
5147 | unsigned long flags; | |
5148 | int order; | |
5149 | ||
5150 | spin_lock_irqsave(&zone->lock, flags); | |
5151 | for (order = 0; order < MAX_ORDER; order++) { | |
5152 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
5153 | ||
5154 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
5155 | break; | |
5156 | } | |
5157 | spin_unlock_irqrestore(&zone->lock, flags); | |
5158 | ||
5159 | return order < MAX_ORDER; | |
5160 | } | |
5161 | #endif |