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> |
0d3d062a | 51 | #include <trace/events/kmem.h> |
1da177e4 LT |
52 | |
53 | #include <asm/tlbflush.h> | |
ac924c60 | 54 | #include <asm/div64.h> |
1da177e4 LT |
55 | #include "internal.h" |
56 | ||
57 | /* | |
13808910 | 58 | * Array of node states. |
1da177e4 | 59 | */ |
13808910 CL |
60 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
61 | [N_POSSIBLE] = NODE_MASK_ALL, | |
62 | [N_ONLINE] = { { [0] = 1UL } }, | |
63 | #ifndef CONFIG_NUMA | |
64 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
65 | #ifdef CONFIG_HIGHMEM | |
66 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
67 | #endif | |
68 | [N_CPU] = { { [0] = 1UL } }, | |
69 | #endif /* NUMA */ | |
70 | }; | |
71 | EXPORT_SYMBOL(node_states); | |
72 | ||
6c231b7b | 73 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 74 | unsigned long totalreserve_pages __read_mostly; |
8ad4b1fb | 75 | int percpu_pagelist_fraction; |
dcce284a | 76 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 77 | |
d9c23400 MG |
78 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
79 | int pageblock_order __read_mostly; | |
80 | #endif | |
81 | ||
d98c7a09 | 82 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 83 | |
1da177e4 LT |
84 | /* |
85 | * results with 256, 32 in the lowmem_reserve sysctl: | |
86 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
87 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
88 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
89 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
90 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
91 | * |
92 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
93 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 94 | */ |
2f1b6248 | 95 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 96 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 97 | 256, |
4b51d669 | 98 | #endif |
fb0e7942 | 99 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 100 | 256, |
fb0e7942 | 101 | #endif |
e53ef38d | 102 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 103 | 32, |
e53ef38d | 104 | #endif |
2a1e274a | 105 | 32, |
2f1b6248 | 106 | }; |
1da177e4 LT |
107 | |
108 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 109 | |
15ad7cdc | 110 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 111 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 112 | "DMA", |
4b51d669 | 113 | #endif |
fb0e7942 | 114 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 115 | "DMA32", |
fb0e7942 | 116 | #endif |
2f1b6248 | 117 | "Normal", |
e53ef38d | 118 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 119 | "HighMem", |
e53ef38d | 120 | #endif |
2a1e274a | 121 | "Movable", |
2f1b6248 CL |
122 | }; |
123 | ||
1da177e4 LT |
124 | int min_free_kbytes = 1024; |
125 | ||
2c85f51d JB |
126 | static unsigned long __meminitdata nr_kernel_pages; |
127 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 128 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 129 | |
c713216d MG |
130 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
131 | /* | |
183ff22b | 132 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
133 | * ranges of memory (RAM) that may be registered with add_active_range(). |
134 | * Ranges passed to add_active_range() will be merged if possible | |
135 | * so the number of times add_active_range() can be called is | |
136 | * related to the number of nodes and the number of holes | |
137 | */ | |
138 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
139 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
140 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
141 | #else | |
142 | #if MAX_NUMNODES >= 32 | |
143 | /* If there can be many nodes, allow up to 50 holes per node */ | |
144 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
145 | #else | |
146 | /* By default, allow up to 256 distinct regions */ | |
147 | #define MAX_ACTIVE_REGIONS 256 | |
148 | #endif | |
149 | #endif | |
150 | ||
98011f56 JB |
151 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
152 | static int __meminitdata nr_nodemap_entries; | |
153 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
154 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
b69a7288 | 155 | static unsigned long __initdata required_kernelcore; |
484f51f8 | 156 | static unsigned long __initdata required_movablecore; |
b69a7288 | 157 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
158 | |
159 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
160 | int movable_zone; | |
161 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
162 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
163 | ||
418508c1 MS |
164 | #if MAX_NUMNODES > 1 |
165 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 166 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 167 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 168 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
169 | #endif |
170 | ||
9ef9acb0 MG |
171 | int page_group_by_mobility_disabled __read_mostly; |
172 | ||
b2a0ac88 MG |
173 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
174 | { | |
49255c61 MG |
175 | |
176 | if (unlikely(page_group_by_mobility_disabled)) | |
177 | migratetype = MIGRATE_UNMOVABLE; | |
178 | ||
b2a0ac88 MG |
179 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
180 | PB_migrate, PB_migrate_end); | |
181 | } | |
182 | ||
7f33d49a RW |
183 | bool oom_killer_disabled __read_mostly; |
184 | ||
13e7444b | 185 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 186 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 187 | { |
bdc8cb98 DH |
188 | int ret = 0; |
189 | unsigned seq; | |
190 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 191 | |
bdc8cb98 DH |
192 | do { |
193 | seq = zone_span_seqbegin(zone); | |
194 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
195 | ret = 1; | |
196 | else if (pfn < zone->zone_start_pfn) | |
197 | ret = 1; | |
198 | } while (zone_span_seqretry(zone, seq)); | |
199 | ||
200 | return ret; | |
c6a57e19 DH |
201 | } |
202 | ||
203 | static int page_is_consistent(struct zone *zone, struct page *page) | |
204 | { | |
14e07298 | 205 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 206 | return 0; |
1da177e4 | 207 | if (zone != page_zone(page)) |
c6a57e19 DH |
208 | return 0; |
209 | ||
210 | return 1; | |
211 | } | |
212 | /* | |
213 | * Temporary debugging check for pages not lying within a given zone. | |
214 | */ | |
215 | static int bad_range(struct zone *zone, struct page *page) | |
216 | { | |
217 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 218 | return 1; |
c6a57e19 DH |
219 | if (!page_is_consistent(zone, page)) |
220 | return 1; | |
221 | ||
1da177e4 LT |
222 | return 0; |
223 | } | |
13e7444b NP |
224 | #else |
225 | static inline int bad_range(struct zone *zone, struct page *page) | |
226 | { | |
227 | return 0; | |
228 | } | |
229 | #endif | |
230 | ||
224abf92 | 231 | static void bad_page(struct page *page) |
1da177e4 | 232 | { |
d936cf9b HD |
233 | static unsigned long resume; |
234 | static unsigned long nr_shown; | |
235 | static unsigned long nr_unshown; | |
236 | ||
2a7684a2 WF |
237 | /* Don't complain about poisoned pages */ |
238 | if (PageHWPoison(page)) { | |
239 | __ClearPageBuddy(page); | |
240 | return; | |
241 | } | |
242 | ||
d936cf9b HD |
243 | /* |
244 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
245 | * or allow a steady drip of one report per second. | |
246 | */ | |
247 | if (nr_shown == 60) { | |
248 | if (time_before(jiffies, resume)) { | |
249 | nr_unshown++; | |
250 | goto out; | |
251 | } | |
252 | if (nr_unshown) { | |
1e9e6365 HD |
253 | printk(KERN_ALERT |
254 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
255 | nr_unshown); |
256 | nr_unshown = 0; | |
257 | } | |
258 | nr_shown = 0; | |
259 | } | |
260 | if (nr_shown++ == 0) | |
261 | resume = jiffies + 60 * HZ; | |
262 | ||
1e9e6365 | 263 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 264 | current->comm, page_to_pfn(page)); |
1e9e6365 | 265 | printk(KERN_ALERT |
3dc14741 HD |
266 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
267 | page, (void *)page->flags, page_count(page), | |
268 | page_mapcount(page), page->mapping, page->index); | |
3dc14741 | 269 | |
1da177e4 | 270 | dump_stack(); |
d936cf9b | 271 | out: |
8cc3b392 HD |
272 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
273 | __ClearPageBuddy(page); | |
9f158333 | 274 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
275 | } |
276 | ||
1da177e4 LT |
277 | /* |
278 | * Higher-order pages are called "compound pages". They are structured thusly: | |
279 | * | |
280 | * The first PAGE_SIZE page is called the "head page". | |
281 | * | |
282 | * The remaining PAGE_SIZE pages are called "tail pages". | |
283 | * | |
284 | * All pages have PG_compound set. All pages have their ->private pointing at | |
285 | * the head page (even the head page has this). | |
286 | * | |
41d78ba5 HD |
287 | * The first tail page's ->lru.next holds the address of the compound page's |
288 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
289 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 290 | */ |
d98c7a09 HD |
291 | |
292 | static void free_compound_page(struct page *page) | |
293 | { | |
d85f3385 | 294 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
295 | } |
296 | ||
01ad1c08 | 297 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
298 | { |
299 | int i; | |
300 | int nr_pages = 1 << order; | |
301 | ||
302 | set_compound_page_dtor(page, free_compound_page); | |
303 | set_compound_order(page, order); | |
304 | __SetPageHead(page); | |
305 | for (i = 1; i < nr_pages; i++) { | |
306 | struct page *p = page + i; | |
307 | ||
308 | __SetPageTail(p); | |
309 | p->first_page = page; | |
310 | } | |
311 | } | |
312 | ||
8cc3b392 | 313 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
314 | { |
315 | int i; | |
316 | int nr_pages = 1 << order; | |
8cc3b392 | 317 | int bad = 0; |
1da177e4 | 318 | |
8cc3b392 HD |
319 | if (unlikely(compound_order(page) != order) || |
320 | unlikely(!PageHead(page))) { | |
224abf92 | 321 | bad_page(page); |
8cc3b392 HD |
322 | bad++; |
323 | } | |
1da177e4 | 324 | |
6d777953 | 325 | __ClearPageHead(page); |
8cc3b392 | 326 | |
18229df5 AW |
327 | for (i = 1; i < nr_pages; i++) { |
328 | struct page *p = page + i; | |
1da177e4 | 329 | |
e713a21d | 330 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 331 | bad_page(page); |
8cc3b392 HD |
332 | bad++; |
333 | } | |
d85f3385 | 334 | __ClearPageTail(p); |
1da177e4 | 335 | } |
8cc3b392 HD |
336 | |
337 | return bad; | |
1da177e4 | 338 | } |
1da177e4 | 339 | |
17cf4406 NP |
340 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
341 | { | |
342 | int i; | |
343 | ||
6626c5d5 AM |
344 | /* |
345 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
346 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
347 | */ | |
725d704e | 348 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
349 | for (i = 0; i < (1 << order); i++) |
350 | clear_highpage(page + i); | |
351 | } | |
352 | ||
6aa3001b AM |
353 | static inline void set_page_order(struct page *page, int order) |
354 | { | |
4c21e2f2 | 355 | set_page_private(page, order); |
676165a8 | 356 | __SetPageBuddy(page); |
1da177e4 LT |
357 | } |
358 | ||
359 | static inline void rmv_page_order(struct page *page) | |
360 | { | |
676165a8 | 361 | __ClearPageBuddy(page); |
4c21e2f2 | 362 | set_page_private(page, 0); |
1da177e4 LT |
363 | } |
364 | ||
365 | /* | |
366 | * Locate the struct page for both the matching buddy in our | |
367 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
368 | * | |
369 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
370 | * the following equation: | |
371 | * B2 = B1 ^ (1 << O) | |
372 | * For example, if the starting buddy (buddy2) is #8 its order | |
373 | * 1 buddy is #10: | |
374 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
375 | * | |
376 | * 2) Any buddy B will have an order O+1 parent P which | |
377 | * satisfies the following equation: | |
378 | * P = B & ~(1 << O) | |
379 | * | |
d6e05edc | 380 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
381 | */ |
382 | static inline struct page * | |
383 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
384 | { | |
385 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
386 | ||
387 | return page + (buddy_idx - page_idx); | |
388 | } | |
389 | ||
390 | static inline unsigned long | |
391 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
392 | { | |
393 | return (page_idx & ~(1 << order)); | |
394 | } | |
395 | ||
396 | /* | |
397 | * This function checks whether a page is free && is the buddy | |
398 | * we can do coalesce a page and its buddy if | |
13e7444b | 399 | * (a) the buddy is not in a hole && |
676165a8 | 400 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
401 | * (c) a page and its buddy have the same order && |
402 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
403 | * |
404 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
405 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 406 | * |
676165a8 | 407 | * For recording page's order, we use page_private(page). |
1da177e4 | 408 | */ |
cb2b95e1 AW |
409 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
410 | int order) | |
1da177e4 | 411 | { |
14e07298 | 412 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 413 | return 0; |
13e7444b | 414 | |
cb2b95e1 AW |
415 | if (page_zone_id(page) != page_zone_id(buddy)) |
416 | return 0; | |
417 | ||
418 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
a3af9c38 | 419 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 420 | return 1; |
676165a8 | 421 | } |
6aa3001b | 422 | return 0; |
1da177e4 LT |
423 | } |
424 | ||
425 | /* | |
426 | * Freeing function for a buddy system allocator. | |
427 | * | |
428 | * The concept of a buddy system is to maintain direct-mapped table | |
429 | * (containing bit values) for memory blocks of various "orders". | |
430 | * The bottom level table contains the map for the smallest allocatable | |
431 | * units of memory (here, pages), and each level above it describes | |
432 | * pairs of units from the levels below, hence, "buddies". | |
433 | * At a high level, all that happens here is marking the table entry | |
434 | * at the bottom level available, and propagating the changes upward | |
435 | * as necessary, plus some accounting needed to play nicely with other | |
436 | * parts of the VM system. | |
437 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 438 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 439 | * order is recorded in page_private(page) field. |
1da177e4 LT |
440 | * So when we are allocating or freeing one, we can derive the state of the |
441 | * other. That is, if we allocate a small block, and both were | |
442 | * free, the remainder of the region must be split into blocks. | |
443 | * If a block is freed, and its buddy is also free, then this | |
444 | * triggers coalescing into a block of larger size. | |
445 | * | |
446 | * -- wli | |
447 | */ | |
448 | ||
48db57f8 | 449 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
450 | struct zone *zone, unsigned int order, |
451 | int migratetype) | |
1da177e4 LT |
452 | { |
453 | unsigned long page_idx; | |
1da177e4 | 454 | |
224abf92 | 455 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
456 | if (unlikely(destroy_compound_page(page, order))) |
457 | return; | |
1da177e4 | 458 | |
ed0ae21d MG |
459 | VM_BUG_ON(migratetype == -1); |
460 | ||
1da177e4 LT |
461 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
462 | ||
f2260e6b | 463 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 464 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 465 | |
1da177e4 LT |
466 | while (order < MAX_ORDER-1) { |
467 | unsigned long combined_idx; | |
1da177e4 LT |
468 | struct page *buddy; |
469 | ||
1da177e4 | 470 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 471 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 472 | break; |
13e7444b | 473 | |
3c82d0ce | 474 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 475 | list_del(&buddy->lru); |
b2a0ac88 | 476 | zone->free_area[order].nr_free--; |
1da177e4 | 477 | rmv_page_order(buddy); |
13e7444b | 478 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
479 | page = page + (combined_idx - page_idx); |
480 | page_idx = combined_idx; | |
481 | order++; | |
482 | } | |
483 | set_page_order(page, order); | |
b2a0ac88 MG |
484 | list_add(&page->lru, |
485 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
486 | zone->free_area[order].nr_free++; |
487 | } | |
488 | ||
092cead6 KM |
489 | /* |
490 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
491 | * Page should not be on lru, so no need to fix that up. | |
492 | * free_pages_check() will verify... | |
493 | */ | |
494 | static inline void free_page_mlock(struct page *page) | |
495 | { | |
092cead6 KM |
496 | __dec_zone_page_state(page, NR_MLOCK); |
497 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
498 | } | |
092cead6 | 499 | |
224abf92 | 500 | static inline int free_pages_check(struct page *page) |
1da177e4 | 501 | { |
92be2e33 NP |
502 | if (unlikely(page_mapcount(page) | |
503 | (page->mapping != NULL) | | |
a3af9c38 | 504 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 505 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
224abf92 | 506 | bad_page(page); |
79f4b7bf | 507 | return 1; |
8cc3b392 | 508 | } |
79f4b7bf HD |
509 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
510 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
511 | return 0; | |
1da177e4 LT |
512 | } |
513 | ||
514 | /* | |
5f8dcc21 | 515 | * Frees a number of pages from the PCP lists |
1da177e4 | 516 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 517 | * count is the number of pages to free. |
1da177e4 LT |
518 | * |
519 | * If the zone was previously in an "all pages pinned" state then look to | |
520 | * see if this freeing clears that state. | |
521 | * | |
522 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
523 | * pinned" detection logic. | |
524 | */ | |
5f8dcc21 MG |
525 | static void free_pcppages_bulk(struct zone *zone, int count, |
526 | struct per_cpu_pages *pcp) | |
1da177e4 | 527 | { |
5f8dcc21 | 528 | int migratetype = 0; |
a6f9edd6 | 529 | int batch_free = 0; |
5f8dcc21 | 530 | |
c54ad30c | 531 | spin_lock(&zone->lock); |
e815af95 | 532 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 533 | zone->pages_scanned = 0; |
f2260e6b | 534 | |
5f8dcc21 | 535 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
a6f9edd6 | 536 | while (count) { |
48db57f8 | 537 | struct page *page; |
5f8dcc21 MG |
538 | struct list_head *list; |
539 | ||
540 | /* | |
a6f9edd6 MG |
541 | * Remove pages from lists in a round-robin fashion. A |
542 | * batch_free count is maintained that is incremented when an | |
543 | * empty list is encountered. This is so more pages are freed | |
544 | * off fuller lists instead of spinning excessively around empty | |
545 | * lists | |
5f8dcc21 MG |
546 | */ |
547 | do { | |
a6f9edd6 | 548 | batch_free++; |
5f8dcc21 MG |
549 | if (++migratetype == MIGRATE_PCPTYPES) |
550 | migratetype = 0; | |
551 | list = &pcp->lists[migratetype]; | |
552 | } while (list_empty(list)); | |
48db57f8 | 553 | |
a6f9edd6 MG |
554 | do { |
555 | page = list_entry(list->prev, struct page, lru); | |
556 | /* must delete as __free_one_page list manipulates */ | |
557 | list_del(&page->lru); | |
558 | __free_one_page(page, zone, 0, migratetype); | |
559 | trace_mm_page_pcpu_drain(page, 0, migratetype); | |
560 | } while (--count && --batch_free && !list_empty(list)); | |
1da177e4 | 561 | } |
c54ad30c | 562 | spin_unlock(&zone->lock); |
1da177e4 LT |
563 | } |
564 | ||
ed0ae21d MG |
565 | static void free_one_page(struct zone *zone, struct page *page, int order, |
566 | int migratetype) | |
1da177e4 | 567 | { |
006d22d9 | 568 | spin_lock(&zone->lock); |
e815af95 | 569 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 570 | zone->pages_scanned = 0; |
f2260e6b MG |
571 | |
572 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); | |
ed0ae21d | 573 | __free_one_page(page, zone, order, migratetype); |
006d22d9 | 574 | spin_unlock(&zone->lock); |
48db57f8 NP |
575 | } |
576 | ||
577 | static void __free_pages_ok(struct page *page, unsigned int order) | |
578 | { | |
579 | unsigned long flags; | |
1da177e4 | 580 | int i; |
8cc3b392 | 581 | int bad = 0; |
451ea25d | 582 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 583 | |
b1eeab67 VN |
584 | kmemcheck_free_shadow(page, order); |
585 | ||
1da177e4 | 586 | for (i = 0 ; i < (1 << order) ; ++i) |
8cc3b392 HD |
587 | bad += free_pages_check(page + i); |
588 | if (bad) | |
689bcebf HD |
589 | return; |
590 | ||
3ac7fe5a | 591 | if (!PageHighMem(page)) { |
9858db50 | 592 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
593 | debug_check_no_obj_freed(page_address(page), |
594 | PAGE_SIZE << order); | |
595 | } | |
dafb1367 | 596 | arch_free_page(page, order); |
48db57f8 | 597 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 598 | |
c54ad30c | 599 | local_irq_save(flags); |
c277331d | 600 | if (unlikely(wasMlocked)) |
da456f14 | 601 | free_page_mlock(page); |
f8891e5e | 602 | __count_vm_events(PGFREE, 1 << order); |
ed0ae21d MG |
603 | free_one_page(page_zone(page), page, order, |
604 | get_pageblock_migratetype(page)); | |
c54ad30c | 605 | local_irq_restore(flags); |
1da177e4 LT |
606 | } |
607 | ||
a226f6c8 DH |
608 | /* |
609 | * permit the bootmem allocator to evade page validation on high-order frees | |
610 | */ | |
af370fb8 | 611 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
612 | { |
613 | if (order == 0) { | |
614 | __ClearPageReserved(page); | |
615 | set_page_count(page, 0); | |
7835e98b | 616 | set_page_refcounted(page); |
545b1ea9 | 617 | __free_page(page); |
a226f6c8 | 618 | } else { |
a226f6c8 DH |
619 | int loop; |
620 | ||
545b1ea9 | 621 | prefetchw(page); |
a226f6c8 DH |
622 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
623 | struct page *p = &page[loop]; | |
624 | ||
545b1ea9 NP |
625 | if (loop + 1 < BITS_PER_LONG) |
626 | prefetchw(p + 1); | |
a226f6c8 DH |
627 | __ClearPageReserved(p); |
628 | set_page_count(p, 0); | |
629 | } | |
630 | ||
7835e98b | 631 | set_page_refcounted(page); |
545b1ea9 | 632 | __free_pages(page, order); |
a226f6c8 DH |
633 | } |
634 | } | |
635 | ||
1da177e4 LT |
636 | |
637 | /* | |
638 | * The order of subdivision here is critical for the IO subsystem. | |
639 | * Please do not alter this order without good reasons and regression | |
640 | * testing. Specifically, as large blocks of memory are subdivided, | |
641 | * the order in which smaller blocks are delivered depends on the order | |
642 | * they're subdivided in this function. This is the primary factor | |
643 | * influencing the order in which pages are delivered to the IO | |
644 | * subsystem according to empirical testing, and this is also justified | |
645 | * by considering the behavior of a buddy system containing a single | |
646 | * large block of memory acted on by a series of small allocations. | |
647 | * This behavior is a critical factor in sglist merging's success. | |
648 | * | |
649 | * -- wli | |
650 | */ | |
085cc7d5 | 651 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
652 | int low, int high, struct free_area *area, |
653 | int migratetype) | |
1da177e4 LT |
654 | { |
655 | unsigned long size = 1 << high; | |
656 | ||
657 | while (high > low) { | |
658 | area--; | |
659 | high--; | |
660 | size >>= 1; | |
725d704e | 661 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 662 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
663 | area->nr_free++; |
664 | set_page_order(&page[size], high); | |
665 | } | |
1da177e4 LT |
666 | } |
667 | ||
1da177e4 LT |
668 | /* |
669 | * This page is about to be returned from the page allocator | |
670 | */ | |
2a7684a2 | 671 | static inline int check_new_page(struct page *page) |
1da177e4 | 672 | { |
92be2e33 NP |
673 | if (unlikely(page_mapcount(page) | |
674 | (page->mapping != NULL) | | |
a3af9c38 | 675 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 676 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
224abf92 | 677 | bad_page(page); |
689bcebf | 678 | return 1; |
8cc3b392 | 679 | } |
2a7684a2 WF |
680 | return 0; |
681 | } | |
682 | ||
683 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
684 | { | |
685 | int i; | |
686 | ||
687 | for (i = 0; i < (1 << order); i++) { | |
688 | struct page *p = page + i; | |
689 | if (unlikely(check_new_page(p))) | |
690 | return 1; | |
691 | } | |
689bcebf | 692 | |
4c21e2f2 | 693 | set_page_private(page, 0); |
7835e98b | 694 | set_page_refcounted(page); |
cc102509 NP |
695 | |
696 | arch_alloc_page(page, order); | |
1da177e4 | 697 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
698 | |
699 | if (gfp_flags & __GFP_ZERO) | |
700 | prep_zero_page(page, order, gfp_flags); | |
701 | ||
702 | if (order && (gfp_flags & __GFP_COMP)) | |
703 | prep_compound_page(page, order); | |
704 | ||
689bcebf | 705 | return 0; |
1da177e4 LT |
706 | } |
707 | ||
56fd56b8 MG |
708 | /* |
709 | * Go through the free lists for the given migratetype and remove | |
710 | * the smallest available page from the freelists | |
711 | */ | |
728ec980 MG |
712 | static inline |
713 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
714 | int migratetype) |
715 | { | |
716 | unsigned int current_order; | |
717 | struct free_area * area; | |
718 | struct page *page; | |
719 | ||
720 | /* Find a page of the appropriate size in the preferred list */ | |
721 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
722 | area = &(zone->free_area[current_order]); | |
723 | if (list_empty(&area->free_list[migratetype])) | |
724 | continue; | |
725 | ||
726 | page = list_entry(area->free_list[migratetype].next, | |
727 | struct page, lru); | |
728 | list_del(&page->lru); | |
729 | rmv_page_order(page); | |
730 | area->nr_free--; | |
56fd56b8 MG |
731 | expand(zone, page, order, current_order, area, migratetype); |
732 | return page; | |
733 | } | |
734 | ||
735 | return NULL; | |
736 | } | |
737 | ||
738 | ||
b2a0ac88 MG |
739 | /* |
740 | * This array describes the order lists are fallen back to when | |
741 | * the free lists for the desirable migrate type are depleted | |
742 | */ | |
743 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
744 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
745 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
746 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
747 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
748 | }; |
749 | ||
c361be55 MG |
750 | /* |
751 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 752 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
753 | * boundary. If alignment is required, use move_freepages_block() |
754 | */ | |
b69a7288 AB |
755 | static int move_freepages(struct zone *zone, |
756 | struct page *start_page, struct page *end_page, | |
757 | int migratetype) | |
c361be55 MG |
758 | { |
759 | struct page *page; | |
760 | unsigned long order; | |
d100313f | 761 | int pages_moved = 0; |
c361be55 MG |
762 | |
763 | #ifndef CONFIG_HOLES_IN_ZONE | |
764 | /* | |
765 | * page_zone is not safe to call in this context when | |
766 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
767 | * anyway as we check zone boundaries in move_freepages_block(). | |
768 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 769 | * grouping pages by mobility |
c361be55 MG |
770 | */ |
771 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
772 | #endif | |
773 | ||
774 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
775 | /* Make sure we are not inadvertently changing nodes */ |
776 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
777 | ||
c361be55 MG |
778 | if (!pfn_valid_within(page_to_pfn(page))) { |
779 | page++; | |
780 | continue; | |
781 | } | |
782 | ||
783 | if (!PageBuddy(page)) { | |
784 | page++; | |
785 | continue; | |
786 | } | |
787 | ||
788 | order = page_order(page); | |
789 | list_del(&page->lru); | |
790 | list_add(&page->lru, | |
791 | &zone->free_area[order].free_list[migratetype]); | |
792 | page += 1 << order; | |
d100313f | 793 | pages_moved += 1 << order; |
c361be55 MG |
794 | } |
795 | ||
d100313f | 796 | return pages_moved; |
c361be55 MG |
797 | } |
798 | ||
b69a7288 AB |
799 | static int move_freepages_block(struct zone *zone, struct page *page, |
800 | int migratetype) | |
c361be55 MG |
801 | { |
802 | unsigned long start_pfn, end_pfn; | |
803 | struct page *start_page, *end_page; | |
804 | ||
805 | start_pfn = page_to_pfn(page); | |
d9c23400 | 806 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 807 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
808 | end_page = start_page + pageblock_nr_pages - 1; |
809 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
810 | |
811 | /* Do not cross zone boundaries */ | |
812 | if (start_pfn < zone->zone_start_pfn) | |
813 | start_page = page; | |
814 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
815 | return 0; | |
816 | ||
817 | return move_freepages(zone, start_page, end_page, migratetype); | |
818 | } | |
819 | ||
2f66a68f MG |
820 | static void change_pageblock_range(struct page *pageblock_page, |
821 | int start_order, int migratetype) | |
822 | { | |
823 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
824 | ||
825 | while (nr_pageblocks--) { | |
826 | set_pageblock_migratetype(pageblock_page, migratetype); | |
827 | pageblock_page += pageblock_nr_pages; | |
828 | } | |
829 | } | |
830 | ||
b2a0ac88 | 831 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
832 | static inline struct page * |
833 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
834 | { |
835 | struct free_area * area; | |
836 | int current_order; | |
837 | struct page *page; | |
838 | int migratetype, i; | |
839 | ||
840 | /* Find the largest possible block of pages in the other list */ | |
841 | for (current_order = MAX_ORDER-1; current_order >= order; | |
842 | --current_order) { | |
843 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
844 | migratetype = fallbacks[start_migratetype][i]; | |
845 | ||
56fd56b8 MG |
846 | /* MIGRATE_RESERVE handled later if necessary */ |
847 | if (migratetype == MIGRATE_RESERVE) | |
848 | continue; | |
e010487d | 849 | |
b2a0ac88 MG |
850 | area = &(zone->free_area[current_order]); |
851 | if (list_empty(&area->free_list[migratetype])) | |
852 | continue; | |
853 | ||
854 | page = list_entry(area->free_list[migratetype].next, | |
855 | struct page, lru); | |
856 | area->nr_free--; | |
857 | ||
858 | /* | |
c361be55 | 859 | * If breaking a large block of pages, move all free |
46dafbca MG |
860 | * pages to the preferred allocation list. If falling |
861 | * back for a reclaimable kernel allocation, be more | |
862 | * agressive about taking ownership of free pages | |
b2a0ac88 | 863 | */ |
d9c23400 | 864 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
dd5d241e MG |
865 | start_migratetype == MIGRATE_RECLAIMABLE || |
866 | page_group_by_mobility_disabled) { | |
46dafbca MG |
867 | unsigned long pages; |
868 | pages = move_freepages_block(zone, page, | |
869 | start_migratetype); | |
870 | ||
871 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
872 | if (pages >= (1 << (pageblock_order-1)) || |
873 | page_group_by_mobility_disabled) | |
46dafbca MG |
874 | set_pageblock_migratetype(page, |
875 | start_migratetype); | |
876 | ||
b2a0ac88 | 877 | migratetype = start_migratetype; |
c361be55 | 878 | } |
b2a0ac88 MG |
879 | |
880 | /* Remove the page from the freelists */ | |
881 | list_del(&page->lru); | |
882 | rmv_page_order(page); | |
b2a0ac88 | 883 | |
2f66a68f MG |
884 | /* Take ownership for orders >= pageblock_order */ |
885 | if (current_order >= pageblock_order) | |
886 | change_pageblock_range(page, current_order, | |
b2a0ac88 MG |
887 | start_migratetype); |
888 | ||
889 | expand(zone, page, order, current_order, area, migratetype); | |
e0fff1bd MG |
890 | |
891 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
892 | start_migratetype, migratetype); | |
893 | ||
b2a0ac88 MG |
894 | return page; |
895 | } | |
896 | } | |
897 | ||
728ec980 | 898 | return NULL; |
b2a0ac88 MG |
899 | } |
900 | ||
56fd56b8 | 901 | /* |
1da177e4 LT |
902 | * Do the hard work of removing an element from the buddy allocator. |
903 | * Call me with the zone->lock already held. | |
904 | */ | |
b2a0ac88 MG |
905 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
906 | int migratetype) | |
1da177e4 | 907 | { |
1da177e4 LT |
908 | struct page *page; |
909 | ||
728ec980 | 910 | retry_reserve: |
56fd56b8 | 911 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 912 | |
728ec980 | 913 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 914 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 915 | |
728ec980 MG |
916 | /* |
917 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
918 | * is used because __rmqueue_smallest is an inline function | |
919 | * and we want just one call site | |
920 | */ | |
921 | if (!page) { | |
922 | migratetype = MIGRATE_RESERVE; | |
923 | goto retry_reserve; | |
924 | } | |
925 | } | |
926 | ||
0d3d062a | 927 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 928 | return page; |
1da177e4 LT |
929 | } |
930 | ||
931 | /* | |
932 | * Obtain a specified number of elements from the buddy allocator, all under | |
933 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
934 | * Returns the number of new pages which were placed at *list. | |
935 | */ | |
936 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 | 937 | unsigned long count, struct list_head *list, |
e084b2d9 | 938 | int migratetype, int cold) |
1da177e4 | 939 | { |
1da177e4 | 940 | int i; |
1da177e4 | 941 | |
c54ad30c | 942 | spin_lock(&zone->lock); |
1da177e4 | 943 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 944 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 945 | if (unlikely(page == NULL)) |
1da177e4 | 946 | break; |
81eabcbe MG |
947 | |
948 | /* | |
949 | * Split buddy pages returned by expand() are received here | |
950 | * in physical page order. The page is added to the callers and | |
951 | * list and the list head then moves forward. From the callers | |
952 | * perspective, the linked list is ordered by page number in | |
953 | * some conditions. This is useful for IO devices that can | |
954 | * merge IO requests if the physical pages are ordered | |
955 | * properly. | |
956 | */ | |
e084b2d9 MG |
957 | if (likely(cold == 0)) |
958 | list_add(&page->lru, list); | |
959 | else | |
960 | list_add_tail(&page->lru, list); | |
535131e6 | 961 | set_page_private(page, migratetype); |
81eabcbe | 962 | list = &page->lru; |
1da177e4 | 963 | } |
f2260e6b | 964 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 965 | spin_unlock(&zone->lock); |
085cc7d5 | 966 | return i; |
1da177e4 LT |
967 | } |
968 | ||
4ae7c039 | 969 | #ifdef CONFIG_NUMA |
8fce4d8e | 970 | /* |
4037d452 CL |
971 | * Called from the vmstat counter updater to drain pagesets of this |
972 | * currently executing processor on remote nodes after they have | |
973 | * expired. | |
974 | * | |
879336c3 CL |
975 | * Note that this function must be called with the thread pinned to |
976 | * a single processor. | |
8fce4d8e | 977 | */ |
4037d452 | 978 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 979 | { |
4ae7c039 | 980 | unsigned long flags; |
4037d452 | 981 | int to_drain; |
4ae7c039 | 982 | |
4037d452 CL |
983 | local_irq_save(flags); |
984 | if (pcp->count >= pcp->batch) | |
985 | to_drain = pcp->batch; | |
986 | else | |
987 | to_drain = pcp->count; | |
5f8dcc21 | 988 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 CL |
989 | pcp->count -= to_drain; |
990 | local_irq_restore(flags); | |
4ae7c039 CL |
991 | } |
992 | #endif | |
993 | ||
9f8f2172 CL |
994 | /* |
995 | * Drain pages of the indicated processor. | |
996 | * | |
997 | * The processor must either be the current processor and the | |
998 | * thread pinned to the current processor or a processor that | |
999 | * is not online. | |
1000 | */ | |
1001 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1002 | { |
c54ad30c | 1003 | unsigned long flags; |
1da177e4 | 1004 | struct zone *zone; |
1da177e4 | 1005 | |
ee99c71c | 1006 | for_each_populated_zone(zone) { |
1da177e4 | 1007 | struct per_cpu_pageset *pset; |
3dfa5721 | 1008 | struct per_cpu_pages *pcp; |
1da177e4 | 1009 | |
e7c8d5c9 | 1010 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
1011 | |
1012 | pcp = &pset->pcp; | |
1013 | local_irq_save(flags); | |
5f8dcc21 | 1014 | free_pcppages_bulk(zone, pcp->count, pcp); |
3dfa5721 CL |
1015 | pcp->count = 0; |
1016 | local_irq_restore(flags); | |
1da177e4 LT |
1017 | } |
1018 | } | |
1da177e4 | 1019 | |
9f8f2172 CL |
1020 | /* |
1021 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1022 | */ | |
1023 | void drain_local_pages(void *arg) | |
1024 | { | |
1025 | drain_pages(smp_processor_id()); | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1030 | */ | |
1031 | void drain_all_pages(void) | |
1032 | { | |
15c8b6c1 | 1033 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
1034 | } |
1035 | ||
296699de | 1036 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1037 | |
1038 | void mark_free_pages(struct zone *zone) | |
1039 | { | |
f623f0db RW |
1040 | unsigned long pfn, max_zone_pfn; |
1041 | unsigned long flags; | |
b2a0ac88 | 1042 | int order, t; |
1da177e4 LT |
1043 | struct list_head *curr; |
1044 | ||
1045 | if (!zone->spanned_pages) | |
1046 | return; | |
1047 | ||
1048 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1049 | |
1050 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1051 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1052 | if (pfn_valid(pfn)) { | |
1053 | struct page *page = pfn_to_page(pfn); | |
1054 | ||
7be98234 RW |
1055 | if (!swsusp_page_is_forbidden(page)) |
1056 | swsusp_unset_page_free(page); | |
f623f0db | 1057 | } |
1da177e4 | 1058 | |
b2a0ac88 MG |
1059 | for_each_migratetype_order(order, t) { |
1060 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1061 | unsigned long i; |
1da177e4 | 1062 | |
f623f0db RW |
1063 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1064 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1065 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1066 | } |
b2a0ac88 | 1067 | } |
1da177e4 LT |
1068 | spin_unlock_irqrestore(&zone->lock, flags); |
1069 | } | |
e2c55dc8 | 1070 | #endif /* CONFIG_PM */ |
1da177e4 | 1071 | |
1da177e4 LT |
1072 | /* |
1073 | * Free a 0-order page | |
1074 | */ | |
920c7a5d | 1075 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1076 | { |
1077 | struct zone *zone = page_zone(page); | |
1078 | struct per_cpu_pages *pcp; | |
1079 | unsigned long flags; | |
5f8dcc21 | 1080 | int migratetype; |
451ea25d | 1081 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 1082 | |
b1eeab67 VN |
1083 | kmemcheck_free_shadow(page, 0); |
1084 | ||
1da177e4 LT |
1085 | if (PageAnon(page)) |
1086 | page->mapping = NULL; | |
224abf92 | 1087 | if (free_pages_check(page)) |
689bcebf HD |
1088 | return; |
1089 | ||
3ac7fe5a | 1090 | if (!PageHighMem(page)) { |
9858db50 | 1091 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); |
3ac7fe5a TG |
1092 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); |
1093 | } | |
dafb1367 | 1094 | arch_free_page(page, 0); |
689bcebf HD |
1095 | kernel_map_pages(page, 1, 0); |
1096 | ||
3dfa5721 | 1097 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
5f8dcc21 MG |
1098 | migratetype = get_pageblock_migratetype(page); |
1099 | set_page_private(page, migratetype); | |
1da177e4 | 1100 | local_irq_save(flags); |
c277331d | 1101 | if (unlikely(wasMlocked)) |
da456f14 | 1102 | free_page_mlock(page); |
f8891e5e | 1103 | __count_vm_event(PGFREE); |
da456f14 | 1104 | |
5f8dcc21 MG |
1105 | /* |
1106 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1107 | * Free ISOLATE pages back to the allocator because they are being | |
1108 | * offlined but treat RESERVE as movable pages so we can get those | |
1109 | * areas back if necessary. Otherwise, we may have to free | |
1110 | * excessively into the page allocator | |
1111 | */ | |
1112 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1113 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1114 | free_one_page(zone, page, 0, migratetype); | |
1115 | goto out; | |
1116 | } | |
1117 | migratetype = MIGRATE_MOVABLE; | |
1118 | } | |
1119 | ||
3dfa5721 | 1120 | if (cold) |
5f8dcc21 | 1121 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1122 | else |
5f8dcc21 | 1123 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1124 | pcp->count++; |
48db57f8 | 1125 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1126 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1127 | pcp->count -= pcp->batch; |
1128 | } | |
5f8dcc21 MG |
1129 | |
1130 | out: | |
1da177e4 LT |
1131 | local_irq_restore(flags); |
1132 | put_cpu(); | |
1133 | } | |
1134 | ||
920c7a5d | 1135 | void free_hot_page(struct page *page) |
1da177e4 | 1136 | { |
4b4f278c | 1137 | trace_mm_page_free_direct(page, 0); |
1da177e4 LT |
1138 | free_hot_cold_page(page, 0); |
1139 | } | |
1140 | ||
8dfcc9ba NP |
1141 | /* |
1142 | * split_page takes a non-compound higher-order page, and splits it into | |
1143 | * n (1<<order) sub-pages: page[0..n] | |
1144 | * Each sub-page must be freed individually. | |
1145 | * | |
1146 | * Note: this is probably too low level an operation for use in drivers. | |
1147 | * Please consult with lkml before using this in your driver. | |
1148 | */ | |
1149 | void split_page(struct page *page, unsigned int order) | |
1150 | { | |
1151 | int i; | |
1152 | ||
725d704e NP |
1153 | VM_BUG_ON(PageCompound(page)); |
1154 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1155 | |
1156 | #ifdef CONFIG_KMEMCHECK | |
1157 | /* | |
1158 | * Split shadow pages too, because free(page[0]) would | |
1159 | * otherwise free the whole shadow. | |
1160 | */ | |
1161 | if (kmemcheck_page_is_tracked(page)) | |
1162 | split_page(virt_to_page(page[0].shadow), order); | |
1163 | #endif | |
1164 | ||
7835e98b NP |
1165 | for (i = 1; i < (1 << order); i++) |
1166 | set_page_refcounted(page + i); | |
8dfcc9ba | 1167 | } |
8dfcc9ba | 1168 | |
1da177e4 LT |
1169 | /* |
1170 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1171 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1172 | * or two. | |
1173 | */ | |
0a15c3e9 MG |
1174 | static inline |
1175 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1176 | struct zone *zone, int order, gfp_t gfp_flags, |
1177 | int migratetype) | |
1da177e4 LT |
1178 | { |
1179 | unsigned long flags; | |
689bcebf | 1180 | struct page *page; |
1da177e4 | 1181 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1182 | int cpu; |
1da177e4 | 1183 | |
689bcebf | 1184 | again: |
a74609fa | 1185 | cpu = get_cpu(); |
48db57f8 | 1186 | if (likely(order == 0)) { |
1da177e4 | 1187 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1188 | struct list_head *list; |
1da177e4 | 1189 | |
3dfa5721 | 1190 | pcp = &zone_pcp(zone, cpu)->pcp; |
5f8dcc21 | 1191 | list = &pcp->lists[migratetype]; |
1da177e4 | 1192 | local_irq_save(flags); |
5f8dcc21 | 1193 | if (list_empty(list)) { |
535131e6 | 1194 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1195 | pcp->batch, list, |
e084b2d9 | 1196 | migratetype, cold); |
5f8dcc21 | 1197 | if (unlikely(list_empty(list))) |
6fb332fa | 1198 | goto failed; |
535131e6 | 1199 | } |
b92a6edd | 1200 | |
5f8dcc21 MG |
1201 | if (cold) |
1202 | page = list_entry(list->prev, struct page, lru); | |
1203 | else | |
1204 | page = list_entry(list->next, struct page, lru); | |
1205 | ||
b92a6edd MG |
1206 | list_del(&page->lru); |
1207 | pcp->count--; | |
7fb1d9fc | 1208 | } else { |
dab48dab AM |
1209 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1210 | /* | |
1211 | * __GFP_NOFAIL is not to be used in new code. | |
1212 | * | |
1213 | * All __GFP_NOFAIL callers should be fixed so that they | |
1214 | * properly detect and handle allocation failures. | |
1215 | * | |
1216 | * We most definitely don't want callers attempting to | |
4923abf9 | 1217 | * allocate greater than order-1 page units with |
dab48dab AM |
1218 | * __GFP_NOFAIL. |
1219 | */ | |
4923abf9 | 1220 | WARN_ON_ONCE(order > 1); |
dab48dab | 1221 | } |
1da177e4 | 1222 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1223 | page = __rmqueue(zone, order, migratetype); |
f2260e6b | 1224 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
a74609fa NP |
1225 | spin_unlock(&zone->lock); |
1226 | if (!page) | |
1227 | goto failed; | |
1da177e4 LT |
1228 | } |
1229 | ||
f8891e5e | 1230 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1231 | zone_statistics(preferred_zone, zone); |
a74609fa NP |
1232 | local_irq_restore(flags); |
1233 | put_cpu(); | |
1da177e4 | 1234 | |
725d704e | 1235 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1236 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1237 | goto again; |
1da177e4 | 1238 | return page; |
a74609fa NP |
1239 | |
1240 | failed: | |
1241 | local_irq_restore(flags); | |
1242 | put_cpu(); | |
1243 | return NULL; | |
1da177e4 LT |
1244 | } |
1245 | ||
41858966 MG |
1246 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
1247 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1248 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1249 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1250 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1251 | ||
1252 | /* Mask to get the watermark bits */ | |
1253 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1254 | ||
3148890b NP |
1255 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
1256 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1257 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1258 | |
933e312e AM |
1259 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1260 | ||
1261 | static struct fail_page_alloc_attr { | |
1262 | struct fault_attr attr; | |
1263 | ||
1264 | u32 ignore_gfp_highmem; | |
1265 | u32 ignore_gfp_wait; | |
54114994 | 1266 | u32 min_order; |
933e312e AM |
1267 | |
1268 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1269 | ||
1270 | struct dentry *ignore_gfp_highmem_file; | |
1271 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1272 | struct dentry *min_order_file; |
933e312e AM |
1273 | |
1274 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1275 | ||
1276 | } fail_page_alloc = { | |
1277 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1278 | .ignore_gfp_wait = 1, |
1279 | .ignore_gfp_highmem = 1, | |
54114994 | 1280 | .min_order = 1, |
933e312e AM |
1281 | }; |
1282 | ||
1283 | static int __init setup_fail_page_alloc(char *str) | |
1284 | { | |
1285 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1286 | } | |
1287 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1288 | ||
1289 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1290 | { | |
54114994 AM |
1291 | if (order < fail_page_alloc.min_order) |
1292 | return 0; | |
933e312e AM |
1293 | if (gfp_mask & __GFP_NOFAIL) |
1294 | return 0; | |
1295 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1296 | return 0; | |
1297 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1298 | return 0; | |
1299 | ||
1300 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1301 | } | |
1302 | ||
1303 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1304 | ||
1305 | static int __init fail_page_alloc_debugfs(void) | |
1306 | { | |
1307 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1308 | struct dentry *dir; | |
1309 | int err; | |
1310 | ||
1311 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1312 | "fail_page_alloc"); | |
1313 | if (err) | |
1314 | return err; | |
1315 | dir = fail_page_alloc.attr.dentries.dir; | |
1316 | ||
1317 | fail_page_alloc.ignore_gfp_wait_file = | |
1318 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1319 | &fail_page_alloc.ignore_gfp_wait); | |
1320 | ||
1321 | fail_page_alloc.ignore_gfp_highmem_file = | |
1322 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1323 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1324 | fail_page_alloc.min_order_file = |
1325 | debugfs_create_u32("min-order", mode, dir, | |
1326 | &fail_page_alloc.min_order); | |
933e312e AM |
1327 | |
1328 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1329 | !fail_page_alloc.ignore_gfp_highmem_file || |
1330 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1331 | err = -ENOMEM; |
1332 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1333 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1334 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1335 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1336 | } | |
1337 | ||
1338 | return err; | |
1339 | } | |
1340 | ||
1341 | late_initcall(fail_page_alloc_debugfs); | |
1342 | ||
1343 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1344 | ||
1345 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1346 | ||
1347 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1348 | { | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1353 | ||
1da177e4 LT |
1354 | /* |
1355 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1356 | * of the allocation. | |
1357 | */ | |
1358 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1359 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1360 | { |
1361 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1362 | long min = mark; |
1363 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1364 | int o; |
1365 | ||
7fb1d9fc | 1366 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1367 | min -= min / 2; |
7fb1d9fc | 1368 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1369 | min -= min / 4; |
1370 | ||
1371 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1372 | return 0; | |
1373 | for (o = 0; o < order; o++) { | |
1374 | /* At the next order, this order's pages become unavailable */ | |
1375 | free_pages -= z->free_area[o].nr_free << o; | |
1376 | ||
1377 | /* Require fewer higher order pages to be free */ | |
1378 | min >>= 1; | |
1379 | ||
1380 | if (free_pages <= min) | |
1381 | return 0; | |
1382 | } | |
1383 | return 1; | |
1384 | } | |
1385 | ||
9276b1bc PJ |
1386 | #ifdef CONFIG_NUMA |
1387 | /* | |
1388 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1389 | * skip over zones that are not allowed by the cpuset, or that have | |
1390 | * been recently (in last second) found to be nearly full. See further | |
1391 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1392 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1393 | * |
1394 | * If the zonelist cache is present in the passed in zonelist, then | |
1395 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1396 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1397 | * |
1398 | * If the zonelist cache is not available for this zonelist, does | |
1399 | * nothing and returns NULL. | |
1400 | * | |
1401 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1402 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1403 | * | |
1404 | * We hold off even calling zlc_setup, until after we've checked the | |
1405 | * first zone in the zonelist, on the theory that most allocations will | |
1406 | * be satisfied from that first zone, so best to examine that zone as | |
1407 | * quickly as we can. | |
1408 | */ | |
1409 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1410 | { | |
1411 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1412 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1413 | ||
1414 | zlc = zonelist->zlcache_ptr; | |
1415 | if (!zlc) | |
1416 | return NULL; | |
1417 | ||
f05111f5 | 1418 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1419 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1420 | zlc->last_full_zap = jiffies; | |
1421 | } | |
1422 | ||
1423 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1424 | &cpuset_current_mems_allowed : | |
37b07e41 | 1425 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1426 | return allowednodes; |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1431 | * if it is worth looking at further for free memory: | |
1432 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1433 | * bit set in the zonelist_cache fullzones BITMAP). | |
1434 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1435 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1436 | * Return true (non-zero) if zone is worth looking at further, or | |
1437 | * else return false (zero) if it is not. | |
1438 | * | |
1439 | * This check -ignores- the distinction between various watermarks, | |
1440 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1441 | * found to be full for any variation of these watermarks, it will | |
1442 | * be considered full for up to one second by all requests, unless | |
1443 | * we are so low on memory on all allowed nodes that we are forced | |
1444 | * into the second scan of the zonelist. | |
1445 | * | |
1446 | * In the second scan we ignore this zonelist cache and exactly | |
1447 | * apply the watermarks to all zones, even it is slower to do so. | |
1448 | * We are low on memory in the second scan, and should leave no stone | |
1449 | * unturned looking for a free page. | |
1450 | */ | |
dd1a239f | 1451 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1452 | nodemask_t *allowednodes) |
1453 | { | |
1454 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1455 | int i; /* index of *z in zonelist zones */ | |
1456 | int n; /* node that zone *z is on */ | |
1457 | ||
1458 | zlc = zonelist->zlcache_ptr; | |
1459 | if (!zlc) | |
1460 | return 1; | |
1461 | ||
dd1a239f | 1462 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1463 | n = zlc->z_to_n[i]; |
1464 | ||
1465 | /* This zone is worth trying if it is allowed but not full */ | |
1466 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1471 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1472 | * from that zone don't waste time re-examining it. | |
1473 | */ | |
dd1a239f | 1474 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1475 | { |
1476 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1477 | int i; /* index of *z in zonelist zones */ | |
1478 | ||
1479 | zlc = zonelist->zlcache_ptr; | |
1480 | if (!zlc) | |
1481 | return; | |
1482 | ||
dd1a239f | 1483 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1484 | |
1485 | set_bit(i, zlc->fullzones); | |
1486 | } | |
1487 | ||
1488 | #else /* CONFIG_NUMA */ | |
1489 | ||
1490 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1491 | { | |
1492 | return NULL; | |
1493 | } | |
1494 | ||
dd1a239f | 1495 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1496 | nodemask_t *allowednodes) |
1497 | { | |
1498 | return 1; | |
1499 | } | |
1500 | ||
dd1a239f | 1501 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1502 | { |
1503 | } | |
1504 | #endif /* CONFIG_NUMA */ | |
1505 | ||
7fb1d9fc | 1506 | /* |
0798e519 | 1507 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1508 | * a page. |
1509 | */ | |
1510 | static struct page * | |
19770b32 | 1511 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1512 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1513 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1514 | { |
dd1a239f | 1515 | struct zoneref *z; |
7fb1d9fc | 1516 | struct page *page = NULL; |
54a6eb5c | 1517 | int classzone_idx; |
5117f45d | 1518 | struct zone *zone; |
9276b1bc PJ |
1519 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1520 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1521 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1522 | |
19770b32 | 1523 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1524 | zonelist_scan: |
7fb1d9fc | 1525 | /* |
9276b1bc | 1526 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1527 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1528 | */ | |
19770b32 MG |
1529 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1530 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1531 | if (NUMA_BUILD && zlc_active && |
1532 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1533 | continue; | |
7fb1d9fc | 1534 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1535 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1536 | goto try_next_zone; |
7fb1d9fc | 1537 | |
41858966 | 1538 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1539 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1540 | unsigned long mark; |
fa5e084e MG |
1541 | int ret; |
1542 | ||
41858966 | 1543 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1544 | if (zone_watermark_ok(zone, order, mark, |
1545 | classzone_idx, alloc_flags)) | |
1546 | goto try_this_zone; | |
1547 | ||
1548 | if (zone_reclaim_mode == 0) | |
1549 | goto this_zone_full; | |
1550 | ||
1551 | ret = zone_reclaim(zone, gfp_mask, order); | |
1552 | switch (ret) { | |
1553 | case ZONE_RECLAIM_NOSCAN: | |
1554 | /* did not scan */ | |
1555 | goto try_next_zone; | |
1556 | case ZONE_RECLAIM_FULL: | |
1557 | /* scanned but unreclaimable */ | |
1558 | goto this_zone_full; | |
1559 | default: | |
1560 | /* did we reclaim enough */ | |
1561 | if (!zone_watermark_ok(zone, order, mark, | |
1562 | classzone_idx, alloc_flags)) | |
9276b1bc | 1563 | goto this_zone_full; |
0798e519 | 1564 | } |
7fb1d9fc RS |
1565 | } |
1566 | ||
fa5e084e | 1567 | try_this_zone: |
3dd28266 MG |
1568 | page = buffered_rmqueue(preferred_zone, zone, order, |
1569 | gfp_mask, migratetype); | |
0798e519 | 1570 | if (page) |
7fb1d9fc | 1571 | break; |
9276b1bc PJ |
1572 | this_zone_full: |
1573 | if (NUMA_BUILD) | |
1574 | zlc_mark_zone_full(zonelist, z); | |
1575 | try_next_zone: | |
62bc62a8 | 1576 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
d395b734 MG |
1577 | /* |
1578 | * we do zlc_setup after the first zone is tried but only | |
1579 | * if there are multiple nodes make it worthwhile | |
1580 | */ | |
9276b1bc PJ |
1581 | allowednodes = zlc_setup(zonelist, alloc_flags); |
1582 | zlc_active = 1; | |
1583 | did_zlc_setup = 1; | |
1584 | } | |
54a6eb5c | 1585 | } |
9276b1bc PJ |
1586 | |
1587 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1588 | /* Disable zlc cache for second zonelist scan */ | |
1589 | zlc_active = 0; | |
1590 | goto zonelist_scan; | |
1591 | } | |
7fb1d9fc | 1592 | return page; |
753ee728 MH |
1593 | } |
1594 | ||
11e33f6a MG |
1595 | static inline int |
1596 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1597 | unsigned long pages_reclaimed) | |
1da177e4 | 1598 | { |
11e33f6a MG |
1599 | /* Do not loop if specifically requested */ |
1600 | if (gfp_mask & __GFP_NORETRY) | |
1601 | return 0; | |
1da177e4 | 1602 | |
11e33f6a MG |
1603 | /* |
1604 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1605 | * means __GFP_NOFAIL, but that may not be true in other | |
1606 | * implementations. | |
1607 | */ | |
1608 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1609 | return 1; | |
1610 | ||
1611 | /* | |
1612 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1613 | * specified, then we retry until we no longer reclaim any pages | |
1614 | * (above), or we've reclaimed an order of pages at least as | |
1615 | * large as the allocation's order. In both cases, if the | |
1616 | * allocation still fails, we stop retrying. | |
1617 | */ | |
1618 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1619 | return 1; | |
cf40bd16 | 1620 | |
11e33f6a MG |
1621 | /* |
1622 | * Don't let big-order allocations loop unless the caller | |
1623 | * explicitly requests that. | |
1624 | */ | |
1625 | if (gfp_mask & __GFP_NOFAIL) | |
1626 | return 1; | |
1da177e4 | 1627 | |
11e33f6a MG |
1628 | return 0; |
1629 | } | |
933e312e | 1630 | |
11e33f6a MG |
1631 | static inline struct page * |
1632 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1633 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1634 | nodemask_t *nodemask, struct zone *preferred_zone, |
1635 | int migratetype) | |
11e33f6a MG |
1636 | { |
1637 | struct page *page; | |
1638 | ||
1639 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
1640 | if (!try_set_zone_oom(zonelist, gfp_mask)) { | |
1641 | schedule_timeout_uninterruptible(1); | |
1da177e4 LT |
1642 | return NULL; |
1643 | } | |
6b1de916 | 1644 | |
11e33f6a MG |
1645 | /* |
1646 | * Go through the zonelist yet one more time, keep very high watermark | |
1647 | * here, this is only to catch a parallel oom killing, we must fail if | |
1648 | * we're still under heavy pressure. | |
1649 | */ | |
1650 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1651 | order, zonelist, high_zoneidx, | |
5117f45d | 1652 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 1653 | preferred_zone, migratetype); |
7fb1d9fc | 1654 | if (page) |
11e33f6a MG |
1655 | goto out; |
1656 | ||
1657 | /* The OOM killer will not help higher order allocs */ | |
82553a93 | 1658 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL)) |
11e33f6a MG |
1659 | goto out; |
1660 | ||
1661 | /* Exhausted what can be done so it's blamo time */ | |
1662 | out_of_memory(zonelist, gfp_mask, order); | |
1663 | ||
1664 | out: | |
1665 | clear_zonelist_oom(zonelist, gfp_mask); | |
1666 | return page; | |
1667 | } | |
1668 | ||
1669 | /* The really slow allocator path where we enter direct reclaim */ | |
1670 | static inline struct page * | |
1671 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1672 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
5117f45d | 1673 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
3dd28266 | 1674 | int migratetype, unsigned long *did_some_progress) |
11e33f6a MG |
1675 | { |
1676 | struct page *page = NULL; | |
1677 | struct reclaim_state reclaim_state; | |
1678 | struct task_struct *p = current; | |
1679 | ||
1680 | cond_resched(); | |
1681 | ||
1682 | /* We now go into synchronous reclaim */ | |
1683 | cpuset_memory_pressure_bump(); | |
11e33f6a MG |
1684 | p->flags |= PF_MEMALLOC; |
1685 | lockdep_set_current_reclaim_state(gfp_mask); | |
1686 | reclaim_state.reclaimed_slab = 0; | |
1687 | p->reclaim_state = &reclaim_state; | |
1688 | ||
1689 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1690 | ||
1691 | p->reclaim_state = NULL; | |
1692 | lockdep_clear_current_reclaim_state(); | |
1693 | p->flags &= ~PF_MEMALLOC; | |
1694 | ||
1695 | cond_resched(); | |
1696 | ||
1697 | if (order != 0) | |
1698 | drain_all_pages(); | |
1699 | ||
1700 | if (likely(*did_some_progress)) | |
1701 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1702 | zonelist, high_zoneidx, |
3dd28266 MG |
1703 | alloc_flags, preferred_zone, |
1704 | migratetype); | |
11e33f6a MG |
1705 | return page; |
1706 | } | |
1707 | ||
1da177e4 | 1708 | /* |
11e33f6a MG |
1709 | * This is called in the allocator slow-path if the allocation request is of |
1710 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 1711 | */ |
11e33f6a MG |
1712 | static inline struct page * |
1713 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1714 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1715 | nodemask_t *nodemask, struct zone *preferred_zone, |
1716 | int migratetype) | |
11e33f6a MG |
1717 | { |
1718 | struct page *page; | |
1719 | ||
1720 | do { | |
1721 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1722 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 1723 | preferred_zone, migratetype); |
11e33f6a MG |
1724 | |
1725 | if (!page && gfp_mask & __GFP_NOFAIL) | |
8aa7e847 | 1726 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
1727 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
1728 | ||
1729 | return page; | |
1730 | } | |
1731 | ||
1732 | static inline | |
1733 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
1734 | enum zone_type high_zoneidx) | |
1da177e4 | 1735 | { |
dd1a239f MG |
1736 | struct zoneref *z; |
1737 | struct zone *zone; | |
1da177e4 | 1738 | |
11e33f6a MG |
1739 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
1740 | wakeup_kswapd(zone, order); | |
1741 | } | |
cf40bd16 | 1742 | |
341ce06f PZ |
1743 | static inline int |
1744 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
1745 | { | |
1746 | struct task_struct *p = current; | |
1747 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; | |
1748 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 1749 | |
a56f57ff MG |
1750 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
1751 | BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH); | |
933e312e | 1752 | |
341ce06f PZ |
1753 | /* |
1754 | * The caller may dip into page reserves a bit more if the caller | |
1755 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1756 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1757 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1758 | */ | |
a56f57ff | 1759 | alloc_flags |= (gfp_mask & __GFP_HIGH); |
1da177e4 | 1760 | |
341ce06f PZ |
1761 | if (!wait) { |
1762 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 1763 | /* |
341ce06f PZ |
1764 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
1765 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 1766 | */ |
341ce06f | 1767 | alloc_flags &= ~ALLOC_CPUSET; |
9d0ed60f | 1768 | } else if (unlikely(rt_task(p)) && !in_interrupt()) |
341ce06f PZ |
1769 | alloc_flags |= ALLOC_HARDER; |
1770 | ||
1771 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
1772 | if (!in_interrupt() && | |
1773 | ((p->flags & PF_MEMALLOC) || | |
1774 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
1775 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1da177e4 | 1776 | } |
6b1de916 | 1777 | |
341ce06f PZ |
1778 | return alloc_flags; |
1779 | } | |
1780 | ||
11e33f6a MG |
1781 | static inline struct page * |
1782 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
1783 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1784 | nodemask_t *nodemask, struct zone *preferred_zone, |
1785 | int migratetype) | |
11e33f6a MG |
1786 | { |
1787 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1788 | struct page *page = NULL; | |
1789 | int alloc_flags; | |
1790 | unsigned long pages_reclaimed = 0; | |
1791 | unsigned long did_some_progress; | |
1792 | struct task_struct *p = current; | |
1da177e4 | 1793 | |
72807a74 MG |
1794 | /* |
1795 | * In the slowpath, we sanity check order to avoid ever trying to | |
1796 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
1797 | * be using allocators in order of preference for an area that is | |
1798 | * too large. | |
1799 | */ | |
1fc28b70 MG |
1800 | if (order >= MAX_ORDER) { |
1801 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 1802 | return NULL; |
1fc28b70 | 1803 | } |
1da177e4 | 1804 | |
952f3b51 CL |
1805 | /* |
1806 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1807 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1808 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1809 | * using a larger set of nodes after it has established that the | |
1810 | * allowed per node queues are empty and that nodes are | |
1811 | * over allocated. | |
1812 | */ | |
1813 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1814 | goto nopage; | |
1815 | ||
cc4a6851 | 1816 | restart: |
11e33f6a | 1817 | wake_all_kswapd(order, zonelist, high_zoneidx); |
1da177e4 | 1818 | |
9bf2229f | 1819 | /* |
7fb1d9fc RS |
1820 | * OK, we're below the kswapd watermark and have kicked background |
1821 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1822 | * to how we want to proceed. | |
9bf2229f | 1823 | */ |
341ce06f | 1824 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 1825 | |
341ce06f | 1826 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 1827 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
1828 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
1829 | preferred_zone, migratetype); | |
7fb1d9fc RS |
1830 | if (page) |
1831 | goto got_pg; | |
1da177e4 | 1832 | |
b43a57bb | 1833 | rebalance: |
11e33f6a | 1834 | /* Allocate without watermarks if the context allows */ |
341ce06f PZ |
1835 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
1836 | page = __alloc_pages_high_priority(gfp_mask, order, | |
1837 | zonelist, high_zoneidx, nodemask, | |
1838 | preferred_zone, migratetype); | |
1839 | if (page) | |
1840 | goto got_pg; | |
1da177e4 LT |
1841 | } |
1842 | ||
1843 | /* Atomic allocations - we can't balance anything */ | |
1844 | if (!wait) | |
1845 | goto nopage; | |
1846 | ||
341ce06f PZ |
1847 | /* Avoid recursion of direct reclaim */ |
1848 | if (p->flags & PF_MEMALLOC) | |
1849 | goto nopage; | |
1850 | ||
6583bb64 DR |
1851 | /* Avoid allocations with no watermarks from looping endlessly */ |
1852 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
1853 | goto nopage; | |
1854 | ||
11e33f6a MG |
1855 | /* Try direct reclaim and then allocating */ |
1856 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
1857 | zonelist, high_zoneidx, | |
1858 | nodemask, | |
5117f45d | 1859 | alloc_flags, preferred_zone, |
3dd28266 | 1860 | migratetype, &did_some_progress); |
11e33f6a MG |
1861 | if (page) |
1862 | goto got_pg; | |
1da177e4 | 1863 | |
e33c3b5e | 1864 | /* |
11e33f6a MG |
1865 | * If we failed to make any progress reclaiming, then we are |
1866 | * running out of options and have to consider going OOM | |
e33c3b5e | 1867 | */ |
11e33f6a MG |
1868 | if (!did_some_progress) { |
1869 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
1870 | if (oom_killer_disabled) |
1871 | goto nopage; | |
11e33f6a MG |
1872 | page = __alloc_pages_may_oom(gfp_mask, order, |
1873 | zonelist, high_zoneidx, | |
3dd28266 MG |
1874 | nodemask, preferred_zone, |
1875 | migratetype); | |
11e33f6a MG |
1876 | if (page) |
1877 | goto got_pg; | |
1da177e4 | 1878 | |
11e33f6a | 1879 | /* |
82553a93 DR |
1880 | * The OOM killer does not trigger for high-order |
1881 | * ~__GFP_NOFAIL allocations so if no progress is being | |
1882 | * made, there are no other options and retrying is | |
1883 | * unlikely to help. | |
11e33f6a | 1884 | */ |
82553a93 DR |
1885 | if (order > PAGE_ALLOC_COSTLY_ORDER && |
1886 | !(gfp_mask & __GFP_NOFAIL)) | |
11e33f6a | 1887 | goto nopage; |
e2c55dc8 | 1888 | |
ff0ceb9d DR |
1889 | goto restart; |
1890 | } | |
1da177e4 LT |
1891 | } |
1892 | ||
11e33f6a | 1893 | /* Check if we should retry the allocation */ |
a41f24ea | 1894 | pages_reclaimed += did_some_progress; |
11e33f6a MG |
1895 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
1896 | /* Wait for some write requests to complete then retry */ | |
8aa7e847 | 1897 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
1da177e4 LT |
1898 | goto rebalance; |
1899 | } | |
1900 | ||
1901 | nopage: | |
1902 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1903 | printk(KERN_WARNING "%s: page allocation failure." | |
1904 | " order:%d, mode:0x%x\n", | |
1905 | p->comm, order, gfp_mask); | |
1906 | dump_stack(); | |
578c2fd6 | 1907 | show_mem(); |
1da177e4 | 1908 | } |
b1eeab67 | 1909 | return page; |
1da177e4 | 1910 | got_pg: |
b1eeab67 VN |
1911 | if (kmemcheck_enabled) |
1912 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1da177e4 | 1913 | return page; |
11e33f6a | 1914 | |
1da177e4 | 1915 | } |
11e33f6a MG |
1916 | |
1917 | /* | |
1918 | * This is the 'heart' of the zoned buddy allocator. | |
1919 | */ | |
1920 | struct page * | |
1921 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
1922 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1923 | { | |
1924 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 1925 | struct zone *preferred_zone; |
11e33f6a | 1926 | struct page *page; |
3dd28266 | 1927 | int migratetype = allocflags_to_migratetype(gfp_mask); |
11e33f6a | 1928 | |
dcce284a BH |
1929 | gfp_mask &= gfp_allowed_mask; |
1930 | ||
11e33f6a MG |
1931 | lockdep_trace_alloc(gfp_mask); |
1932 | ||
1933 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1934 | ||
1935 | if (should_fail_alloc_page(gfp_mask, order)) | |
1936 | return NULL; | |
1937 | ||
1938 | /* | |
1939 | * Check the zones suitable for the gfp_mask contain at least one | |
1940 | * valid zone. It's possible to have an empty zonelist as a result | |
1941 | * of GFP_THISNODE and a memoryless node | |
1942 | */ | |
1943 | if (unlikely(!zonelist->_zonerefs->zone)) | |
1944 | return NULL; | |
1945 | ||
5117f45d MG |
1946 | /* The preferred zone is used for statistics later */ |
1947 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); | |
1948 | if (!preferred_zone) | |
1949 | return NULL; | |
1950 | ||
1951 | /* First allocation attempt */ | |
11e33f6a | 1952 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
5117f45d | 1953 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
3dd28266 | 1954 | preferred_zone, migratetype); |
11e33f6a MG |
1955 | if (unlikely(!page)) |
1956 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 1957 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 1958 | preferred_zone, migratetype); |
11e33f6a | 1959 | |
4b4f278c | 1960 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
11e33f6a | 1961 | return page; |
1da177e4 | 1962 | } |
d239171e | 1963 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
1964 | |
1965 | /* | |
1966 | * Common helper functions. | |
1967 | */ | |
920c7a5d | 1968 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 1969 | { |
945a1113 AM |
1970 | struct page *page; |
1971 | ||
1972 | /* | |
1973 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
1974 | * a highmem page | |
1975 | */ | |
1976 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
1977 | ||
1da177e4 LT |
1978 | page = alloc_pages(gfp_mask, order); |
1979 | if (!page) | |
1980 | return 0; | |
1981 | return (unsigned long) page_address(page); | |
1982 | } | |
1da177e4 LT |
1983 | EXPORT_SYMBOL(__get_free_pages); |
1984 | ||
920c7a5d | 1985 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 1986 | { |
945a1113 | 1987 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 1988 | } |
1da177e4 LT |
1989 | EXPORT_SYMBOL(get_zeroed_page); |
1990 | ||
1991 | void __pagevec_free(struct pagevec *pvec) | |
1992 | { | |
1993 | int i = pagevec_count(pvec); | |
1994 | ||
4b4f278c MG |
1995 | while (--i >= 0) { |
1996 | trace_mm_pagevec_free(pvec->pages[i], pvec->cold); | |
1da177e4 | 1997 | free_hot_cold_page(pvec->pages[i], pvec->cold); |
4b4f278c | 1998 | } |
1da177e4 LT |
1999 | } |
2000 | ||
920c7a5d | 2001 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2002 | { |
b5810039 | 2003 | if (put_page_testzero(page)) { |
4b4f278c | 2004 | trace_mm_page_free_direct(page, order); |
1da177e4 LT |
2005 | if (order == 0) |
2006 | free_hot_page(page); | |
2007 | else | |
2008 | __free_pages_ok(page, order); | |
2009 | } | |
2010 | } | |
2011 | ||
2012 | EXPORT_SYMBOL(__free_pages); | |
2013 | ||
920c7a5d | 2014 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2015 | { |
2016 | if (addr != 0) { | |
725d704e | 2017 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2018 | __free_pages(virt_to_page((void *)addr), order); |
2019 | } | |
2020 | } | |
2021 | ||
2022 | EXPORT_SYMBOL(free_pages); | |
2023 | ||
2be0ffe2 TT |
2024 | /** |
2025 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2026 | * @size: the number of bytes to allocate | |
2027 | * @gfp_mask: GFP flags for the allocation | |
2028 | * | |
2029 | * This function is similar to alloc_pages(), except that it allocates the | |
2030 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2031 | * allocate memory in power-of-two pages. | |
2032 | * | |
2033 | * This function is also limited by MAX_ORDER. | |
2034 | * | |
2035 | * Memory allocated by this function must be released by free_pages_exact(). | |
2036 | */ | |
2037 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2038 | { | |
2039 | unsigned int order = get_order(size); | |
2040 | unsigned long addr; | |
2041 | ||
2042 | addr = __get_free_pages(gfp_mask, order); | |
2043 | if (addr) { | |
2044 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2045 | unsigned long used = addr + PAGE_ALIGN(size); | |
2046 | ||
5bfd7560 | 2047 | split_page(virt_to_page((void *)addr), order); |
2be0ffe2 TT |
2048 | while (used < alloc_end) { |
2049 | free_page(used); | |
2050 | used += PAGE_SIZE; | |
2051 | } | |
2052 | } | |
2053 | ||
2054 | return (void *)addr; | |
2055 | } | |
2056 | EXPORT_SYMBOL(alloc_pages_exact); | |
2057 | ||
2058 | /** | |
2059 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2060 | * @virt: the value returned by alloc_pages_exact. | |
2061 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2062 | * | |
2063 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2064 | */ | |
2065 | void free_pages_exact(void *virt, size_t size) | |
2066 | { | |
2067 | unsigned long addr = (unsigned long)virt; | |
2068 | unsigned long end = addr + PAGE_ALIGN(size); | |
2069 | ||
2070 | while (addr < end) { | |
2071 | free_page(addr); | |
2072 | addr += PAGE_SIZE; | |
2073 | } | |
2074 | } | |
2075 | EXPORT_SYMBOL(free_pages_exact); | |
2076 | ||
1da177e4 LT |
2077 | static unsigned int nr_free_zone_pages(int offset) |
2078 | { | |
dd1a239f | 2079 | struct zoneref *z; |
54a6eb5c MG |
2080 | struct zone *zone; |
2081 | ||
e310fd43 | 2082 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2083 | unsigned int sum = 0; |
2084 | ||
0e88460d | 2085 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2086 | |
54a6eb5c | 2087 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2088 | unsigned long size = zone->present_pages; |
41858966 | 2089 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2090 | if (size > high) |
2091 | sum += size - high; | |
1da177e4 LT |
2092 | } |
2093 | ||
2094 | return sum; | |
2095 | } | |
2096 | ||
2097 | /* | |
2098 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2099 | */ | |
2100 | unsigned int nr_free_buffer_pages(void) | |
2101 | { | |
af4ca457 | 2102 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2103 | } |
c2f1a551 | 2104 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2105 | |
2106 | /* | |
2107 | * Amount of free RAM allocatable within all zones | |
2108 | */ | |
2109 | unsigned int nr_free_pagecache_pages(void) | |
2110 | { | |
2a1e274a | 2111 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2112 | } |
08e0f6a9 CL |
2113 | |
2114 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2115 | { |
08e0f6a9 | 2116 | if (NUMA_BUILD) |
25ba77c1 | 2117 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2118 | } |
1da177e4 | 2119 | |
1da177e4 LT |
2120 | void si_meminfo(struct sysinfo *val) |
2121 | { | |
2122 | val->totalram = totalram_pages; | |
2123 | val->sharedram = 0; | |
d23ad423 | 2124 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2125 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2126 | val->totalhigh = totalhigh_pages; |
2127 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2128 | val->mem_unit = PAGE_SIZE; |
2129 | } | |
2130 | ||
2131 | EXPORT_SYMBOL(si_meminfo); | |
2132 | ||
2133 | #ifdef CONFIG_NUMA | |
2134 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2135 | { | |
2136 | pg_data_t *pgdat = NODE_DATA(nid); | |
2137 | ||
2138 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2139 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2140 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2141 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2142 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2143 | NR_FREE_PAGES); | |
98d2b0eb CL |
2144 | #else |
2145 | val->totalhigh = 0; | |
2146 | val->freehigh = 0; | |
2147 | #endif | |
1da177e4 LT |
2148 | val->mem_unit = PAGE_SIZE; |
2149 | } | |
2150 | #endif | |
2151 | ||
2152 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
2153 | ||
2154 | /* | |
2155 | * Show free area list (used inside shift_scroll-lock stuff) | |
2156 | * We also calculate the percentage fragmentation. We do this by counting the | |
2157 | * memory on each free list with the exception of the first item on the list. | |
2158 | */ | |
2159 | void show_free_areas(void) | |
2160 | { | |
c7241913 | 2161 | int cpu; |
1da177e4 LT |
2162 | struct zone *zone; |
2163 | ||
ee99c71c | 2164 | for_each_populated_zone(zone) { |
c7241913 JS |
2165 | show_node(zone); |
2166 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2167 | |
6b482c67 | 2168 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2169 | struct per_cpu_pageset *pageset; |
2170 | ||
e7c8d5c9 | 2171 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 2172 | |
3dfa5721 CL |
2173 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2174 | cpu, pageset->pcp.high, | |
2175 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2176 | } |
2177 | } | |
2178 | ||
a731286d KM |
2179 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2180 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2181 | " unevictable:%lu" |
b76146ed | 2182 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2183 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
4b02108a | 2184 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe | 2185 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2186 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2187 | global_page_state(NR_ISOLATED_ANON), |
2188 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2189 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2190 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2191 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2192 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2193 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2194 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2195 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2196 | global_page_state(NR_SLAB_RECLAIMABLE), |
2197 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2198 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2199 | global_page_state(NR_SHMEM), |
a25700a5 AM |
2200 | global_page_state(NR_PAGETABLE), |
2201 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2202 | |
ee99c71c | 2203 | for_each_populated_zone(zone) { |
1da177e4 LT |
2204 | int i; |
2205 | ||
2206 | show_node(zone); | |
2207 | printk("%s" | |
2208 | " free:%lukB" | |
2209 | " min:%lukB" | |
2210 | " low:%lukB" | |
2211 | " high:%lukB" | |
4f98a2fe RR |
2212 | " active_anon:%lukB" |
2213 | " inactive_anon:%lukB" | |
2214 | " active_file:%lukB" | |
2215 | " inactive_file:%lukB" | |
7b854121 | 2216 | " unevictable:%lukB" |
a731286d KM |
2217 | " isolated(anon):%lukB" |
2218 | " isolated(file):%lukB" | |
1da177e4 | 2219 | " present:%lukB" |
4a0aa73f KM |
2220 | " mlocked:%lukB" |
2221 | " dirty:%lukB" | |
2222 | " writeback:%lukB" | |
2223 | " mapped:%lukB" | |
4b02108a | 2224 | " shmem:%lukB" |
4a0aa73f KM |
2225 | " slab_reclaimable:%lukB" |
2226 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 2227 | " kernel_stack:%lukB" |
4a0aa73f KM |
2228 | " pagetables:%lukB" |
2229 | " unstable:%lukB" | |
2230 | " bounce:%lukB" | |
2231 | " writeback_tmp:%lukB" | |
1da177e4 LT |
2232 | " pages_scanned:%lu" |
2233 | " all_unreclaimable? %s" | |
2234 | "\n", | |
2235 | zone->name, | |
d23ad423 | 2236 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
2237 | K(min_wmark_pages(zone)), |
2238 | K(low_wmark_pages(zone)), | |
2239 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
2240 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2241 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2242 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2243 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 2244 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
2245 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
2246 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 2247 | K(zone->present_pages), |
4a0aa73f KM |
2248 | K(zone_page_state(zone, NR_MLOCK)), |
2249 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
2250 | K(zone_page_state(zone, NR_WRITEBACK)), | |
2251 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 2252 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
2253 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
2254 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
2255 | zone_page_state(zone, NR_KERNEL_STACK) * |
2256 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
2257 | K(zone_page_state(zone, NR_PAGETABLE)), |
2258 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
2259 | K(zone_page_state(zone, NR_BOUNCE)), | |
2260 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), | |
1da177e4 | 2261 | zone->pages_scanned, |
e815af95 | 2262 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
2263 | ); |
2264 | printk("lowmem_reserve[]:"); | |
2265 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2266 | printk(" %lu", zone->lowmem_reserve[i]); | |
2267 | printk("\n"); | |
2268 | } | |
2269 | ||
ee99c71c | 2270 | for_each_populated_zone(zone) { |
8f9de51a | 2271 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
2272 | |
2273 | show_node(zone); | |
2274 | printk("%s: ", zone->name); | |
1da177e4 LT |
2275 | |
2276 | spin_lock_irqsave(&zone->lock, flags); | |
2277 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2278 | nr[order] = zone->free_area[order].nr_free; |
2279 | total += nr[order] << order; | |
1da177e4 LT |
2280 | } |
2281 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2282 | for (order = 0; order < MAX_ORDER; order++) |
2283 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2284 | printk("= %lukB\n", K(total)); |
2285 | } | |
2286 | ||
e6f3602d LW |
2287 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2288 | ||
1da177e4 LT |
2289 | show_swap_cache_info(); |
2290 | } | |
2291 | ||
19770b32 MG |
2292 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2293 | { | |
2294 | zoneref->zone = zone; | |
2295 | zoneref->zone_idx = zone_idx(zone); | |
2296 | } | |
2297 | ||
1da177e4 LT |
2298 | /* |
2299 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2300 | * |
2301 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2302 | */ |
f0c0b2b8 KH |
2303 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2304 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2305 | { |
1a93205b CL |
2306 | struct zone *zone; |
2307 | ||
98d2b0eb | 2308 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2309 | zone_type++; |
02a68a5e CL |
2310 | |
2311 | do { | |
2f6726e5 | 2312 | zone_type--; |
070f8032 | 2313 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2314 | if (populated_zone(zone)) { |
dd1a239f MG |
2315 | zoneref_set_zone(zone, |
2316 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2317 | check_highest_zone(zone_type); |
1da177e4 | 2318 | } |
02a68a5e | 2319 | |
2f6726e5 | 2320 | } while (zone_type); |
070f8032 | 2321 | return nr_zones; |
1da177e4 LT |
2322 | } |
2323 | ||
f0c0b2b8 KH |
2324 | |
2325 | /* | |
2326 | * zonelist_order: | |
2327 | * 0 = automatic detection of better ordering. | |
2328 | * 1 = order by ([node] distance, -zonetype) | |
2329 | * 2 = order by (-zonetype, [node] distance) | |
2330 | * | |
2331 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2332 | * the same zonelist. So only NUMA can configure this param. | |
2333 | */ | |
2334 | #define ZONELIST_ORDER_DEFAULT 0 | |
2335 | #define ZONELIST_ORDER_NODE 1 | |
2336 | #define ZONELIST_ORDER_ZONE 2 | |
2337 | ||
2338 | /* zonelist order in the kernel. | |
2339 | * set_zonelist_order() will set this to NODE or ZONE. | |
2340 | */ | |
2341 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2342 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2343 | ||
2344 | ||
1da177e4 | 2345 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2346 | /* The value user specified ....changed by config */ |
2347 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2348 | /* string for sysctl */ | |
2349 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2350 | char numa_zonelist_order[16] = "default"; | |
2351 | ||
2352 | /* | |
2353 | * interface for configure zonelist ordering. | |
2354 | * command line option "numa_zonelist_order" | |
2355 | * = "[dD]efault - default, automatic configuration. | |
2356 | * = "[nN]ode - order by node locality, then by zone within node | |
2357 | * = "[zZ]one - order by zone, then by locality within zone | |
2358 | */ | |
2359 | ||
2360 | static int __parse_numa_zonelist_order(char *s) | |
2361 | { | |
2362 | if (*s == 'd' || *s == 'D') { | |
2363 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2364 | } else if (*s == 'n' || *s == 'N') { | |
2365 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2366 | } else if (*s == 'z' || *s == 'Z') { | |
2367 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2368 | } else { | |
2369 | printk(KERN_WARNING | |
2370 | "Ignoring invalid numa_zonelist_order value: " | |
2371 | "%s\n", s); | |
2372 | return -EINVAL; | |
2373 | } | |
2374 | return 0; | |
2375 | } | |
2376 | ||
2377 | static __init int setup_numa_zonelist_order(char *s) | |
2378 | { | |
2379 | if (s) | |
2380 | return __parse_numa_zonelist_order(s); | |
2381 | return 0; | |
2382 | } | |
2383 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2384 | ||
2385 | /* | |
2386 | * sysctl handler for numa_zonelist_order | |
2387 | */ | |
2388 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 2389 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
2390 | loff_t *ppos) |
2391 | { | |
2392 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2393 | int ret; | |
2394 | ||
2395 | if (write) | |
2396 | strncpy(saved_string, (char*)table->data, | |
2397 | NUMA_ZONELIST_ORDER_LEN); | |
8d65af78 | 2398 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 KH |
2399 | if (ret) |
2400 | return ret; | |
2401 | if (write) { | |
2402 | int oldval = user_zonelist_order; | |
2403 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2404 | /* | |
2405 | * bogus value. restore saved string | |
2406 | */ | |
2407 | strncpy((char*)table->data, saved_string, | |
2408 | NUMA_ZONELIST_ORDER_LEN); | |
2409 | user_zonelist_order = oldval; | |
2410 | } else if (oldval != user_zonelist_order) | |
2411 | build_all_zonelists(); | |
2412 | } | |
2413 | return 0; | |
2414 | } | |
2415 | ||
2416 | ||
62bc62a8 | 2417 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
2418 | static int node_load[MAX_NUMNODES]; |
2419 | ||
1da177e4 | 2420 | /** |
4dc3b16b | 2421 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2422 | * @node: node whose fallback list we're appending |
2423 | * @used_node_mask: nodemask_t of already used nodes | |
2424 | * | |
2425 | * We use a number of factors to determine which is the next node that should | |
2426 | * appear on a given node's fallback list. The node should not have appeared | |
2427 | * already in @node's fallback list, and it should be the next closest node | |
2428 | * according to the distance array (which contains arbitrary distance values | |
2429 | * from each node to each node in the system), and should also prefer nodes | |
2430 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2431 | * on them otherwise. | |
2432 | * It returns -1 if no node is found. | |
2433 | */ | |
f0c0b2b8 | 2434 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2435 | { |
4cf808eb | 2436 | int n, val; |
1da177e4 LT |
2437 | int min_val = INT_MAX; |
2438 | int best_node = -1; | |
a70f7302 | 2439 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2440 | |
4cf808eb LT |
2441 | /* Use the local node if we haven't already */ |
2442 | if (!node_isset(node, *used_node_mask)) { | |
2443 | node_set(node, *used_node_mask); | |
2444 | return node; | |
2445 | } | |
1da177e4 | 2446 | |
37b07e41 | 2447 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2448 | |
2449 | /* Don't want a node to appear more than once */ | |
2450 | if (node_isset(n, *used_node_mask)) | |
2451 | continue; | |
2452 | ||
1da177e4 LT |
2453 | /* Use the distance array to find the distance */ |
2454 | val = node_distance(node, n); | |
2455 | ||
4cf808eb LT |
2456 | /* Penalize nodes under us ("prefer the next node") */ |
2457 | val += (n < node); | |
2458 | ||
1da177e4 | 2459 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2460 | tmp = cpumask_of_node(n); |
2461 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2462 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2463 | ||
2464 | /* Slight preference for less loaded node */ | |
2465 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2466 | val += node_load[n]; | |
2467 | ||
2468 | if (val < min_val) { | |
2469 | min_val = val; | |
2470 | best_node = n; | |
2471 | } | |
2472 | } | |
2473 | ||
2474 | if (best_node >= 0) | |
2475 | node_set(best_node, *used_node_mask); | |
2476 | ||
2477 | return best_node; | |
2478 | } | |
2479 | ||
f0c0b2b8 KH |
2480 | |
2481 | /* | |
2482 | * Build zonelists ordered by node and zones within node. | |
2483 | * This results in maximum locality--normal zone overflows into local | |
2484 | * DMA zone, if any--but risks exhausting DMA zone. | |
2485 | */ | |
2486 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2487 | { |
f0c0b2b8 | 2488 | int j; |
1da177e4 | 2489 | struct zonelist *zonelist; |
f0c0b2b8 | 2490 | |
54a6eb5c | 2491 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2492 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2493 | ; |
2494 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2495 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2496 | zonelist->_zonerefs[j].zone = NULL; |
2497 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2498 | } |
2499 | ||
523b9458 CL |
2500 | /* |
2501 | * Build gfp_thisnode zonelists | |
2502 | */ | |
2503 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2504 | { | |
523b9458 CL |
2505 | int j; |
2506 | struct zonelist *zonelist; | |
2507 | ||
54a6eb5c MG |
2508 | zonelist = &pgdat->node_zonelists[1]; |
2509 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2510 | zonelist->_zonerefs[j].zone = NULL; |
2511 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2512 | } |
2513 | ||
f0c0b2b8 KH |
2514 | /* |
2515 | * Build zonelists ordered by zone and nodes within zones. | |
2516 | * This results in conserving DMA zone[s] until all Normal memory is | |
2517 | * exhausted, but results in overflowing to remote node while memory | |
2518 | * may still exist in local DMA zone. | |
2519 | */ | |
2520 | static int node_order[MAX_NUMNODES]; | |
2521 | ||
2522 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2523 | { | |
f0c0b2b8 KH |
2524 | int pos, j, node; |
2525 | int zone_type; /* needs to be signed */ | |
2526 | struct zone *z; | |
2527 | struct zonelist *zonelist; | |
2528 | ||
54a6eb5c MG |
2529 | zonelist = &pgdat->node_zonelists[0]; |
2530 | pos = 0; | |
2531 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2532 | for (j = 0; j < nr_nodes; j++) { | |
2533 | node = node_order[j]; | |
2534 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2535 | if (populated_zone(z)) { | |
dd1a239f MG |
2536 | zoneref_set_zone(z, |
2537 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2538 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2539 | } |
2540 | } | |
f0c0b2b8 | 2541 | } |
dd1a239f MG |
2542 | zonelist->_zonerefs[pos].zone = NULL; |
2543 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2544 | } |
2545 | ||
2546 | static int default_zonelist_order(void) | |
2547 | { | |
2548 | int nid, zone_type; | |
2549 | unsigned long low_kmem_size,total_size; | |
2550 | struct zone *z; | |
2551 | int average_size; | |
2552 | /* | |
2553 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2554 | * If they are really small and used heavily, the system can fall | |
2555 | * into OOM very easily. | |
2556 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2557 | */ | |
2558 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2559 | low_kmem_size = 0; | |
2560 | total_size = 0; | |
2561 | for_each_online_node(nid) { | |
2562 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2563 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2564 | if (populated_zone(z)) { | |
2565 | if (zone_type < ZONE_NORMAL) | |
2566 | low_kmem_size += z->present_pages; | |
2567 | total_size += z->present_pages; | |
2568 | } | |
2569 | } | |
2570 | } | |
2571 | if (!low_kmem_size || /* there are no DMA area. */ | |
2572 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2573 | return ZONELIST_ORDER_NODE; | |
2574 | /* | |
2575 | * look into each node's config. | |
2576 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2577 | * local memory, NODE_ORDER may be suitable. | |
2578 | */ | |
37b07e41 LS |
2579 | average_size = total_size / |
2580 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2581 | for_each_online_node(nid) { |
2582 | low_kmem_size = 0; | |
2583 | total_size = 0; | |
2584 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2585 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2586 | if (populated_zone(z)) { | |
2587 | if (zone_type < ZONE_NORMAL) | |
2588 | low_kmem_size += z->present_pages; | |
2589 | total_size += z->present_pages; | |
2590 | } | |
2591 | } | |
2592 | if (low_kmem_size && | |
2593 | total_size > average_size && /* ignore small node */ | |
2594 | low_kmem_size > total_size * 70/100) | |
2595 | return ZONELIST_ORDER_NODE; | |
2596 | } | |
2597 | return ZONELIST_ORDER_ZONE; | |
2598 | } | |
2599 | ||
2600 | static void set_zonelist_order(void) | |
2601 | { | |
2602 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2603 | current_zonelist_order = default_zonelist_order(); | |
2604 | else | |
2605 | current_zonelist_order = user_zonelist_order; | |
2606 | } | |
2607 | ||
2608 | static void build_zonelists(pg_data_t *pgdat) | |
2609 | { | |
2610 | int j, node, load; | |
2611 | enum zone_type i; | |
1da177e4 | 2612 | nodemask_t used_mask; |
f0c0b2b8 KH |
2613 | int local_node, prev_node; |
2614 | struct zonelist *zonelist; | |
2615 | int order = current_zonelist_order; | |
1da177e4 LT |
2616 | |
2617 | /* initialize zonelists */ | |
523b9458 | 2618 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2619 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2620 | zonelist->_zonerefs[0].zone = NULL; |
2621 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2622 | } |
2623 | ||
2624 | /* NUMA-aware ordering of nodes */ | |
2625 | local_node = pgdat->node_id; | |
62bc62a8 | 2626 | load = nr_online_nodes; |
1da177e4 LT |
2627 | prev_node = local_node; |
2628 | nodes_clear(used_mask); | |
f0c0b2b8 | 2629 | |
f0c0b2b8 KH |
2630 | memset(node_order, 0, sizeof(node_order)); |
2631 | j = 0; | |
2632 | ||
1da177e4 | 2633 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2634 | int distance = node_distance(local_node, node); |
2635 | ||
2636 | /* | |
2637 | * If another node is sufficiently far away then it is better | |
2638 | * to reclaim pages in a zone before going off node. | |
2639 | */ | |
2640 | if (distance > RECLAIM_DISTANCE) | |
2641 | zone_reclaim_mode = 1; | |
2642 | ||
1da177e4 LT |
2643 | /* |
2644 | * We don't want to pressure a particular node. | |
2645 | * So adding penalty to the first node in same | |
2646 | * distance group to make it round-robin. | |
2647 | */ | |
9eeff239 | 2648 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2649 | node_load[node] = load; |
2650 | ||
1da177e4 LT |
2651 | prev_node = node; |
2652 | load--; | |
f0c0b2b8 KH |
2653 | if (order == ZONELIST_ORDER_NODE) |
2654 | build_zonelists_in_node_order(pgdat, node); | |
2655 | else | |
2656 | node_order[j++] = node; /* remember order */ | |
2657 | } | |
1da177e4 | 2658 | |
f0c0b2b8 KH |
2659 | if (order == ZONELIST_ORDER_ZONE) { |
2660 | /* calculate node order -- i.e., DMA last! */ | |
2661 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2662 | } |
523b9458 CL |
2663 | |
2664 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2665 | } |
2666 | ||
9276b1bc | 2667 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2668 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2669 | { |
54a6eb5c MG |
2670 | struct zonelist *zonelist; |
2671 | struct zonelist_cache *zlc; | |
dd1a239f | 2672 | struct zoneref *z; |
9276b1bc | 2673 | |
54a6eb5c MG |
2674 | zonelist = &pgdat->node_zonelists[0]; |
2675 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2676 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2677 | for (z = zonelist->_zonerefs; z->zone; z++) |
2678 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2679 | } |
2680 | ||
f0c0b2b8 | 2681 | |
1da177e4 LT |
2682 | #else /* CONFIG_NUMA */ |
2683 | ||
f0c0b2b8 KH |
2684 | static void set_zonelist_order(void) |
2685 | { | |
2686 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2687 | } | |
2688 | ||
2689 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2690 | { |
19655d34 | 2691 | int node, local_node; |
54a6eb5c MG |
2692 | enum zone_type j; |
2693 | struct zonelist *zonelist; | |
1da177e4 LT |
2694 | |
2695 | local_node = pgdat->node_id; | |
1da177e4 | 2696 | |
54a6eb5c MG |
2697 | zonelist = &pgdat->node_zonelists[0]; |
2698 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2699 | |
54a6eb5c MG |
2700 | /* |
2701 | * Now we build the zonelist so that it contains the zones | |
2702 | * of all the other nodes. | |
2703 | * We don't want to pressure a particular node, so when | |
2704 | * building the zones for node N, we make sure that the | |
2705 | * zones coming right after the local ones are those from | |
2706 | * node N+1 (modulo N) | |
2707 | */ | |
2708 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2709 | if (!node_online(node)) | |
2710 | continue; | |
2711 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2712 | MAX_NR_ZONES - 1); | |
1da177e4 | 2713 | } |
54a6eb5c MG |
2714 | for (node = 0; node < local_node; node++) { |
2715 | if (!node_online(node)) | |
2716 | continue; | |
2717 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2718 | MAX_NR_ZONES - 1); | |
2719 | } | |
2720 | ||
dd1a239f MG |
2721 | zonelist->_zonerefs[j].zone = NULL; |
2722 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
2723 | } |
2724 | ||
9276b1bc | 2725 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2726 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2727 | { |
54a6eb5c | 2728 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
2729 | } |
2730 | ||
1da177e4 LT |
2731 | #endif /* CONFIG_NUMA */ |
2732 | ||
9b1a4d38 | 2733 | /* return values int ....just for stop_machine() */ |
f0c0b2b8 | 2734 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2735 | { |
6811378e | 2736 | int nid; |
9276b1bc | 2737 | |
7f9cfb31 BL |
2738 | #ifdef CONFIG_NUMA |
2739 | memset(node_load, 0, sizeof(node_load)); | |
2740 | #endif | |
9276b1bc | 2741 | for_each_online_node(nid) { |
7ea1530a CL |
2742 | pg_data_t *pgdat = NODE_DATA(nid); |
2743 | ||
2744 | build_zonelists(pgdat); | |
2745 | build_zonelist_cache(pgdat); | |
9276b1bc | 2746 | } |
6811378e YG |
2747 | return 0; |
2748 | } | |
2749 | ||
f0c0b2b8 | 2750 | void build_all_zonelists(void) |
6811378e | 2751 | { |
f0c0b2b8 KH |
2752 | set_zonelist_order(); |
2753 | ||
6811378e | 2754 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2755 | __build_all_zonelists(NULL); |
68ad8df4 | 2756 | mminit_verify_zonelist(); |
6811378e YG |
2757 | cpuset_init_current_mems_allowed(); |
2758 | } else { | |
183ff22b | 2759 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 2760 | of zonelist */ |
9b1a4d38 | 2761 | stop_machine(__build_all_zonelists, NULL, NULL); |
6811378e YG |
2762 | /* cpuset refresh routine should be here */ |
2763 | } | |
bd1e22b8 | 2764 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2765 | /* |
2766 | * Disable grouping by mobility if the number of pages in the | |
2767 | * system is too low to allow the mechanism to work. It would be | |
2768 | * more accurate, but expensive to check per-zone. This check is | |
2769 | * made on memory-hotadd so a system can start with mobility | |
2770 | * disabled and enable it later | |
2771 | */ | |
d9c23400 | 2772 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2773 | page_group_by_mobility_disabled = 1; |
2774 | else | |
2775 | page_group_by_mobility_disabled = 0; | |
2776 | ||
2777 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2778 | "Total pages: %ld\n", | |
62bc62a8 | 2779 | nr_online_nodes, |
f0c0b2b8 | 2780 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 2781 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2782 | vm_total_pages); |
2783 | #ifdef CONFIG_NUMA | |
2784 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2785 | #endif | |
1da177e4 LT |
2786 | } |
2787 | ||
2788 | /* | |
2789 | * Helper functions to size the waitqueue hash table. | |
2790 | * Essentially these want to choose hash table sizes sufficiently | |
2791 | * large so that collisions trying to wait on pages are rare. | |
2792 | * But in fact, the number of active page waitqueues on typical | |
2793 | * systems is ridiculously low, less than 200. So this is even | |
2794 | * conservative, even though it seems large. | |
2795 | * | |
2796 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2797 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2798 | */ | |
2799 | #define PAGES_PER_WAITQUEUE 256 | |
2800 | ||
cca448fe | 2801 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2802 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2803 | { |
2804 | unsigned long size = 1; | |
2805 | ||
2806 | pages /= PAGES_PER_WAITQUEUE; | |
2807 | ||
2808 | while (size < pages) | |
2809 | size <<= 1; | |
2810 | ||
2811 | /* | |
2812 | * Once we have dozens or even hundreds of threads sleeping | |
2813 | * on IO we've got bigger problems than wait queue collision. | |
2814 | * Limit the size of the wait table to a reasonable size. | |
2815 | */ | |
2816 | size = min(size, 4096UL); | |
2817 | ||
2818 | return max(size, 4UL); | |
2819 | } | |
cca448fe YG |
2820 | #else |
2821 | /* | |
2822 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2823 | * a suitable size for its wait_table. So we use the maximum size now. | |
2824 | * | |
2825 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2826 | * | |
2827 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2828 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2829 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2830 | * | |
2831 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2832 | * or more by the traditional way. (See above). It equals: | |
2833 | * | |
2834 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2835 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2836 | * powerpc (64K page size) : = (32G +16M)byte. | |
2837 | */ | |
2838 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2839 | { | |
2840 | return 4096UL; | |
2841 | } | |
2842 | #endif | |
1da177e4 LT |
2843 | |
2844 | /* | |
2845 | * This is an integer logarithm so that shifts can be used later | |
2846 | * to extract the more random high bits from the multiplicative | |
2847 | * hash function before the remainder is taken. | |
2848 | */ | |
2849 | static inline unsigned long wait_table_bits(unsigned long size) | |
2850 | { | |
2851 | return ffz(~size); | |
2852 | } | |
2853 | ||
2854 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2855 | ||
56fd56b8 | 2856 | /* |
d9c23400 | 2857 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
2858 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
2859 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
2860 | * higher will lead to a bigger reserve which will get freed as contiguous |
2861 | * blocks as reclaim kicks in | |
2862 | */ | |
2863 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2864 | { | |
2865 | unsigned long start_pfn, pfn, end_pfn; | |
2866 | struct page *page; | |
78986a67 MG |
2867 | unsigned long block_migratetype; |
2868 | int reserve; | |
56fd56b8 MG |
2869 | |
2870 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2871 | start_pfn = zone->zone_start_pfn; | |
2872 | end_pfn = start_pfn + zone->spanned_pages; | |
41858966 | 2873 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 2874 | pageblock_order; |
56fd56b8 | 2875 | |
78986a67 MG |
2876 | /* |
2877 | * Reserve blocks are generally in place to help high-order atomic | |
2878 | * allocations that are short-lived. A min_free_kbytes value that | |
2879 | * would result in more than 2 reserve blocks for atomic allocations | |
2880 | * is assumed to be in place to help anti-fragmentation for the | |
2881 | * future allocation of hugepages at runtime. | |
2882 | */ | |
2883 | reserve = min(2, reserve); | |
2884 | ||
d9c23400 | 2885 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2886 | if (!pfn_valid(pfn)) |
2887 | continue; | |
2888 | page = pfn_to_page(pfn); | |
2889 | ||
344c790e AL |
2890 | /* Watch out for overlapping nodes */ |
2891 | if (page_to_nid(page) != zone_to_nid(zone)) | |
2892 | continue; | |
2893 | ||
56fd56b8 MG |
2894 | /* Blocks with reserved pages will never free, skip them. */ |
2895 | if (PageReserved(page)) | |
2896 | continue; | |
2897 | ||
2898 | block_migratetype = get_pageblock_migratetype(page); | |
2899 | ||
2900 | /* If this block is reserved, account for it */ | |
2901 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2902 | reserve--; | |
2903 | continue; | |
2904 | } | |
2905 | ||
2906 | /* Suitable for reserving if this block is movable */ | |
2907 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2908 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2909 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2910 | reserve--; | |
2911 | continue; | |
2912 | } | |
2913 | ||
2914 | /* | |
2915 | * If the reserve is met and this is a previous reserved block, | |
2916 | * take it back | |
2917 | */ | |
2918 | if (block_migratetype == MIGRATE_RESERVE) { | |
2919 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2920 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2921 | } | |
2922 | } | |
2923 | } | |
ac0e5b7a | 2924 | |
1da177e4 LT |
2925 | /* |
2926 | * Initially all pages are reserved - free ones are freed | |
2927 | * up by free_all_bootmem() once the early boot process is | |
2928 | * done. Non-atomic initialization, single-pass. | |
2929 | */ | |
c09b4240 | 2930 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2931 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2932 | { |
1da177e4 | 2933 | struct page *page; |
29751f69 AW |
2934 | unsigned long end_pfn = start_pfn + size; |
2935 | unsigned long pfn; | |
86051ca5 | 2936 | struct zone *z; |
1da177e4 | 2937 | |
22b31eec HD |
2938 | if (highest_memmap_pfn < end_pfn - 1) |
2939 | highest_memmap_pfn = end_pfn - 1; | |
2940 | ||
86051ca5 | 2941 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 2942 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2943 | /* |
2944 | * There can be holes in boot-time mem_map[]s | |
2945 | * handed to this function. They do not | |
2946 | * exist on hotplugged memory. | |
2947 | */ | |
2948 | if (context == MEMMAP_EARLY) { | |
2949 | if (!early_pfn_valid(pfn)) | |
2950 | continue; | |
2951 | if (!early_pfn_in_nid(pfn, nid)) | |
2952 | continue; | |
2953 | } | |
d41dee36 AW |
2954 | page = pfn_to_page(pfn); |
2955 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 2956 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 2957 | init_page_count(page); |
1da177e4 LT |
2958 | reset_page_mapcount(page); |
2959 | SetPageReserved(page); | |
b2a0ac88 MG |
2960 | /* |
2961 | * Mark the block movable so that blocks are reserved for | |
2962 | * movable at startup. This will force kernel allocations | |
2963 | * to reserve their blocks rather than leaking throughout | |
2964 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2965 | * kernel allocations are made. Later some blocks near |
2966 | * the start are marked MIGRATE_RESERVE by | |
2967 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
2968 | * |
2969 | * bitmap is created for zone's valid pfn range. but memmap | |
2970 | * can be created for invalid pages (for alignment) | |
2971 | * check here not to call set_pageblock_migratetype() against | |
2972 | * pfn out of zone. | |
b2a0ac88 | 2973 | */ |
86051ca5 KH |
2974 | if ((z->zone_start_pfn <= pfn) |
2975 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
2976 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 2977 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2978 | |
1da177e4 LT |
2979 | INIT_LIST_HEAD(&page->lru); |
2980 | #ifdef WANT_PAGE_VIRTUAL | |
2981 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2982 | if (!is_highmem_idx(zone)) | |
3212c6be | 2983 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2984 | #endif |
1da177e4 LT |
2985 | } |
2986 | } | |
2987 | ||
1e548deb | 2988 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 2989 | { |
b2a0ac88 MG |
2990 | int order, t; |
2991 | for_each_migratetype_order(order, t) { | |
2992 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2993 | zone->free_area[order].nr_free = 0; |
2994 | } | |
2995 | } | |
2996 | ||
2997 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2998 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2999 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3000 | #endif |
3001 | ||
1d6f4e60 | 3002 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3003 | { |
3a6be87f | 3004 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3005 | int batch; |
3006 | ||
3007 | /* | |
3008 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3009 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3010 | * |
3011 | * OK, so we don't know how big the cache is. So guess. | |
3012 | */ | |
3013 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3014 | if (batch * PAGE_SIZE > 512 * 1024) |
3015 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3016 | batch /= 4; /* We effectively *= 4 below */ |
3017 | if (batch < 1) | |
3018 | batch = 1; | |
3019 | ||
3020 | /* | |
0ceaacc9 NP |
3021 | * Clamp the batch to a 2^n - 1 value. Having a power |
3022 | * of 2 value was found to be more likely to have | |
3023 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3024 | * |
0ceaacc9 NP |
3025 | * For example if 2 tasks are alternately allocating |
3026 | * batches of pages, one task can end up with a lot | |
3027 | * of pages of one half of the possible page colors | |
3028 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3029 | */ |
9155203a | 3030 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3031 | |
e7c8d5c9 | 3032 | return batch; |
3a6be87f DH |
3033 | |
3034 | #else | |
3035 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3036 | * conditions. | |
3037 | * | |
3038 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3039 | * of contiguous memory as there's no hardware page translation to | |
3040 | * assemble apparent contiguous memory from discontiguous pages. | |
3041 | * | |
3042 | * Queueing large contiguous runs of pages for batching, however, | |
3043 | * causes the pages to actually be freed in smaller chunks. As there | |
3044 | * can be a significant delay between the individual batches being | |
3045 | * recycled, this leads to the once large chunks of space being | |
3046 | * fragmented and becoming unavailable for high-order allocations. | |
3047 | */ | |
3048 | return 0; | |
3049 | #endif | |
e7c8d5c9 CL |
3050 | } |
3051 | ||
b69a7288 | 3052 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3053 | { |
3054 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3055 | int migratetype; |
2caaad41 | 3056 | |
1c6fe946 MD |
3057 | memset(p, 0, sizeof(*p)); |
3058 | ||
3dfa5721 | 3059 | pcp = &p->pcp; |
2caaad41 | 3060 | pcp->count = 0; |
2caaad41 CL |
3061 | pcp->high = 6 * batch; |
3062 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3063 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3064 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3065 | } |
3066 | ||
8ad4b1fb RS |
3067 | /* |
3068 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3069 | * to the value high for the pageset p. | |
3070 | */ | |
3071 | ||
3072 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3073 | unsigned long high) | |
3074 | { | |
3075 | struct per_cpu_pages *pcp; | |
3076 | ||
3dfa5721 | 3077 | pcp = &p->pcp; |
8ad4b1fb RS |
3078 | pcp->high = high; |
3079 | pcp->batch = max(1UL, high/4); | |
3080 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3081 | pcp->batch = PAGE_SHIFT * 8; | |
3082 | } | |
3083 | ||
3084 | ||
e7c8d5c9 CL |
3085 | #ifdef CONFIG_NUMA |
3086 | /* | |
2caaad41 CL |
3087 | * Boot pageset table. One per cpu which is going to be used for all |
3088 | * zones and all nodes. The parameters will be set in such a way | |
3089 | * that an item put on a list will immediately be handed over to | |
3090 | * the buddy list. This is safe since pageset manipulation is done | |
3091 | * with interrupts disabled. | |
3092 | * | |
3093 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
3094 | * |
3095 | * The boot_pagesets must be kept even after bootup is complete for | |
3096 | * unused processors and/or zones. They do play a role for bootstrapping | |
3097 | * hotplugged processors. | |
3098 | * | |
3099 | * zoneinfo_show() and maybe other functions do | |
3100 | * not check if the processor is online before following the pageset pointer. | |
3101 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 3102 | */ |
88a2a4ac | 3103 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
3104 | |
3105 | /* | |
3106 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
3107 | * per cpu pageset array in struct zone. |
3108 | */ | |
6292d9aa | 3109 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
3110 | { |
3111 | struct zone *zone, *dzone; | |
37c0708d CL |
3112 | int node = cpu_to_node(cpu); |
3113 | ||
3114 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 | 3115 | |
ee99c71c | 3116 | for_each_populated_zone(zone) { |
23316bc8 | 3117 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 3118 | GFP_KERNEL, node); |
23316bc8 | 3119 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 3120 | goto bad; |
e7c8d5c9 | 3121 | |
23316bc8 | 3122 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
3123 | |
3124 | if (percpu_pagelist_fraction) | |
3125 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
3126 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
3127 | } |
3128 | ||
3129 | return 0; | |
3130 | bad: | |
3131 | for_each_zone(dzone) { | |
64191688 AM |
3132 | if (!populated_zone(dzone)) |
3133 | continue; | |
e7c8d5c9 CL |
3134 | if (dzone == zone) |
3135 | break; | |
23316bc8 | 3136 | kfree(zone_pcp(dzone, cpu)); |
364df0eb | 3137 | zone_pcp(dzone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 CL |
3138 | } |
3139 | return -ENOMEM; | |
3140 | } | |
3141 | ||
3142 | static inline void free_zone_pagesets(int cpu) | |
3143 | { | |
e7c8d5c9 CL |
3144 | struct zone *zone; |
3145 | ||
3146 | for_each_zone(zone) { | |
3147 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
3148 | ||
f3ef9ead DR |
3149 | /* Free per_cpu_pageset if it is slab allocated */ |
3150 | if (pset != &boot_pageset[cpu]) | |
3151 | kfree(pset); | |
364df0eb | 3152 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
e7c8d5c9 | 3153 | } |
e7c8d5c9 CL |
3154 | } |
3155 | ||
9c7b216d | 3156 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
3157 | unsigned long action, |
3158 | void *hcpu) | |
3159 | { | |
3160 | int cpu = (long)hcpu; | |
3161 | int ret = NOTIFY_OK; | |
3162 | ||
3163 | switch (action) { | |
ce421c79 | 3164 | case CPU_UP_PREPARE: |
8bb78442 | 3165 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
3166 | if (process_zones(cpu)) |
3167 | ret = NOTIFY_BAD; | |
3168 | break; | |
3169 | case CPU_UP_CANCELED: | |
8bb78442 | 3170 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 3171 | case CPU_DEAD: |
8bb78442 | 3172 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
3173 | free_zone_pagesets(cpu); |
3174 | break; | |
3175 | default: | |
3176 | break; | |
e7c8d5c9 CL |
3177 | } |
3178 | return ret; | |
3179 | } | |
3180 | ||
74b85f37 | 3181 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
3182 | { &pageset_cpuup_callback, NULL, 0 }; |
3183 | ||
78d9955b | 3184 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
3185 | { |
3186 | int err; | |
3187 | ||
3188 | /* Initialize per_cpu_pageset for cpu 0. | |
3189 | * A cpuup callback will do this for every cpu | |
3190 | * as it comes online | |
3191 | */ | |
3192 | err = process_zones(smp_processor_id()); | |
3193 | BUG_ON(err); | |
3194 | register_cpu_notifier(&pageset_notifier); | |
3195 | } | |
3196 | ||
3197 | #endif | |
3198 | ||
577a32f6 | 3199 | static noinline __init_refok |
cca448fe | 3200 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
3201 | { |
3202 | int i; | |
3203 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 3204 | size_t alloc_size; |
ed8ece2e DH |
3205 | |
3206 | /* | |
3207 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3208 | * per zone. | |
3209 | */ | |
02b694de YG |
3210 | zone->wait_table_hash_nr_entries = |
3211 | wait_table_hash_nr_entries(zone_size_pages); | |
3212 | zone->wait_table_bits = | |
3213 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
3214 | alloc_size = zone->wait_table_hash_nr_entries |
3215 | * sizeof(wait_queue_head_t); | |
3216 | ||
cd94b9db | 3217 | if (!slab_is_available()) { |
cca448fe YG |
3218 | zone->wait_table = (wait_queue_head_t *) |
3219 | alloc_bootmem_node(pgdat, alloc_size); | |
3220 | } else { | |
3221 | /* | |
3222 | * This case means that a zone whose size was 0 gets new memory | |
3223 | * via memory hot-add. | |
3224 | * But it may be the case that a new node was hot-added. In | |
3225 | * this case vmalloc() will not be able to use this new node's | |
3226 | * memory - this wait_table must be initialized to use this new | |
3227 | * node itself as well. | |
3228 | * To use this new node's memory, further consideration will be | |
3229 | * necessary. | |
3230 | */ | |
8691f3a7 | 3231 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3232 | } |
3233 | if (!zone->wait_table) | |
3234 | return -ENOMEM; | |
ed8ece2e | 3235 | |
02b694de | 3236 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3237 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3238 | |
3239 | return 0; | |
ed8ece2e DH |
3240 | } |
3241 | ||
112067f0 SL |
3242 | static int __zone_pcp_update(void *data) |
3243 | { | |
3244 | struct zone *zone = data; | |
3245 | int cpu; | |
3246 | unsigned long batch = zone_batchsize(zone), flags; | |
3247 | ||
3248 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3249 | struct per_cpu_pageset *pset; | |
3250 | struct per_cpu_pages *pcp; | |
3251 | ||
3252 | pset = zone_pcp(zone, cpu); | |
3253 | pcp = &pset->pcp; | |
3254 | ||
3255 | local_irq_save(flags); | |
5f8dcc21 | 3256 | free_pcppages_bulk(zone, pcp->count, pcp); |
112067f0 SL |
3257 | setup_pageset(pset, batch); |
3258 | local_irq_restore(flags); | |
3259 | } | |
3260 | return 0; | |
3261 | } | |
3262 | ||
3263 | void zone_pcp_update(struct zone *zone) | |
3264 | { | |
3265 | stop_machine(__zone_pcp_update, zone, NULL); | |
3266 | } | |
3267 | ||
c09b4240 | 3268 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
3269 | { |
3270 | int cpu; | |
3271 | unsigned long batch = zone_batchsize(zone); | |
3272 | ||
3273 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3274 | #ifdef CONFIG_NUMA | |
3275 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 3276 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
3277 | setup_pageset(&boot_pageset[cpu],0); |
3278 | #else | |
3279 | setup_pageset(zone_pcp(zone,cpu), batch); | |
3280 | #endif | |
3281 | } | |
f5335c0f AB |
3282 | if (zone->present_pages) |
3283 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
3284 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
3285 | } |
3286 | ||
718127cc YG |
3287 | __meminit int init_currently_empty_zone(struct zone *zone, |
3288 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3289 | unsigned long size, |
3290 | enum memmap_context context) | |
ed8ece2e DH |
3291 | { |
3292 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3293 | int ret; |
3294 | ret = zone_wait_table_init(zone, size); | |
3295 | if (ret) | |
3296 | return ret; | |
ed8ece2e DH |
3297 | pgdat->nr_zones = zone_idx(zone) + 1; |
3298 | ||
ed8ece2e DH |
3299 | zone->zone_start_pfn = zone_start_pfn; |
3300 | ||
708614e6 MG |
3301 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3302 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3303 | pgdat->node_id, | |
3304 | (unsigned long)zone_idx(zone), | |
3305 | zone_start_pfn, (zone_start_pfn + size)); | |
3306 | ||
1e548deb | 3307 | zone_init_free_lists(zone); |
718127cc YG |
3308 | |
3309 | return 0; | |
ed8ece2e DH |
3310 | } |
3311 | ||
c713216d MG |
3312 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3313 | /* | |
3314 | * Basic iterator support. Return the first range of PFNs for a node | |
3315 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3316 | */ | |
a3142c8e | 3317 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
3318 | { |
3319 | int i; | |
3320 | ||
3321 | for (i = 0; i < nr_nodemap_entries; i++) | |
3322 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3323 | return i; | |
3324 | ||
3325 | return -1; | |
3326 | } | |
3327 | ||
3328 | /* | |
3329 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 3330 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 3331 | */ |
a3142c8e | 3332 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
3333 | { |
3334 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3335 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3336 | return index; | |
3337 | ||
3338 | return -1; | |
3339 | } | |
3340 | ||
3341 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3342 | /* | |
3343 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3344 | * Architectures may implement their own version but if add_active_range() | |
3345 | * was used and there are no special requirements, this is a convenient | |
3346 | * alternative | |
3347 | */ | |
f2dbcfa7 | 3348 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
3349 | { |
3350 | int i; | |
3351 | ||
3352 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3353 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3354 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3355 | ||
3356 | if (start_pfn <= pfn && pfn < end_pfn) | |
3357 | return early_node_map[i].nid; | |
3358 | } | |
cc2559bc KH |
3359 | /* This is a memory hole */ |
3360 | return -1; | |
c713216d MG |
3361 | } |
3362 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3363 | ||
f2dbcfa7 KH |
3364 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3365 | { | |
cc2559bc KH |
3366 | int nid; |
3367 | ||
3368 | nid = __early_pfn_to_nid(pfn); | |
3369 | if (nid >= 0) | |
3370 | return nid; | |
3371 | /* just returns 0 */ | |
3372 | return 0; | |
f2dbcfa7 KH |
3373 | } |
3374 | ||
cc2559bc KH |
3375 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3376 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3377 | { | |
3378 | int nid; | |
3379 | ||
3380 | nid = __early_pfn_to_nid(pfn); | |
3381 | if (nid >= 0 && nid != node) | |
3382 | return false; | |
3383 | return true; | |
3384 | } | |
3385 | #endif | |
f2dbcfa7 | 3386 | |
c713216d MG |
3387 | /* Basic iterator support to walk early_node_map[] */ |
3388 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3389 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3390 | i = next_active_region_index_in_nid(i, nid)) | |
3391 | ||
3392 | /** | |
3393 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3394 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3395 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3396 | * |
3397 | * If an architecture guarantees that all ranges registered with | |
3398 | * add_active_ranges() contain no holes and may be freed, this | |
3399 | * this function may be used instead of calling free_bootmem() manually. | |
3400 | */ | |
3401 | void __init free_bootmem_with_active_regions(int nid, | |
3402 | unsigned long max_low_pfn) | |
3403 | { | |
3404 | int i; | |
3405 | ||
3406 | for_each_active_range_index_in_nid(i, nid) { | |
3407 | unsigned long size_pages = 0; | |
3408 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3409 | ||
3410 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3411 | continue; | |
3412 | ||
3413 | if (end_pfn > max_low_pfn) | |
3414 | end_pfn = max_low_pfn; | |
3415 | ||
3416 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3417 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3418 | PFN_PHYS(early_node_map[i].start_pfn), | |
3419 | size_pages << PAGE_SHIFT); | |
3420 | } | |
3421 | } | |
3422 | ||
b5bc6c0e YL |
3423 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
3424 | { | |
3425 | int i; | |
d52d53b8 | 3426 | int ret; |
b5bc6c0e | 3427 | |
d52d53b8 YL |
3428 | for_each_active_range_index_in_nid(i, nid) { |
3429 | ret = work_fn(early_node_map[i].start_pfn, | |
3430 | early_node_map[i].end_pfn, data); | |
3431 | if (ret) | |
3432 | break; | |
3433 | } | |
b5bc6c0e | 3434 | } |
c713216d MG |
3435 | /** |
3436 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3437 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3438 | * |
3439 | * If an architecture guarantees that all ranges registered with | |
3440 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3441 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3442 | */ |
3443 | void __init sparse_memory_present_with_active_regions(int nid) | |
3444 | { | |
3445 | int i; | |
3446 | ||
3447 | for_each_active_range_index_in_nid(i, nid) | |
3448 | memory_present(early_node_map[i].nid, | |
3449 | early_node_map[i].start_pfn, | |
3450 | early_node_map[i].end_pfn); | |
3451 | } | |
3452 | ||
3453 | /** | |
3454 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3455 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3456 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3457 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3458 | * |
3459 | * It returns the start and end page frame of a node based on information | |
3460 | * provided by an arch calling add_active_range(). If called for a node | |
3461 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3462 | * PFNs will be 0. |
c713216d | 3463 | */ |
a3142c8e | 3464 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3465 | unsigned long *start_pfn, unsigned long *end_pfn) |
3466 | { | |
3467 | int i; | |
3468 | *start_pfn = -1UL; | |
3469 | *end_pfn = 0; | |
3470 | ||
3471 | for_each_active_range_index_in_nid(i, nid) { | |
3472 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3473 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3474 | } | |
3475 | ||
633c0666 | 3476 | if (*start_pfn == -1UL) |
c713216d | 3477 | *start_pfn = 0; |
c713216d MG |
3478 | } |
3479 | ||
2a1e274a MG |
3480 | /* |
3481 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3482 | * assumption is made that zones within a node are ordered in monotonic | |
3483 | * increasing memory addresses so that the "highest" populated zone is used | |
3484 | */ | |
b69a7288 | 3485 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3486 | { |
3487 | int zone_index; | |
3488 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3489 | if (zone_index == ZONE_MOVABLE) | |
3490 | continue; | |
3491 | ||
3492 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3493 | arch_zone_lowest_possible_pfn[zone_index]) | |
3494 | break; | |
3495 | } | |
3496 | ||
3497 | VM_BUG_ON(zone_index == -1); | |
3498 | movable_zone = zone_index; | |
3499 | } | |
3500 | ||
3501 | /* | |
3502 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3503 | * because it is sized independant of architecture. Unlike the other zones, | |
3504 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3505 | * in each node depending on the size of each node and how evenly kernelcore | |
3506 | * is distributed. This helper function adjusts the zone ranges | |
3507 | * provided by the architecture for a given node by using the end of the | |
3508 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3509 | * zones within a node are in order of monotonic increases memory addresses | |
3510 | */ | |
b69a7288 | 3511 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
3512 | unsigned long zone_type, |
3513 | unsigned long node_start_pfn, | |
3514 | unsigned long node_end_pfn, | |
3515 | unsigned long *zone_start_pfn, | |
3516 | unsigned long *zone_end_pfn) | |
3517 | { | |
3518 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3519 | if (zone_movable_pfn[nid]) { | |
3520 | /* Size ZONE_MOVABLE */ | |
3521 | if (zone_type == ZONE_MOVABLE) { | |
3522 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3523 | *zone_end_pfn = min(node_end_pfn, | |
3524 | arch_zone_highest_possible_pfn[movable_zone]); | |
3525 | ||
3526 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3527 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3528 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3529 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3530 | ||
3531 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3532 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3533 | *zone_start_pfn = *zone_end_pfn; | |
3534 | } | |
3535 | } | |
3536 | ||
c713216d MG |
3537 | /* |
3538 | * Return the number of pages a zone spans in a node, including holes | |
3539 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3540 | */ | |
6ea6e688 | 3541 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3542 | unsigned long zone_type, |
3543 | unsigned long *ignored) | |
3544 | { | |
3545 | unsigned long node_start_pfn, node_end_pfn; | |
3546 | unsigned long zone_start_pfn, zone_end_pfn; | |
3547 | ||
3548 | /* Get the start and end of the node and zone */ | |
3549 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3550 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3551 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3552 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3553 | node_start_pfn, node_end_pfn, | |
3554 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3555 | |
3556 | /* Check that this node has pages within the zone's required range */ | |
3557 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3558 | return 0; | |
3559 | ||
3560 | /* Move the zone boundaries inside the node if necessary */ | |
3561 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3562 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3563 | ||
3564 | /* Return the spanned pages */ | |
3565 | return zone_end_pfn - zone_start_pfn; | |
3566 | } | |
3567 | ||
3568 | /* | |
3569 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3570 | * then all holes in the requested range will be accounted for. |
c713216d | 3571 | */ |
b69a7288 | 3572 | static unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3573 | unsigned long range_start_pfn, |
3574 | unsigned long range_end_pfn) | |
3575 | { | |
3576 | int i = 0; | |
3577 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3578 | unsigned long start_pfn; | |
3579 | ||
3580 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3581 | i = first_active_region_index_in_nid(nid); | |
3582 | if (i == -1) | |
3583 | return 0; | |
3584 | ||
b5445f95 MG |
3585 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3586 | ||
9c7cd687 MG |
3587 | /* Account for ranges before physical memory on this node */ |
3588 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3589 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3590 | |
3591 | /* Find all holes for the zone within the node */ | |
3592 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3593 | ||
3594 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3595 | if (prev_end_pfn >= range_end_pfn) | |
3596 | break; | |
3597 | ||
3598 | /* Make sure the end of the zone is not within the hole */ | |
3599 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3600 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3601 | ||
3602 | /* Update the hole size cound and move on */ | |
3603 | if (start_pfn > range_start_pfn) { | |
3604 | BUG_ON(prev_end_pfn > start_pfn); | |
3605 | hole_pages += start_pfn - prev_end_pfn; | |
3606 | } | |
3607 | prev_end_pfn = early_node_map[i].end_pfn; | |
3608 | } | |
3609 | ||
9c7cd687 MG |
3610 | /* Account for ranges past physical memory on this node */ |
3611 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3612 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3613 | max(range_start_pfn, prev_end_pfn); |
3614 | ||
c713216d MG |
3615 | return hole_pages; |
3616 | } | |
3617 | ||
3618 | /** | |
3619 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3620 | * @start_pfn: The start PFN to start searching for holes | |
3621 | * @end_pfn: The end PFN to stop searching for holes | |
3622 | * | |
88ca3b94 | 3623 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3624 | */ |
3625 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3626 | unsigned long end_pfn) | |
3627 | { | |
3628 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3629 | } | |
3630 | ||
3631 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3632 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3633 | unsigned long zone_type, |
3634 | unsigned long *ignored) | |
3635 | { | |
9c7cd687 MG |
3636 | unsigned long node_start_pfn, node_end_pfn; |
3637 | unsigned long zone_start_pfn, zone_end_pfn; | |
3638 | ||
3639 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3640 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3641 | node_start_pfn); | |
3642 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3643 | node_end_pfn); | |
3644 | ||
2a1e274a MG |
3645 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3646 | node_start_pfn, node_end_pfn, | |
3647 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3648 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3649 | } |
0e0b864e | 3650 | |
c713216d | 3651 | #else |
6ea6e688 | 3652 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3653 | unsigned long zone_type, |
3654 | unsigned long *zones_size) | |
3655 | { | |
3656 | return zones_size[zone_type]; | |
3657 | } | |
3658 | ||
6ea6e688 | 3659 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3660 | unsigned long zone_type, |
3661 | unsigned long *zholes_size) | |
3662 | { | |
3663 | if (!zholes_size) | |
3664 | return 0; | |
3665 | ||
3666 | return zholes_size[zone_type]; | |
3667 | } | |
0e0b864e | 3668 | |
c713216d MG |
3669 | #endif |
3670 | ||
a3142c8e | 3671 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3672 | unsigned long *zones_size, unsigned long *zholes_size) |
3673 | { | |
3674 | unsigned long realtotalpages, totalpages = 0; | |
3675 | enum zone_type i; | |
3676 | ||
3677 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3678 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3679 | zones_size); | |
3680 | pgdat->node_spanned_pages = totalpages; | |
3681 | ||
3682 | realtotalpages = totalpages; | |
3683 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3684 | realtotalpages -= | |
3685 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3686 | zholes_size); | |
3687 | pgdat->node_present_pages = realtotalpages; | |
3688 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3689 | realtotalpages); | |
3690 | } | |
3691 | ||
835c134e MG |
3692 | #ifndef CONFIG_SPARSEMEM |
3693 | /* | |
3694 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3695 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3696 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3697 | * round what is now in bits to nearest long in bits, then return it in |
3698 | * bytes. | |
3699 | */ | |
3700 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3701 | { | |
3702 | unsigned long usemapsize; | |
3703 | ||
d9c23400 MG |
3704 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3705 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3706 | usemapsize *= NR_PAGEBLOCK_BITS; |
3707 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3708 | ||
3709 | return usemapsize / 8; | |
3710 | } | |
3711 | ||
3712 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3713 | struct zone *zone, unsigned long zonesize) | |
3714 | { | |
3715 | unsigned long usemapsize = usemap_size(zonesize); | |
3716 | zone->pageblock_flags = NULL; | |
58a01a45 | 3717 | if (usemapsize) |
835c134e | 3718 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
835c134e MG |
3719 | } |
3720 | #else | |
3721 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3722 | struct zone *zone, unsigned long zonesize) {} | |
3723 | #endif /* CONFIG_SPARSEMEM */ | |
3724 | ||
d9c23400 | 3725 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3726 | |
3727 | /* Return a sensible default order for the pageblock size. */ | |
3728 | static inline int pageblock_default_order(void) | |
3729 | { | |
3730 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3731 | return HUGETLB_PAGE_ORDER; | |
3732 | ||
3733 | return MAX_ORDER-1; | |
3734 | } | |
3735 | ||
d9c23400 MG |
3736 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3737 | static inline void __init set_pageblock_order(unsigned int order) | |
3738 | { | |
3739 | /* Check that pageblock_nr_pages has not already been setup */ | |
3740 | if (pageblock_order) | |
3741 | return; | |
3742 | ||
3743 | /* | |
3744 | * Assume the largest contiguous order of interest is a huge page. | |
3745 | * This value may be variable depending on boot parameters on IA64 | |
3746 | */ | |
3747 | pageblock_order = order; | |
3748 | } | |
3749 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3750 | ||
ba72cb8c MG |
3751 | /* |
3752 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3753 | * and pageblock_default_order() are unused as pageblock_order is set | |
3754 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3755 | * pageblock_order based on the kernel config | |
3756 | */ | |
3757 | static inline int pageblock_default_order(unsigned int order) | |
3758 | { | |
3759 | return MAX_ORDER-1; | |
3760 | } | |
d9c23400 MG |
3761 | #define set_pageblock_order(x) do {} while (0) |
3762 | ||
3763 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3764 | ||
1da177e4 LT |
3765 | /* |
3766 | * Set up the zone data structures: | |
3767 | * - mark all pages reserved | |
3768 | * - mark all memory queues empty | |
3769 | * - clear the memory bitmaps | |
3770 | */ | |
b5a0e011 | 3771 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3772 | unsigned long *zones_size, unsigned long *zholes_size) |
3773 | { | |
2f1b6248 | 3774 | enum zone_type j; |
ed8ece2e | 3775 | int nid = pgdat->node_id; |
1da177e4 | 3776 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3777 | int ret; |
1da177e4 | 3778 | |
208d54e5 | 3779 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3780 | pgdat->nr_zones = 0; |
3781 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3782 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 3783 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
3784 | |
3785 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3786 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3787 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 3788 | enum lru_list l; |
1da177e4 | 3789 | |
c713216d MG |
3790 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3791 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3792 | zholes_size); | |
1da177e4 | 3793 | |
0e0b864e MG |
3794 | /* |
3795 | * Adjust realsize so that it accounts for how much memory | |
3796 | * is used by this zone for memmap. This affects the watermark | |
3797 | * and per-cpu initialisations | |
3798 | */ | |
f7232154 JW |
3799 | memmap_pages = |
3800 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
3801 | if (realsize >= memmap_pages) { |
3802 | realsize -= memmap_pages; | |
5594c8c8 YL |
3803 | if (memmap_pages) |
3804 | printk(KERN_DEBUG | |
3805 | " %s zone: %lu pages used for memmap\n", | |
3806 | zone_names[j], memmap_pages); | |
0e0b864e MG |
3807 | } else |
3808 | printk(KERN_WARNING | |
3809 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3810 | zone_names[j], memmap_pages, realsize); | |
3811 | ||
6267276f CL |
3812 | /* Account for reserved pages */ |
3813 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3814 | realsize -= dma_reserve; |
d903ef9f | 3815 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 3816 | zone_names[0], dma_reserve); |
0e0b864e MG |
3817 | } |
3818 | ||
98d2b0eb | 3819 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3820 | nr_kernel_pages += realsize; |
3821 | nr_all_pages += realsize; | |
3822 | ||
3823 | zone->spanned_pages = size; | |
3824 | zone->present_pages = realsize; | |
9614634f | 3825 | #ifdef CONFIG_NUMA |
d5f541ed | 3826 | zone->node = nid; |
8417bba4 | 3827 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3828 | / 100; |
0ff38490 | 3829 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3830 | #endif |
1da177e4 LT |
3831 | zone->name = zone_names[j]; |
3832 | spin_lock_init(&zone->lock); | |
3833 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3834 | zone_seqlock_init(zone); |
1da177e4 | 3835 | zone->zone_pgdat = pgdat; |
1da177e4 | 3836 | |
3bb1a852 | 3837 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3838 | |
ed8ece2e | 3839 | zone_pcp_init(zone); |
b69408e8 CL |
3840 | for_each_lru(l) { |
3841 | INIT_LIST_HEAD(&zone->lru[l].list); | |
f8629631 | 3842 | zone->reclaim_stat.nr_saved_scan[l] = 0; |
b69408e8 | 3843 | } |
6e901571 KM |
3844 | zone->reclaim_stat.recent_rotated[0] = 0; |
3845 | zone->reclaim_stat.recent_rotated[1] = 0; | |
3846 | zone->reclaim_stat.recent_scanned[0] = 0; | |
3847 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 3848 | zap_zone_vm_stats(zone); |
e815af95 | 3849 | zone->flags = 0; |
1da177e4 LT |
3850 | if (!size) |
3851 | continue; | |
3852 | ||
ba72cb8c | 3853 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3854 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3855 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3856 | size, MEMMAP_EARLY); | |
718127cc | 3857 | BUG_ON(ret); |
76cdd58e | 3858 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 3859 | zone_start_pfn += size; |
1da177e4 LT |
3860 | } |
3861 | } | |
3862 | ||
577a32f6 | 3863 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3864 | { |
1da177e4 LT |
3865 | /* Skip empty nodes */ |
3866 | if (!pgdat->node_spanned_pages) | |
3867 | return; | |
3868 | ||
d41dee36 | 3869 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3870 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3871 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3872 | unsigned long size, start, end; |
d41dee36 AW |
3873 | struct page *map; |
3874 | ||
e984bb43 BP |
3875 | /* |
3876 | * The zone's endpoints aren't required to be MAX_ORDER | |
3877 | * aligned but the node_mem_map endpoints must be in order | |
3878 | * for the buddy allocator to function correctly. | |
3879 | */ | |
3880 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3881 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3882 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3883 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3884 | map = alloc_remap(pgdat->node_id, size); |
3885 | if (!map) | |
3886 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3887 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3888 | } |
12d810c1 | 3889 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3890 | /* |
3891 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3892 | */ | |
c713216d | 3893 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3894 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3895 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3896 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3897 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3898 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3899 | } | |
1da177e4 | 3900 | #endif |
d41dee36 | 3901 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3902 | } |
3903 | ||
9109fb7b JW |
3904 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
3905 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 3906 | { |
9109fb7b JW |
3907 | pg_data_t *pgdat = NODE_DATA(nid); |
3908 | ||
1da177e4 LT |
3909 | pgdat->node_id = nid; |
3910 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3911 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3912 | |
3913 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
3914 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
3915 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3916 | nid, (unsigned long)pgdat, | |
3917 | (unsigned long)pgdat->node_mem_map); | |
3918 | #endif | |
1da177e4 LT |
3919 | |
3920 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3921 | } | |
3922 | ||
c713216d | 3923 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3924 | |
3925 | #if MAX_NUMNODES > 1 | |
3926 | /* | |
3927 | * Figure out the number of possible node ids. | |
3928 | */ | |
3929 | static void __init setup_nr_node_ids(void) | |
3930 | { | |
3931 | unsigned int node; | |
3932 | unsigned int highest = 0; | |
3933 | ||
3934 | for_each_node_mask(node, node_possible_map) | |
3935 | highest = node; | |
3936 | nr_node_ids = highest + 1; | |
3937 | } | |
3938 | #else | |
3939 | static inline void setup_nr_node_ids(void) | |
3940 | { | |
3941 | } | |
3942 | #endif | |
3943 | ||
c713216d MG |
3944 | /** |
3945 | * add_active_range - Register a range of PFNs backed by physical memory | |
3946 | * @nid: The node ID the range resides on | |
3947 | * @start_pfn: The start PFN of the available physical memory | |
3948 | * @end_pfn: The end PFN of the available physical memory | |
3949 | * | |
3950 | * These ranges are stored in an early_node_map[] and later used by | |
3951 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3952 | * range spans a memory hole, it is up to the architecture to ensure | |
3953 | * the memory is not freed by the bootmem allocator. If possible | |
3954 | * the range being registered will be merged with existing ranges. | |
3955 | */ | |
3956 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3957 | unsigned long end_pfn) | |
3958 | { | |
3959 | int i; | |
3960 | ||
6b74ab97 MG |
3961 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
3962 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3963 | "%d entries of %d used\n", | |
3964 | nid, start_pfn, end_pfn, | |
3965 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 3966 | |
2dbb51c4 MG |
3967 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
3968 | ||
c713216d MG |
3969 | /* Merge with existing active regions if possible */ |
3970 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3971 | if (early_node_map[i].nid != nid) | |
3972 | continue; | |
3973 | ||
3974 | /* Skip if an existing region covers this new one */ | |
3975 | if (start_pfn >= early_node_map[i].start_pfn && | |
3976 | end_pfn <= early_node_map[i].end_pfn) | |
3977 | return; | |
3978 | ||
3979 | /* Merge forward if suitable */ | |
3980 | if (start_pfn <= early_node_map[i].end_pfn && | |
3981 | end_pfn > early_node_map[i].end_pfn) { | |
3982 | early_node_map[i].end_pfn = end_pfn; | |
3983 | return; | |
3984 | } | |
3985 | ||
3986 | /* Merge backward if suitable */ | |
3987 | if (start_pfn < early_node_map[i].end_pfn && | |
3988 | end_pfn >= early_node_map[i].start_pfn) { | |
3989 | early_node_map[i].start_pfn = start_pfn; | |
3990 | return; | |
3991 | } | |
3992 | } | |
3993 | ||
3994 | /* Check that early_node_map is large enough */ | |
3995 | if (i >= MAX_ACTIVE_REGIONS) { | |
3996 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3997 | MAX_ACTIVE_REGIONS); | |
3998 | return; | |
3999 | } | |
4000 | ||
4001 | early_node_map[i].nid = nid; | |
4002 | early_node_map[i].start_pfn = start_pfn; | |
4003 | early_node_map[i].end_pfn = end_pfn; | |
4004 | nr_nodemap_entries = i + 1; | |
4005 | } | |
4006 | ||
4007 | /** | |
cc1050ba | 4008 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 4009 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
4010 | * @start_pfn: The new PFN of the range |
4011 | * @end_pfn: The new PFN of the range | |
c713216d MG |
4012 | * |
4013 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
4014 | * The map is kept near the end physical page range that has already been |
4015 | * registered. This function allows an arch to shrink an existing registered | |
4016 | * range. | |
c713216d | 4017 | */ |
cc1050ba YL |
4018 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
4019 | unsigned long end_pfn) | |
c713216d | 4020 | { |
cc1a9d86 YL |
4021 | int i, j; |
4022 | int removed = 0; | |
c713216d | 4023 | |
cc1050ba YL |
4024 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
4025 | nid, start_pfn, end_pfn); | |
4026 | ||
c713216d | 4027 | /* Find the old active region end and shrink */ |
cc1a9d86 | 4028 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
4029 | if (early_node_map[i].start_pfn >= start_pfn && |
4030 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 4031 | /* clear it */ |
cc1050ba | 4032 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
4033 | early_node_map[i].end_pfn = 0; |
4034 | removed = 1; | |
4035 | continue; | |
4036 | } | |
cc1050ba YL |
4037 | if (early_node_map[i].start_pfn < start_pfn && |
4038 | early_node_map[i].end_pfn > start_pfn) { | |
4039 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
4040 | early_node_map[i].end_pfn = start_pfn; | |
4041 | if (temp_end_pfn > end_pfn) | |
4042 | add_active_range(nid, end_pfn, temp_end_pfn); | |
4043 | continue; | |
4044 | } | |
4045 | if (early_node_map[i].start_pfn >= start_pfn && | |
4046 | early_node_map[i].end_pfn > end_pfn && | |
4047 | early_node_map[i].start_pfn < end_pfn) { | |
4048 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 4049 | continue; |
c713216d | 4050 | } |
cc1a9d86 YL |
4051 | } |
4052 | ||
4053 | if (!removed) | |
4054 | return; | |
4055 | ||
4056 | /* remove the blank ones */ | |
4057 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
4058 | if (early_node_map[i].nid != nid) | |
4059 | continue; | |
4060 | if (early_node_map[i].end_pfn) | |
4061 | continue; | |
4062 | /* we found it, get rid of it */ | |
4063 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
4064 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
4065 | sizeof(early_node_map[j])); | |
4066 | j = nr_nodemap_entries - 1; | |
4067 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
4068 | nr_nodemap_entries--; | |
4069 | } | |
c713216d MG |
4070 | } |
4071 | ||
4072 | /** | |
4073 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 4074 | * |
c713216d MG |
4075 | * During discovery, it may be found that a table like SRAT is invalid |
4076 | * and an alternative discovery method must be used. This function removes | |
4077 | * all currently registered regions. | |
4078 | */ | |
88ca3b94 | 4079 | void __init remove_all_active_ranges(void) |
c713216d MG |
4080 | { |
4081 | memset(early_node_map, 0, sizeof(early_node_map)); | |
4082 | nr_nodemap_entries = 0; | |
4083 | } | |
4084 | ||
4085 | /* Compare two active node_active_regions */ | |
4086 | static int __init cmp_node_active_region(const void *a, const void *b) | |
4087 | { | |
4088 | struct node_active_region *arange = (struct node_active_region *)a; | |
4089 | struct node_active_region *brange = (struct node_active_region *)b; | |
4090 | ||
4091 | /* Done this way to avoid overflows */ | |
4092 | if (arange->start_pfn > brange->start_pfn) | |
4093 | return 1; | |
4094 | if (arange->start_pfn < brange->start_pfn) | |
4095 | return -1; | |
4096 | ||
4097 | return 0; | |
4098 | } | |
4099 | ||
4100 | /* sort the node_map by start_pfn */ | |
4101 | static void __init sort_node_map(void) | |
4102 | { | |
4103 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
4104 | sizeof(struct node_active_region), | |
4105 | cmp_node_active_region, NULL); | |
4106 | } | |
4107 | ||
a6af2bc3 | 4108 | /* Find the lowest pfn for a node */ |
b69a7288 | 4109 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
4110 | { |
4111 | int i; | |
a6af2bc3 | 4112 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 4113 | |
c713216d MG |
4114 | /* Assuming a sorted map, the first range found has the starting pfn */ |
4115 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 4116 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 4117 | |
a6af2bc3 MG |
4118 | if (min_pfn == ULONG_MAX) { |
4119 | printk(KERN_WARNING | |
2bc0d261 | 4120 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4121 | return 0; |
4122 | } | |
4123 | ||
4124 | return min_pfn; | |
c713216d MG |
4125 | } |
4126 | ||
4127 | /** | |
4128 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4129 | * | |
4130 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4131 | * add_active_range(). |
c713216d MG |
4132 | */ |
4133 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4134 | { | |
4135 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4136 | } | |
4137 | ||
37b07e41 LS |
4138 | /* |
4139 | * early_calculate_totalpages() | |
4140 | * Sum pages in active regions for movable zone. | |
4141 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4142 | */ | |
484f51f8 | 4143 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
4144 | { |
4145 | int i; | |
4146 | unsigned long totalpages = 0; | |
4147 | ||
37b07e41 LS |
4148 | for (i = 0; i < nr_nodemap_entries; i++) { |
4149 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 4150 | early_node_map[i].start_pfn; |
37b07e41 LS |
4151 | totalpages += pages; |
4152 | if (pages) | |
4153 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
4154 | } | |
4155 | return totalpages; | |
7e63efef MG |
4156 | } |
4157 | ||
2a1e274a MG |
4158 | /* |
4159 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4160 | * is spread evenly between nodes as long as the nodes have enough | |
4161 | * memory. When they don't, some nodes will have more kernelcore than | |
4162 | * others | |
4163 | */ | |
b69a7288 | 4164 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
4165 | { |
4166 | int i, nid; | |
4167 | unsigned long usable_startpfn; | |
4168 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4169 | /* save the state before borrow the nodemask */ |
4170 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4171 | unsigned long totalpages = early_calculate_totalpages(); |
4172 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4173 | |
7e63efef MG |
4174 | /* |
4175 | * If movablecore was specified, calculate what size of | |
4176 | * kernelcore that corresponds so that memory usable for | |
4177 | * any allocation type is evenly spread. If both kernelcore | |
4178 | * and movablecore are specified, then the value of kernelcore | |
4179 | * will be used for required_kernelcore if it's greater than | |
4180 | * what movablecore would have allowed. | |
4181 | */ | |
4182 | if (required_movablecore) { | |
7e63efef MG |
4183 | unsigned long corepages; |
4184 | ||
4185 | /* | |
4186 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4187 | * was requested by the user | |
4188 | */ | |
4189 | required_movablecore = | |
4190 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4191 | corepages = totalpages - required_movablecore; | |
4192 | ||
4193 | required_kernelcore = max(required_kernelcore, corepages); | |
4194 | } | |
4195 | ||
2a1e274a MG |
4196 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4197 | if (!required_kernelcore) | |
66918dcd | 4198 | goto out; |
2a1e274a MG |
4199 | |
4200 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4201 | find_usable_zone_for_movable(); | |
4202 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4203 | ||
4204 | restart: | |
4205 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4206 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4207 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
4208 | /* |
4209 | * Recalculate kernelcore_node if the division per node | |
4210 | * now exceeds what is necessary to satisfy the requested | |
4211 | * amount of memory for the kernel | |
4212 | */ | |
4213 | if (required_kernelcore < kernelcore_node) | |
4214 | kernelcore_node = required_kernelcore / usable_nodes; | |
4215 | ||
4216 | /* | |
4217 | * As the map is walked, we track how much memory is usable | |
4218 | * by the kernel using kernelcore_remaining. When it is | |
4219 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4220 | */ | |
4221 | kernelcore_remaining = kernelcore_node; | |
4222 | ||
4223 | /* Go through each range of PFNs within this node */ | |
4224 | for_each_active_range_index_in_nid(i, nid) { | |
4225 | unsigned long start_pfn, end_pfn; | |
4226 | unsigned long size_pages; | |
4227 | ||
4228 | start_pfn = max(early_node_map[i].start_pfn, | |
4229 | zone_movable_pfn[nid]); | |
4230 | end_pfn = early_node_map[i].end_pfn; | |
4231 | if (start_pfn >= end_pfn) | |
4232 | continue; | |
4233 | ||
4234 | /* Account for what is only usable for kernelcore */ | |
4235 | if (start_pfn < usable_startpfn) { | |
4236 | unsigned long kernel_pages; | |
4237 | kernel_pages = min(end_pfn, usable_startpfn) | |
4238 | - start_pfn; | |
4239 | ||
4240 | kernelcore_remaining -= min(kernel_pages, | |
4241 | kernelcore_remaining); | |
4242 | required_kernelcore -= min(kernel_pages, | |
4243 | required_kernelcore); | |
4244 | ||
4245 | /* Continue if range is now fully accounted */ | |
4246 | if (end_pfn <= usable_startpfn) { | |
4247 | ||
4248 | /* | |
4249 | * Push zone_movable_pfn to the end so | |
4250 | * that if we have to rebalance | |
4251 | * kernelcore across nodes, we will | |
4252 | * not double account here | |
4253 | */ | |
4254 | zone_movable_pfn[nid] = end_pfn; | |
4255 | continue; | |
4256 | } | |
4257 | start_pfn = usable_startpfn; | |
4258 | } | |
4259 | ||
4260 | /* | |
4261 | * The usable PFN range for ZONE_MOVABLE is from | |
4262 | * start_pfn->end_pfn. Calculate size_pages as the | |
4263 | * number of pages used as kernelcore | |
4264 | */ | |
4265 | size_pages = end_pfn - start_pfn; | |
4266 | if (size_pages > kernelcore_remaining) | |
4267 | size_pages = kernelcore_remaining; | |
4268 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4269 | ||
4270 | /* | |
4271 | * Some kernelcore has been met, update counts and | |
4272 | * break if the kernelcore for this node has been | |
4273 | * satisified | |
4274 | */ | |
4275 | required_kernelcore -= min(required_kernelcore, | |
4276 | size_pages); | |
4277 | kernelcore_remaining -= size_pages; | |
4278 | if (!kernelcore_remaining) | |
4279 | break; | |
4280 | } | |
4281 | } | |
4282 | ||
4283 | /* | |
4284 | * If there is still required_kernelcore, we do another pass with one | |
4285 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4286 | * along on the nodes that still have memory until kernelcore is | |
4287 | * satisified | |
4288 | */ | |
4289 | usable_nodes--; | |
4290 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4291 | goto restart; | |
4292 | ||
4293 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4294 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4295 | zone_movable_pfn[nid] = | |
4296 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4297 | |
4298 | out: | |
4299 | /* restore the node_state */ | |
4300 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4301 | } |
4302 | ||
37b07e41 LS |
4303 | /* Any regular memory on that node ? */ |
4304 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4305 | { | |
4306 | #ifdef CONFIG_HIGHMEM | |
4307 | enum zone_type zone_type; | |
4308 | ||
4309 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4310 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4311 | if (zone->present_pages) | |
4312 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4313 | } | |
4314 | #endif | |
4315 | } | |
4316 | ||
c713216d MG |
4317 | /** |
4318 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4319 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4320 | * |
4321 | * This will call free_area_init_node() for each active node in the system. | |
4322 | * Using the page ranges provided by add_active_range(), the size of each | |
4323 | * zone in each node and their holes is calculated. If the maximum PFN | |
4324 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4325 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4326 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4327 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4328 | * at arch_max_dma_pfn. | |
4329 | */ | |
4330 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4331 | { | |
4332 | unsigned long nid; | |
db99100d | 4333 | int i; |
c713216d | 4334 | |
a6af2bc3 MG |
4335 | /* Sort early_node_map as initialisation assumes it is sorted */ |
4336 | sort_node_map(); | |
4337 | ||
c713216d MG |
4338 | /* Record where the zone boundaries are */ |
4339 | memset(arch_zone_lowest_possible_pfn, 0, | |
4340 | sizeof(arch_zone_lowest_possible_pfn)); | |
4341 | memset(arch_zone_highest_possible_pfn, 0, | |
4342 | sizeof(arch_zone_highest_possible_pfn)); | |
4343 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4344 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4345 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4346 | if (i == ZONE_MOVABLE) |
4347 | continue; | |
c713216d MG |
4348 | arch_zone_lowest_possible_pfn[i] = |
4349 | arch_zone_highest_possible_pfn[i-1]; | |
4350 | arch_zone_highest_possible_pfn[i] = | |
4351 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4352 | } | |
2a1e274a MG |
4353 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4354 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4355 | ||
4356 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4357 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4358 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4359 | |
c713216d MG |
4360 | /* Print out the zone ranges */ |
4361 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4362 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4363 | if (i == ZONE_MOVABLE) | |
4364 | continue; | |
5dab8ec1 | 4365 | printk(" %-8s %0#10lx -> %0#10lx\n", |
c713216d MG |
4366 | zone_names[i], |
4367 | arch_zone_lowest_possible_pfn[i], | |
4368 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4369 | } |
4370 | ||
4371 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4372 | printk("Movable zone start PFN for each node\n"); | |
4373 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4374 | if (zone_movable_pfn[i]) | |
4375 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4376 | } | |
c713216d MG |
4377 | |
4378 | /* Print out the early_node_map[] */ | |
4379 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4380 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 4381 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
4382 | early_node_map[i].start_pfn, |
4383 | early_node_map[i].end_pfn); | |
4384 | ||
4385 | /* Initialise every node */ | |
708614e6 | 4386 | mminit_verify_pageflags_layout(); |
8ef82866 | 4387 | setup_nr_node_ids(); |
c713216d MG |
4388 | for_each_online_node(nid) { |
4389 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4390 | free_area_init_node(nid, NULL, |
c713216d | 4391 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4392 | |
4393 | /* Any memory on that node */ | |
4394 | if (pgdat->node_present_pages) | |
4395 | node_set_state(nid, N_HIGH_MEMORY); | |
4396 | check_for_regular_memory(pgdat); | |
c713216d MG |
4397 | } |
4398 | } | |
2a1e274a | 4399 | |
7e63efef | 4400 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4401 | { |
4402 | unsigned long long coremem; | |
4403 | if (!p) | |
4404 | return -EINVAL; | |
4405 | ||
4406 | coremem = memparse(p, &p); | |
7e63efef | 4407 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4408 | |
7e63efef | 4409 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4410 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4411 | ||
4412 | return 0; | |
4413 | } | |
ed7ed365 | 4414 | |
7e63efef MG |
4415 | /* |
4416 | * kernelcore=size sets the amount of memory for use for allocations that | |
4417 | * cannot be reclaimed or migrated. | |
4418 | */ | |
4419 | static int __init cmdline_parse_kernelcore(char *p) | |
4420 | { | |
4421 | return cmdline_parse_core(p, &required_kernelcore); | |
4422 | } | |
4423 | ||
4424 | /* | |
4425 | * movablecore=size sets the amount of memory for use for allocations that | |
4426 | * can be reclaimed or migrated. | |
4427 | */ | |
4428 | static int __init cmdline_parse_movablecore(char *p) | |
4429 | { | |
4430 | return cmdline_parse_core(p, &required_movablecore); | |
4431 | } | |
4432 | ||
ed7ed365 | 4433 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4434 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4435 | |
c713216d MG |
4436 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4437 | ||
0e0b864e | 4438 | /** |
88ca3b94 RD |
4439 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4440 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4441 | * |
4442 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4443 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4444 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4445 | * function may optionally be used to account for unfreeable pages in the |
4446 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4447 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4448 | */ |
4449 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4450 | { | |
4451 | dma_reserve = new_dma_reserve; | |
4452 | } | |
4453 | ||
93b7504e | 4454 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
52765583 | 4455 | struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] }; |
1da177e4 | 4456 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4457 | #endif |
1da177e4 LT |
4458 | |
4459 | void __init free_area_init(unsigned long *zones_size) | |
4460 | { | |
9109fb7b | 4461 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4462 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4463 | } | |
1da177e4 | 4464 | |
1da177e4 LT |
4465 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4466 | unsigned long action, void *hcpu) | |
4467 | { | |
4468 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4469 | |
8bb78442 | 4470 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4471 | drain_pages(cpu); |
4472 | ||
4473 | /* | |
4474 | * Spill the event counters of the dead processor | |
4475 | * into the current processors event counters. | |
4476 | * This artificially elevates the count of the current | |
4477 | * processor. | |
4478 | */ | |
f8891e5e | 4479 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4480 | |
4481 | /* | |
4482 | * Zero the differential counters of the dead processor | |
4483 | * so that the vm statistics are consistent. | |
4484 | * | |
4485 | * This is only okay since the processor is dead and cannot | |
4486 | * race with what we are doing. | |
4487 | */ | |
2244b95a | 4488 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4489 | } |
4490 | return NOTIFY_OK; | |
4491 | } | |
1da177e4 LT |
4492 | |
4493 | void __init page_alloc_init(void) | |
4494 | { | |
4495 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4496 | } | |
4497 | ||
cb45b0e9 HA |
4498 | /* |
4499 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4500 | * or min_free_kbytes changes. | |
4501 | */ | |
4502 | static void calculate_totalreserve_pages(void) | |
4503 | { | |
4504 | struct pglist_data *pgdat; | |
4505 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4506 | enum zone_type i, j; |
cb45b0e9 HA |
4507 | |
4508 | for_each_online_pgdat(pgdat) { | |
4509 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4510 | struct zone *zone = pgdat->node_zones + i; | |
4511 | unsigned long max = 0; | |
4512 | ||
4513 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4514 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4515 | if (zone->lowmem_reserve[j] > max) | |
4516 | max = zone->lowmem_reserve[j]; | |
4517 | } | |
4518 | ||
41858966 MG |
4519 | /* we treat the high watermark as reserved pages. */ |
4520 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
4521 | |
4522 | if (max > zone->present_pages) | |
4523 | max = zone->present_pages; | |
4524 | reserve_pages += max; | |
4525 | } | |
4526 | } | |
4527 | totalreserve_pages = reserve_pages; | |
4528 | } | |
4529 | ||
1da177e4 LT |
4530 | /* |
4531 | * setup_per_zone_lowmem_reserve - called whenever | |
4532 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4533 | * has a correct pages reserved value, so an adequate number of | |
4534 | * pages are left in the zone after a successful __alloc_pages(). | |
4535 | */ | |
4536 | static void setup_per_zone_lowmem_reserve(void) | |
4537 | { | |
4538 | struct pglist_data *pgdat; | |
2f6726e5 | 4539 | enum zone_type j, idx; |
1da177e4 | 4540 | |
ec936fc5 | 4541 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4542 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4543 | struct zone *zone = pgdat->node_zones + j; | |
4544 | unsigned long present_pages = zone->present_pages; | |
4545 | ||
4546 | zone->lowmem_reserve[j] = 0; | |
4547 | ||
2f6726e5 CL |
4548 | idx = j; |
4549 | while (idx) { | |
1da177e4 LT |
4550 | struct zone *lower_zone; |
4551 | ||
2f6726e5 CL |
4552 | idx--; |
4553 | ||
1da177e4 LT |
4554 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4555 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4556 | ||
4557 | lower_zone = pgdat->node_zones + idx; | |
4558 | lower_zone->lowmem_reserve[j] = present_pages / | |
4559 | sysctl_lowmem_reserve_ratio[idx]; | |
4560 | present_pages += lower_zone->present_pages; | |
4561 | } | |
4562 | } | |
4563 | } | |
cb45b0e9 HA |
4564 | |
4565 | /* update totalreserve_pages */ | |
4566 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4567 | } |
4568 | ||
88ca3b94 | 4569 | /** |
bc75d33f | 4570 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
bce7394a | 4571 | * or when memory is hot-{added|removed} |
88ca3b94 | 4572 | * |
bc75d33f MK |
4573 | * Ensures that the watermark[min,low,high] values for each zone are set |
4574 | * correctly with respect to min_free_kbytes. | |
1da177e4 | 4575 | */ |
bc75d33f | 4576 | void setup_per_zone_wmarks(void) |
1da177e4 LT |
4577 | { |
4578 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4579 | unsigned long lowmem_pages = 0; | |
4580 | struct zone *zone; | |
4581 | unsigned long flags; | |
4582 | ||
4583 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4584 | for_each_zone(zone) { | |
4585 | if (!is_highmem(zone)) | |
4586 | lowmem_pages += zone->present_pages; | |
4587 | } | |
4588 | ||
4589 | for_each_zone(zone) { | |
ac924c60 AM |
4590 | u64 tmp; |
4591 | ||
1125b4e3 | 4592 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4593 | tmp = (u64)pages_min * zone->present_pages; |
4594 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4595 | if (is_highmem(zone)) { |
4596 | /* | |
669ed175 NP |
4597 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4598 | * need highmem pages, so cap pages_min to a small | |
4599 | * value here. | |
4600 | * | |
41858966 | 4601 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
4602 | * deltas controls asynch page reclaim, and so should |
4603 | * not be capped for highmem. | |
1da177e4 LT |
4604 | */ |
4605 | int min_pages; | |
4606 | ||
4607 | min_pages = zone->present_pages / 1024; | |
4608 | if (min_pages < SWAP_CLUSTER_MAX) | |
4609 | min_pages = SWAP_CLUSTER_MAX; | |
4610 | if (min_pages > 128) | |
4611 | min_pages = 128; | |
41858966 | 4612 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 4613 | } else { |
669ed175 NP |
4614 | /* |
4615 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4616 | * proportionate to the zone's size. |
4617 | */ | |
41858966 | 4618 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
4619 | } |
4620 | ||
41858966 MG |
4621 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
4622 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
56fd56b8 | 4623 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4624 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4625 | } |
cb45b0e9 HA |
4626 | |
4627 | /* update totalreserve_pages */ | |
4628 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4629 | } |
4630 | ||
55a4462a | 4631 | /* |
556adecb RR |
4632 | * The inactive anon list should be small enough that the VM never has to |
4633 | * do too much work, but large enough that each inactive page has a chance | |
4634 | * to be referenced again before it is swapped out. | |
4635 | * | |
4636 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4637 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4638 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4639 | * the anonymous pages are kept on the inactive list. | |
4640 | * | |
4641 | * total target max | |
4642 | * memory ratio inactive anon | |
4643 | * ------------------------------------- | |
4644 | * 10MB 1 5MB | |
4645 | * 100MB 1 50MB | |
4646 | * 1GB 3 250MB | |
4647 | * 10GB 10 0.9GB | |
4648 | * 100GB 31 3GB | |
4649 | * 1TB 101 10GB | |
4650 | * 10TB 320 32GB | |
4651 | */ | |
96cb4df5 | 4652 | void calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 4653 | { |
96cb4df5 | 4654 | unsigned int gb, ratio; |
556adecb | 4655 | |
96cb4df5 MK |
4656 | /* Zone size in gigabytes */ |
4657 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
4658 | if (gb) | |
556adecb | 4659 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
4660 | else |
4661 | ratio = 1; | |
556adecb | 4662 | |
96cb4df5 MK |
4663 | zone->inactive_ratio = ratio; |
4664 | } | |
556adecb | 4665 | |
96cb4df5 MK |
4666 | static void __init setup_per_zone_inactive_ratio(void) |
4667 | { | |
4668 | struct zone *zone; | |
4669 | ||
4670 | for_each_zone(zone) | |
4671 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
4672 | } |
4673 | ||
1da177e4 LT |
4674 | /* |
4675 | * Initialise min_free_kbytes. | |
4676 | * | |
4677 | * For small machines we want it small (128k min). For large machines | |
4678 | * we want it large (64MB max). But it is not linear, because network | |
4679 | * bandwidth does not increase linearly with machine size. We use | |
4680 | * | |
4681 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4682 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4683 | * | |
4684 | * which yields | |
4685 | * | |
4686 | * 16MB: 512k | |
4687 | * 32MB: 724k | |
4688 | * 64MB: 1024k | |
4689 | * 128MB: 1448k | |
4690 | * 256MB: 2048k | |
4691 | * 512MB: 2896k | |
4692 | * 1024MB: 4096k | |
4693 | * 2048MB: 5792k | |
4694 | * 4096MB: 8192k | |
4695 | * 8192MB: 11584k | |
4696 | * 16384MB: 16384k | |
4697 | */ | |
bc75d33f | 4698 | static int __init init_per_zone_wmark_min(void) |
1da177e4 LT |
4699 | { |
4700 | unsigned long lowmem_kbytes; | |
4701 | ||
4702 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4703 | ||
4704 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4705 | if (min_free_kbytes < 128) | |
4706 | min_free_kbytes = 128; | |
4707 | if (min_free_kbytes > 65536) | |
4708 | min_free_kbytes = 65536; | |
bc75d33f | 4709 | setup_per_zone_wmarks(); |
1da177e4 | 4710 | setup_per_zone_lowmem_reserve(); |
556adecb | 4711 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
4712 | return 0; |
4713 | } | |
bc75d33f | 4714 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
4715 | |
4716 | /* | |
4717 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4718 | * that we can call two helper functions whenever min_free_kbytes | |
4719 | * changes. | |
4720 | */ | |
4721 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4722 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 4723 | { |
8d65af78 | 4724 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 4725 | if (write) |
bc75d33f | 4726 | setup_per_zone_wmarks(); |
1da177e4 LT |
4727 | return 0; |
4728 | } | |
4729 | ||
9614634f CL |
4730 | #ifdef CONFIG_NUMA |
4731 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4732 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
4733 | { |
4734 | struct zone *zone; | |
4735 | int rc; | |
4736 | ||
8d65af78 | 4737 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
4738 | if (rc) |
4739 | return rc; | |
4740 | ||
4741 | for_each_zone(zone) | |
8417bba4 | 4742 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4743 | sysctl_min_unmapped_ratio) / 100; |
4744 | return 0; | |
4745 | } | |
0ff38490 CL |
4746 | |
4747 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4748 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
4749 | { |
4750 | struct zone *zone; | |
4751 | int rc; | |
4752 | ||
8d65af78 | 4753 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
4754 | if (rc) |
4755 | return rc; | |
4756 | ||
4757 | for_each_zone(zone) | |
4758 | zone->min_slab_pages = (zone->present_pages * | |
4759 | sysctl_min_slab_ratio) / 100; | |
4760 | return 0; | |
4761 | } | |
9614634f CL |
4762 | #endif |
4763 | ||
1da177e4 LT |
4764 | /* |
4765 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4766 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4767 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4768 | * | |
4769 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 4770 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
4771 | * if in function of the boot time zone sizes. |
4772 | */ | |
4773 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4774 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 4775 | { |
8d65af78 | 4776 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
4777 | setup_per_zone_lowmem_reserve(); |
4778 | return 0; | |
4779 | } | |
4780 | ||
8ad4b1fb RS |
4781 | /* |
4782 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4783 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4784 | * can have before it gets flushed back to buddy allocator. | |
4785 | */ | |
4786 | ||
4787 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 4788 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
4789 | { |
4790 | struct zone *zone; | |
4791 | unsigned int cpu; | |
4792 | int ret; | |
4793 | ||
8d65af78 | 4794 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
8ad4b1fb RS |
4795 | if (!write || (ret == -EINVAL)) |
4796 | return ret; | |
364df0eb | 4797 | for_each_populated_zone(zone) { |
8ad4b1fb RS |
4798 | for_each_online_cpu(cpu) { |
4799 | unsigned long high; | |
4800 | high = zone->present_pages / percpu_pagelist_fraction; | |
4801 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4802 | } | |
4803 | } | |
4804 | return 0; | |
4805 | } | |
4806 | ||
f034b5d4 | 4807 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4808 | |
4809 | #ifdef CONFIG_NUMA | |
4810 | static int __init set_hashdist(char *str) | |
4811 | { | |
4812 | if (!str) | |
4813 | return 0; | |
4814 | hashdist = simple_strtoul(str, &str, 0); | |
4815 | return 1; | |
4816 | } | |
4817 | __setup("hashdist=", set_hashdist); | |
4818 | #endif | |
4819 | ||
4820 | /* | |
4821 | * allocate a large system hash table from bootmem | |
4822 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4823 | * quantity of entries | |
4824 | * - limit is the number of hash buckets, not the total allocation size | |
4825 | */ | |
4826 | void *__init alloc_large_system_hash(const char *tablename, | |
4827 | unsigned long bucketsize, | |
4828 | unsigned long numentries, | |
4829 | int scale, | |
4830 | int flags, | |
4831 | unsigned int *_hash_shift, | |
4832 | unsigned int *_hash_mask, | |
4833 | unsigned long limit) | |
4834 | { | |
4835 | unsigned long long max = limit; | |
4836 | unsigned long log2qty, size; | |
4837 | void *table = NULL; | |
4838 | ||
4839 | /* allow the kernel cmdline to have a say */ | |
4840 | if (!numentries) { | |
4841 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4842 | numentries = nr_kernel_pages; |
1da177e4 LT |
4843 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4844 | numentries >>= 20 - PAGE_SHIFT; | |
4845 | numentries <<= 20 - PAGE_SHIFT; | |
4846 | ||
4847 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4848 | if (scale > PAGE_SHIFT) | |
4849 | numentries >>= (scale - PAGE_SHIFT); | |
4850 | else | |
4851 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4852 | |
4853 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
4854 | if (unlikely(flags & HASH_SMALL)) { |
4855 | /* Makes no sense without HASH_EARLY */ | |
4856 | WARN_ON(!(flags & HASH_EARLY)); | |
4857 | if (!(numentries >> *_hash_shift)) { | |
4858 | numentries = 1UL << *_hash_shift; | |
4859 | BUG_ON(!numentries); | |
4860 | } | |
4861 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 4862 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 4863 | } |
6e692ed3 | 4864 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4865 | |
4866 | /* limit allocation size to 1/16 total memory by default */ | |
4867 | if (max == 0) { | |
4868 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4869 | do_div(max, bucketsize); | |
4870 | } | |
4871 | ||
4872 | if (numentries > max) | |
4873 | numentries = max; | |
4874 | ||
f0d1b0b3 | 4875 | log2qty = ilog2(numentries); |
1da177e4 LT |
4876 | |
4877 | do { | |
4878 | size = bucketsize << log2qty; | |
4879 | if (flags & HASH_EARLY) | |
74768ed8 | 4880 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
4881 | else if (hashdist) |
4882 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4883 | else { | |
1037b83b ED |
4884 | /* |
4885 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
4886 | * some pages at the end of hash table which |
4887 | * alloc_pages_exact() automatically does | |
1037b83b | 4888 | */ |
264ef8a9 | 4889 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 4890 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
4891 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
4892 | } | |
1da177e4 LT |
4893 | } |
4894 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4895 | ||
4896 | if (!table) | |
4897 | panic("Failed to allocate %s hash table\n", tablename); | |
4898 | ||
b49ad484 | 4899 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4900 | tablename, |
4901 | (1U << log2qty), | |
f0d1b0b3 | 4902 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4903 | size); |
4904 | ||
4905 | if (_hash_shift) | |
4906 | *_hash_shift = log2qty; | |
4907 | if (_hash_mask) | |
4908 | *_hash_mask = (1 << log2qty) - 1; | |
4909 | ||
4910 | return table; | |
4911 | } | |
a117e66e | 4912 | |
835c134e MG |
4913 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4914 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4915 | unsigned long pfn) | |
4916 | { | |
4917 | #ifdef CONFIG_SPARSEMEM | |
4918 | return __pfn_to_section(pfn)->pageblock_flags; | |
4919 | #else | |
4920 | return zone->pageblock_flags; | |
4921 | #endif /* CONFIG_SPARSEMEM */ | |
4922 | } | |
4923 | ||
4924 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4925 | { | |
4926 | #ifdef CONFIG_SPARSEMEM | |
4927 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4928 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4929 | #else |
4930 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4931 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4932 | #endif /* CONFIG_SPARSEMEM */ |
4933 | } | |
4934 | ||
4935 | /** | |
d9c23400 | 4936 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4937 | * @page: The page within the block of interest |
4938 | * @start_bitidx: The first bit of interest to retrieve | |
4939 | * @end_bitidx: The last bit of interest | |
4940 | * returns pageblock_bits flags | |
4941 | */ | |
4942 | unsigned long get_pageblock_flags_group(struct page *page, | |
4943 | int start_bitidx, int end_bitidx) | |
4944 | { | |
4945 | struct zone *zone; | |
4946 | unsigned long *bitmap; | |
4947 | unsigned long pfn, bitidx; | |
4948 | unsigned long flags = 0; | |
4949 | unsigned long value = 1; | |
4950 | ||
4951 | zone = page_zone(page); | |
4952 | pfn = page_to_pfn(page); | |
4953 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4954 | bitidx = pfn_to_bitidx(zone, pfn); | |
4955 | ||
4956 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4957 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4958 | flags |= value; | |
6220ec78 | 4959 | |
835c134e MG |
4960 | return flags; |
4961 | } | |
4962 | ||
4963 | /** | |
d9c23400 | 4964 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4965 | * @page: The page within the block of interest |
4966 | * @start_bitidx: The first bit of interest | |
4967 | * @end_bitidx: The last bit of interest | |
4968 | * @flags: The flags to set | |
4969 | */ | |
4970 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4971 | int start_bitidx, int end_bitidx) | |
4972 | { | |
4973 | struct zone *zone; | |
4974 | unsigned long *bitmap; | |
4975 | unsigned long pfn, bitidx; | |
4976 | unsigned long value = 1; | |
4977 | ||
4978 | zone = page_zone(page); | |
4979 | pfn = page_to_pfn(page); | |
4980 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4981 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
4982 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
4983 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
4984 | |
4985 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4986 | if (flags & value) | |
4987 | __set_bit(bitidx + start_bitidx, bitmap); | |
4988 | else | |
4989 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4990 | } | |
a5d76b54 KH |
4991 | |
4992 | /* | |
4993 | * This is designed as sub function...plz see page_isolation.c also. | |
4994 | * set/clear page block's type to be ISOLATE. | |
4995 | * page allocater never alloc memory from ISOLATE block. | |
4996 | */ | |
4997 | ||
4998 | int set_migratetype_isolate(struct page *page) | |
4999 | { | |
5000 | struct zone *zone; | |
5001 | unsigned long flags; | |
5002 | int ret = -EBUSY; | |
8e7e40d9 | 5003 | int zone_idx; |
a5d76b54 KH |
5004 | |
5005 | zone = page_zone(page); | |
8e7e40d9 | 5006 | zone_idx = zone_idx(zone); |
a5d76b54 KH |
5007 | spin_lock_irqsave(&zone->lock, flags); |
5008 | /* | |
5009 | * In future, more migrate types will be able to be isolation target. | |
5010 | */ | |
8e7e40d9 SL |
5011 | if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE && |
5012 | zone_idx != ZONE_MOVABLE) | |
a5d76b54 KH |
5013 | goto out; |
5014 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
5015 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
5016 | ret = 0; | |
5017 | out: | |
5018 | spin_unlock_irqrestore(&zone->lock, flags); | |
5019 | if (!ret) | |
9f8f2172 | 5020 | drain_all_pages(); |
a5d76b54 KH |
5021 | return ret; |
5022 | } | |
5023 | ||
5024 | void unset_migratetype_isolate(struct page *page) | |
5025 | { | |
5026 | struct zone *zone; | |
5027 | unsigned long flags; | |
5028 | zone = page_zone(page); | |
5029 | spin_lock_irqsave(&zone->lock, flags); | |
5030 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
5031 | goto out; | |
5032 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5033 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
5034 | out: | |
5035 | spin_unlock_irqrestore(&zone->lock, flags); | |
5036 | } | |
0c0e6195 KH |
5037 | |
5038 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5039 | /* | |
5040 | * All pages in the range must be isolated before calling this. | |
5041 | */ | |
5042 | void | |
5043 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5044 | { | |
5045 | struct page *page; | |
5046 | struct zone *zone; | |
5047 | int order, i; | |
5048 | unsigned long pfn; | |
5049 | unsigned long flags; | |
5050 | /* find the first valid pfn */ | |
5051 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5052 | if (pfn_valid(pfn)) | |
5053 | break; | |
5054 | if (pfn == end_pfn) | |
5055 | return; | |
5056 | zone = page_zone(pfn_to_page(pfn)); | |
5057 | spin_lock_irqsave(&zone->lock, flags); | |
5058 | pfn = start_pfn; | |
5059 | while (pfn < end_pfn) { | |
5060 | if (!pfn_valid(pfn)) { | |
5061 | pfn++; | |
5062 | continue; | |
5063 | } | |
5064 | page = pfn_to_page(pfn); | |
5065 | BUG_ON(page_count(page)); | |
5066 | BUG_ON(!PageBuddy(page)); | |
5067 | order = page_order(page); | |
5068 | #ifdef CONFIG_DEBUG_VM | |
5069 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5070 | pfn, 1 << order, end_pfn); | |
5071 | #endif | |
5072 | list_del(&page->lru); | |
5073 | rmv_page_order(page); | |
5074 | zone->free_area[order].nr_free--; | |
5075 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
5076 | - (1UL << order)); | |
5077 | for (i = 0; i < (1 << order); i++) | |
5078 | SetPageReserved((page+i)); | |
5079 | pfn += (1 << order); | |
5080 | } | |
5081 | spin_unlock_irqrestore(&zone->lock, flags); | |
5082 | } | |
5083 | #endif |