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 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.h> |
1da177e4 LT |
28 | #include <linux/module.h> |
29 | #include <linux/suspend.h> | |
30 | #include <linux/pagevec.h> | |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/slab.h> | |
5a3135c2 | 33 | #include <linux/oom.h> |
1da177e4 LT |
34 | #include <linux/notifier.h> |
35 | #include <linux/topology.h> | |
36 | #include <linux/sysctl.h> | |
37 | #include <linux/cpu.h> | |
38 | #include <linux/cpuset.h> | |
bdc8cb98 | 39 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
40 | #include <linux/nodemask.h> |
41 | #include <linux/vmalloc.h> | |
4be38e35 | 42 | #include <linux/mempolicy.h> |
6811378e | 43 | #include <linux/stop_machine.h> |
c713216d MG |
44 | #include <linux/sort.h> |
45 | #include <linux/pfn.h> | |
3fcfab16 | 46 | #include <linux/backing-dev.h> |
933e312e | 47 | #include <linux/fault-inject.h> |
a5d76b54 | 48 | #include <linux/page-isolation.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
3ac7fe5a | 50 | #include <linux/debugobjects.h> |
dbb1f81c | 51 | #include <linux/kmemleak.h> |
925cc71e | 52 | #include <linux/memory.h> |
56de7263 | 53 | #include <linux/compaction.h> |
0d3d062a | 54 | #include <trace/events/kmem.h> |
718a3821 | 55 | #include <linux/ftrace_event.h> |
1da177e4 LT |
56 | |
57 | #include <asm/tlbflush.h> | |
ac924c60 | 58 | #include <asm/div64.h> |
1da177e4 LT |
59 | #include "internal.h" |
60 | ||
72812019 LS |
61 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
62 | DEFINE_PER_CPU(int, numa_node); | |
63 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
64 | #endif | |
65 | ||
7aac7898 LS |
66 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
67 | /* | |
68 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
69 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
70 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
71 | * defined in <linux/topology.h>. | |
72 | */ | |
73 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
74 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
75 | #endif | |
76 | ||
1da177e4 | 77 | /* |
13808910 | 78 | * Array of node states. |
1da177e4 | 79 | */ |
13808910 CL |
80 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
81 | [N_POSSIBLE] = NODE_MASK_ALL, | |
82 | [N_ONLINE] = { { [0] = 1UL } }, | |
83 | #ifndef CONFIG_NUMA | |
84 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
85 | #ifdef CONFIG_HIGHMEM | |
86 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
87 | #endif | |
88 | [N_CPU] = { { [0] = 1UL } }, | |
89 | #endif /* NUMA */ | |
90 | }; | |
91 | EXPORT_SYMBOL(node_states); | |
92 | ||
6c231b7b | 93 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 94 | unsigned long totalreserve_pages __read_mostly; |
8ad4b1fb | 95 | int percpu_pagelist_fraction; |
dcce284a | 96 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 97 | |
452aa699 RW |
98 | #ifdef CONFIG_PM_SLEEP |
99 | /* | |
100 | * The following functions are used by the suspend/hibernate code to temporarily | |
101 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
102 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
103 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
104 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
105 | * guaranteed not to run in parallel with that modification). | |
106 | */ | |
c9e664f1 RW |
107 | |
108 | static gfp_t saved_gfp_mask; | |
109 | ||
110 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
111 | { |
112 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
113 | if (saved_gfp_mask) { |
114 | gfp_allowed_mask = saved_gfp_mask; | |
115 | saved_gfp_mask = 0; | |
116 | } | |
452aa699 RW |
117 | } |
118 | ||
c9e664f1 | 119 | void pm_restrict_gfp_mask(void) |
452aa699 | 120 | { |
452aa699 | 121 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
122 | WARN_ON(saved_gfp_mask); |
123 | saved_gfp_mask = gfp_allowed_mask; | |
124 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 RW |
125 | } |
126 | #endif /* CONFIG_PM_SLEEP */ | |
127 | ||
d9c23400 MG |
128 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
129 | int pageblock_order __read_mostly; | |
130 | #endif | |
131 | ||
d98c7a09 | 132 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 133 | |
1da177e4 LT |
134 | /* |
135 | * results with 256, 32 in the lowmem_reserve sysctl: | |
136 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
137 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
138 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
139 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
140 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
141 | * |
142 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
143 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 144 | */ |
2f1b6248 | 145 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 146 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 147 | 256, |
4b51d669 | 148 | #endif |
fb0e7942 | 149 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 150 | 256, |
fb0e7942 | 151 | #endif |
e53ef38d | 152 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 153 | 32, |
e53ef38d | 154 | #endif |
2a1e274a | 155 | 32, |
2f1b6248 | 156 | }; |
1da177e4 LT |
157 | |
158 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 159 | |
15ad7cdc | 160 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 161 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 162 | "DMA", |
4b51d669 | 163 | #endif |
fb0e7942 | 164 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 165 | "DMA32", |
fb0e7942 | 166 | #endif |
2f1b6248 | 167 | "Normal", |
e53ef38d | 168 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 169 | "HighMem", |
e53ef38d | 170 | #endif |
2a1e274a | 171 | "Movable", |
2f1b6248 CL |
172 | }; |
173 | ||
1da177e4 LT |
174 | int min_free_kbytes = 1024; |
175 | ||
2c85f51d JB |
176 | static unsigned long __meminitdata nr_kernel_pages; |
177 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 178 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 179 | |
c713216d MG |
180 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
181 | /* | |
183ff22b | 182 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
183 | * ranges of memory (RAM) that may be registered with add_active_range(). |
184 | * Ranges passed to add_active_range() will be merged if possible | |
185 | * so the number of times add_active_range() can be called is | |
186 | * related to the number of nodes and the number of holes | |
187 | */ | |
188 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
189 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
190 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
191 | #else | |
192 | #if MAX_NUMNODES >= 32 | |
193 | /* If there can be many nodes, allow up to 50 holes per node */ | |
194 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
195 | #else | |
196 | /* By default, allow up to 256 distinct regions */ | |
197 | #define MAX_ACTIVE_REGIONS 256 | |
198 | #endif | |
199 | #endif | |
200 | ||
98011f56 JB |
201 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
202 | static int __meminitdata nr_nodemap_entries; | |
203 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
204 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
b69a7288 | 205 | static unsigned long __initdata required_kernelcore; |
484f51f8 | 206 | static unsigned long __initdata required_movablecore; |
b69a7288 | 207 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
208 | |
209 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
210 | int movable_zone; | |
211 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
212 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
213 | ||
418508c1 MS |
214 | #if MAX_NUMNODES > 1 |
215 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 216 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 217 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 218 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
219 | #endif |
220 | ||
9ef9acb0 MG |
221 | int page_group_by_mobility_disabled __read_mostly; |
222 | ||
b2a0ac88 MG |
223 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
224 | { | |
49255c61 MG |
225 | |
226 | if (unlikely(page_group_by_mobility_disabled)) | |
227 | migratetype = MIGRATE_UNMOVABLE; | |
228 | ||
b2a0ac88 MG |
229 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
230 | PB_migrate, PB_migrate_end); | |
231 | } | |
232 | ||
7f33d49a RW |
233 | bool oom_killer_disabled __read_mostly; |
234 | ||
13e7444b | 235 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 236 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 237 | { |
bdc8cb98 DH |
238 | int ret = 0; |
239 | unsigned seq; | |
240 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 241 | |
bdc8cb98 DH |
242 | do { |
243 | seq = zone_span_seqbegin(zone); | |
244 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
245 | ret = 1; | |
246 | else if (pfn < zone->zone_start_pfn) | |
247 | ret = 1; | |
248 | } while (zone_span_seqretry(zone, seq)); | |
249 | ||
250 | return ret; | |
c6a57e19 DH |
251 | } |
252 | ||
253 | static int page_is_consistent(struct zone *zone, struct page *page) | |
254 | { | |
14e07298 | 255 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 256 | return 0; |
1da177e4 | 257 | if (zone != page_zone(page)) |
c6a57e19 DH |
258 | return 0; |
259 | ||
260 | return 1; | |
261 | } | |
262 | /* | |
263 | * Temporary debugging check for pages not lying within a given zone. | |
264 | */ | |
265 | static int bad_range(struct zone *zone, struct page *page) | |
266 | { | |
267 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 268 | return 1; |
c6a57e19 DH |
269 | if (!page_is_consistent(zone, page)) |
270 | return 1; | |
271 | ||
1da177e4 LT |
272 | return 0; |
273 | } | |
13e7444b NP |
274 | #else |
275 | static inline int bad_range(struct zone *zone, struct page *page) | |
276 | { | |
277 | return 0; | |
278 | } | |
279 | #endif | |
280 | ||
224abf92 | 281 | static void bad_page(struct page *page) |
1da177e4 | 282 | { |
d936cf9b HD |
283 | static unsigned long resume; |
284 | static unsigned long nr_shown; | |
285 | static unsigned long nr_unshown; | |
286 | ||
2a7684a2 WF |
287 | /* Don't complain about poisoned pages */ |
288 | if (PageHWPoison(page)) { | |
289 | __ClearPageBuddy(page); | |
290 | return; | |
291 | } | |
292 | ||
d936cf9b HD |
293 | /* |
294 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
295 | * or allow a steady drip of one report per second. | |
296 | */ | |
297 | if (nr_shown == 60) { | |
298 | if (time_before(jiffies, resume)) { | |
299 | nr_unshown++; | |
300 | goto out; | |
301 | } | |
302 | if (nr_unshown) { | |
1e9e6365 HD |
303 | printk(KERN_ALERT |
304 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
305 | nr_unshown); |
306 | nr_unshown = 0; | |
307 | } | |
308 | nr_shown = 0; | |
309 | } | |
310 | if (nr_shown++ == 0) | |
311 | resume = jiffies + 60 * HZ; | |
312 | ||
1e9e6365 | 313 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 314 | current->comm, page_to_pfn(page)); |
718a3821 | 315 | dump_page(page); |
3dc14741 | 316 | |
1da177e4 | 317 | dump_stack(); |
d936cf9b | 318 | out: |
8cc3b392 HD |
319 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
320 | __ClearPageBuddy(page); | |
9f158333 | 321 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
322 | } |
323 | ||
1da177e4 LT |
324 | /* |
325 | * Higher-order pages are called "compound pages". They are structured thusly: | |
326 | * | |
327 | * The first PAGE_SIZE page is called the "head page". | |
328 | * | |
329 | * The remaining PAGE_SIZE pages are called "tail pages". | |
330 | * | |
331 | * All pages have PG_compound set. All pages have their ->private pointing at | |
332 | * the head page (even the head page has this). | |
333 | * | |
41d78ba5 HD |
334 | * The first tail page's ->lru.next holds the address of the compound page's |
335 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
336 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 337 | */ |
d98c7a09 HD |
338 | |
339 | static void free_compound_page(struct page *page) | |
340 | { | |
d85f3385 | 341 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
342 | } |
343 | ||
01ad1c08 | 344 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
345 | { |
346 | int i; | |
347 | int nr_pages = 1 << order; | |
348 | ||
349 | set_compound_page_dtor(page, free_compound_page); | |
350 | set_compound_order(page, order); | |
351 | __SetPageHead(page); | |
352 | for (i = 1; i < nr_pages; i++) { | |
353 | struct page *p = page + i; | |
354 | ||
355 | __SetPageTail(p); | |
356 | p->first_page = page; | |
357 | } | |
358 | } | |
359 | ||
59ff4216 | 360 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 361 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
362 | { |
363 | int i; | |
364 | int nr_pages = 1 << order; | |
8cc3b392 | 365 | int bad = 0; |
1da177e4 | 366 | |
8cc3b392 HD |
367 | if (unlikely(compound_order(page) != order) || |
368 | unlikely(!PageHead(page))) { | |
224abf92 | 369 | bad_page(page); |
8cc3b392 HD |
370 | bad++; |
371 | } | |
1da177e4 | 372 | |
6d777953 | 373 | __ClearPageHead(page); |
8cc3b392 | 374 | |
18229df5 AW |
375 | for (i = 1; i < nr_pages; i++) { |
376 | struct page *p = page + i; | |
1da177e4 | 377 | |
e713a21d | 378 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 379 | bad_page(page); |
8cc3b392 HD |
380 | bad++; |
381 | } | |
d85f3385 | 382 | __ClearPageTail(p); |
1da177e4 | 383 | } |
8cc3b392 HD |
384 | |
385 | return bad; | |
1da177e4 | 386 | } |
1da177e4 | 387 | |
17cf4406 NP |
388 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
389 | { | |
390 | int i; | |
391 | ||
6626c5d5 AM |
392 | /* |
393 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
394 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
395 | */ | |
725d704e | 396 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
397 | for (i = 0; i < (1 << order); i++) |
398 | clear_highpage(page + i); | |
399 | } | |
400 | ||
6aa3001b AM |
401 | static inline void set_page_order(struct page *page, int order) |
402 | { | |
4c21e2f2 | 403 | set_page_private(page, order); |
676165a8 | 404 | __SetPageBuddy(page); |
1da177e4 LT |
405 | } |
406 | ||
407 | static inline void rmv_page_order(struct page *page) | |
408 | { | |
676165a8 | 409 | __ClearPageBuddy(page); |
4c21e2f2 | 410 | set_page_private(page, 0); |
1da177e4 LT |
411 | } |
412 | ||
413 | /* | |
414 | * Locate the struct page for both the matching buddy in our | |
415 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
416 | * | |
417 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
418 | * the following equation: | |
419 | * B2 = B1 ^ (1 << O) | |
420 | * For example, if the starting buddy (buddy2) is #8 its order | |
421 | * 1 buddy is #10: | |
422 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
423 | * | |
424 | * 2) Any buddy B will have an order O+1 parent P which | |
425 | * satisfies the following equation: | |
426 | * P = B & ~(1 << O) | |
427 | * | |
d6e05edc | 428 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
429 | */ |
430 | static inline struct page * | |
431 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
432 | { | |
433 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
434 | ||
435 | return page + (buddy_idx - page_idx); | |
436 | } | |
437 | ||
438 | static inline unsigned long | |
439 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
440 | { | |
441 | return (page_idx & ~(1 << order)); | |
442 | } | |
443 | ||
444 | /* | |
445 | * This function checks whether a page is free && is the buddy | |
446 | * we can do coalesce a page and its buddy if | |
13e7444b | 447 | * (a) the buddy is not in a hole && |
676165a8 | 448 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
449 | * (c) a page and its buddy have the same order && |
450 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
451 | * |
452 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
453 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 454 | * |
676165a8 | 455 | * For recording page's order, we use page_private(page). |
1da177e4 | 456 | */ |
cb2b95e1 AW |
457 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
458 | int order) | |
1da177e4 | 459 | { |
14e07298 | 460 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 461 | return 0; |
13e7444b | 462 | |
cb2b95e1 AW |
463 | if (page_zone_id(page) != page_zone_id(buddy)) |
464 | return 0; | |
465 | ||
466 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
a3af9c38 | 467 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 468 | return 1; |
676165a8 | 469 | } |
6aa3001b | 470 | return 0; |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
474 | * Freeing function for a buddy system allocator. | |
475 | * | |
476 | * The concept of a buddy system is to maintain direct-mapped table | |
477 | * (containing bit values) for memory blocks of various "orders". | |
478 | * The bottom level table contains the map for the smallest allocatable | |
479 | * units of memory (here, pages), and each level above it describes | |
480 | * pairs of units from the levels below, hence, "buddies". | |
481 | * At a high level, all that happens here is marking the table entry | |
482 | * at the bottom level available, and propagating the changes upward | |
483 | * as necessary, plus some accounting needed to play nicely with other | |
484 | * parts of the VM system. | |
485 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 486 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 487 | * order is recorded in page_private(page) field. |
1da177e4 LT |
488 | * So when we are allocating or freeing one, we can derive the state of the |
489 | * other. That is, if we allocate a small block, and both were | |
490 | * free, the remainder of the region must be split into blocks. | |
491 | * If a block is freed, and its buddy is also free, then this | |
492 | * triggers coalescing into a block of larger size. | |
493 | * | |
494 | * -- wli | |
495 | */ | |
496 | ||
48db57f8 | 497 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
498 | struct zone *zone, unsigned int order, |
499 | int migratetype) | |
1da177e4 LT |
500 | { |
501 | unsigned long page_idx; | |
6dda9d55 CZ |
502 | unsigned long combined_idx; |
503 | struct page *buddy; | |
1da177e4 | 504 | |
224abf92 | 505 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
506 | if (unlikely(destroy_compound_page(page, order))) |
507 | return; | |
1da177e4 | 508 | |
ed0ae21d MG |
509 | VM_BUG_ON(migratetype == -1); |
510 | ||
1da177e4 LT |
511 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
512 | ||
f2260e6b | 513 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 514 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 515 | |
1da177e4 | 516 | while (order < MAX_ORDER-1) { |
1da177e4 | 517 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 518 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 519 | break; |
13e7444b | 520 | |
3c82d0ce | 521 | /* Our buddy is free, merge with it and move up one order. */ |
1da177e4 | 522 | list_del(&buddy->lru); |
b2a0ac88 | 523 | zone->free_area[order].nr_free--; |
1da177e4 | 524 | rmv_page_order(buddy); |
13e7444b | 525 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
526 | page = page + (combined_idx - page_idx); |
527 | page_idx = combined_idx; | |
528 | order++; | |
529 | } | |
530 | set_page_order(page, order); | |
6dda9d55 CZ |
531 | |
532 | /* | |
533 | * If this is not the largest possible page, check if the buddy | |
534 | * of the next-highest order is free. If it is, it's possible | |
535 | * that pages are being freed that will coalesce soon. In case, | |
536 | * that is happening, add the free page to the tail of the list | |
537 | * so it's less likely to be used soon and more likely to be merged | |
538 | * as a higher order page | |
539 | */ | |
b7f50cfa | 540 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 CZ |
541 | struct page *higher_page, *higher_buddy; |
542 | combined_idx = __find_combined_index(page_idx, order); | |
543 | higher_page = page + combined_idx - page_idx; | |
544 | higher_buddy = __page_find_buddy(higher_page, combined_idx, order + 1); | |
545 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
546 | list_add_tail(&page->lru, | |
547 | &zone->free_area[order].free_list[migratetype]); | |
548 | goto out; | |
549 | } | |
550 | } | |
551 | ||
552 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
553 | out: | |
1da177e4 LT |
554 | zone->free_area[order].nr_free++; |
555 | } | |
556 | ||
092cead6 KM |
557 | /* |
558 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
559 | * Page should not be on lru, so no need to fix that up. | |
560 | * free_pages_check() will verify... | |
561 | */ | |
562 | static inline void free_page_mlock(struct page *page) | |
563 | { | |
092cead6 KM |
564 | __dec_zone_page_state(page, NR_MLOCK); |
565 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
566 | } | |
092cead6 | 567 | |
224abf92 | 568 | static inline int free_pages_check(struct page *page) |
1da177e4 | 569 | { |
92be2e33 NP |
570 | if (unlikely(page_mapcount(page) | |
571 | (page->mapping != NULL) | | |
a3af9c38 | 572 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 573 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
224abf92 | 574 | bad_page(page); |
79f4b7bf | 575 | return 1; |
8cc3b392 | 576 | } |
79f4b7bf HD |
577 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
578 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
579 | return 0; | |
1da177e4 LT |
580 | } |
581 | ||
582 | /* | |
5f8dcc21 | 583 | * Frees a number of pages from the PCP lists |
1da177e4 | 584 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 585 | * count is the number of pages to free. |
1da177e4 LT |
586 | * |
587 | * If the zone was previously in an "all pages pinned" state then look to | |
588 | * see if this freeing clears that state. | |
589 | * | |
590 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
591 | * pinned" detection logic. | |
592 | */ | |
5f8dcc21 MG |
593 | static void free_pcppages_bulk(struct zone *zone, int count, |
594 | struct per_cpu_pages *pcp) | |
1da177e4 | 595 | { |
5f8dcc21 | 596 | int migratetype = 0; |
a6f9edd6 | 597 | int batch_free = 0; |
72853e29 | 598 | int to_free = count; |
5f8dcc21 | 599 | |
c54ad30c | 600 | spin_lock(&zone->lock); |
93e4a89a | 601 | zone->all_unreclaimable = 0; |
1da177e4 | 602 | zone->pages_scanned = 0; |
f2260e6b | 603 | |
72853e29 | 604 | while (to_free) { |
48db57f8 | 605 | struct page *page; |
5f8dcc21 MG |
606 | struct list_head *list; |
607 | ||
608 | /* | |
a6f9edd6 MG |
609 | * Remove pages from lists in a round-robin fashion. A |
610 | * batch_free count is maintained that is incremented when an | |
611 | * empty list is encountered. This is so more pages are freed | |
612 | * off fuller lists instead of spinning excessively around empty | |
613 | * lists | |
5f8dcc21 MG |
614 | */ |
615 | do { | |
a6f9edd6 | 616 | batch_free++; |
5f8dcc21 MG |
617 | if (++migratetype == MIGRATE_PCPTYPES) |
618 | migratetype = 0; | |
619 | list = &pcp->lists[migratetype]; | |
620 | } while (list_empty(list)); | |
48db57f8 | 621 | |
a6f9edd6 MG |
622 | do { |
623 | page = list_entry(list->prev, struct page, lru); | |
624 | /* must delete as __free_one_page list manipulates */ | |
625 | list_del(&page->lru); | |
a7016235 HD |
626 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
627 | __free_one_page(page, zone, 0, page_private(page)); | |
628 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); | |
72853e29 | 629 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 630 | } |
72853e29 | 631 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
c54ad30c | 632 | spin_unlock(&zone->lock); |
1da177e4 LT |
633 | } |
634 | ||
ed0ae21d MG |
635 | static void free_one_page(struct zone *zone, struct page *page, int order, |
636 | int migratetype) | |
1da177e4 | 637 | { |
006d22d9 | 638 | spin_lock(&zone->lock); |
93e4a89a | 639 | zone->all_unreclaimable = 0; |
006d22d9 | 640 | zone->pages_scanned = 0; |
f2260e6b | 641 | |
ed0ae21d | 642 | __free_one_page(page, zone, order, migratetype); |
72853e29 | 643 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); |
006d22d9 | 644 | spin_unlock(&zone->lock); |
48db57f8 NP |
645 | } |
646 | ||
ec95f53a | 647 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 648 | { |
1da177e4 | 649 | int i; |
8cc3b392 | 650 | int bad = 0; |
1da177e4 | 651 | |
f650316c | 652 | trace_mm_page_free_direct(page, order); |
b1eeab67 VN |
653 | kmemcheck_free_shadow(page, order); |
654 | ||
8dd60a3a AA |
655 | if (PageAnon(page)) |
656 | page->mapping = NULL; | |
657 | for (i = 0; i < (1 << order); i++) | |
658 | bad += free_pages_check(page + i); | |
8cc3b392 | 659 | if (bad) |
ec95f53a | 660 | return false; |
689bcebf | 661 | |
3ac7fe5a | 662 | if (!PageHighMem(page)) { |
9858db50 | 663 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
664 | debug_check_no_obj_freed(page_address(page), |
665 | PAGE_SIZE << order); | |
666 | } | |
dafb1367 | 667 | arch_free_page(page, order); |
48db57f8 | 668 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 669 | |
ec95f53a KM |
670 | return true; |
671 | } | |
672 | ||
673 | static void __free_pages_ok(struct page *page, unsigned int order) | |
674 | { | |
675 | unsigned long flags; | |
676 | int wasMlocked = __TestClearPageMlocked(page); | |
677 | ||
678 | if (!free_pages_prepare(page, order)) | |
679 | return; | |
680 | ||
c54ad30c | 681 | local_irq_save(flags); |
c277331d | 682 | if (unlikely(wasMlocked)) |
da456f14 | 683 | free_page_mlock(page); |
f8891e5e | 684 | __count_vm_events(PGFREE, 1 << order); |
ed0ae21d MG |
685 | free_one_page(page_zone(page), page, order, |
686 | get_pageblock_migratetype(page)); | |
c54ad30c | 687 | local_irq_restore(flags); |
1da177e4 LT |
688 | } |
689 | ||
a226f6c8 DH |
690 | /* |
691 | * permit the bootmem allocator to evade page validation on high-order frees | |
692 | */ | |
af370fb8 | 693 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
694 | { |
695 | if (order == 0) { | |
696 | __ClearPageReserved(page); | |
697 | set_page_count(page, 0); | |
7835e98b | 698 | set_page_refcounted(page); |
545b1ea9 | 699 | __free_page(page); |
a226f6c8 | 700 | } else { |
a226f6c8 DH |
701 | int loop; |
702 | ||
545b1ea9 | 703 | prefetchw(page); |
a226f6c8 DH |
704 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
705 | struct page *p = &page[loop]; | |
706 | ||
545b1ea9 NP |
707 | if (loop + 1 < BITS_PER_LONG) |
708 | prefetchw(p + 1); | |
a226f6c8 DH |
709 | __ClearPageReserved(p); |
710 | set_page_count(p, 0); | |
711 | } | |
712 | ||
7835e98b | 713 | set_page_refcounted(page); |
545b1ea9 | 714 | __free_pages(page, order); |
a226f6c8 DH |
715 | } |
716 | } | |
717 | ||
1da177e4 LT |
718 | |
719 | /* | |
720 | * The order of subdivision here is critical for the IO subsystem. | |
721 | * Please do not alter this order without good reasons and regression | |
722 | * testing. Specifically, as large blocks of memory are subdivided, | |
723 | * the order in which smaller blocks are delivered depends on the order | |
724 | * they're subdivided in this function. This is the primary factor | |
725 | * influencing the order in which pages are delivered to the IO | |
726 | * subsystem according to empirical testing, and this is also justified | |
727 | * by considering the behavior of a buddy system containing a single | |
728 | * large block of memory acted on by a series of small allocations. | |
729 | * This behavior is a critical factor in sglist merging's success. | |
730 | * | |
731 | * -- wli | |
732 | */ | |
085cc7d5 | 733 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
734 | int low, int high, struct free_area *area, |
735 | int migratetype) | |
1da177e4 LT |
736 | { |
737 | unsigned long size = 1 << high; | |
738 | ||
739 | while (high > low) { | |
740 | area--; | |
741 | high--; | |
742 | size >>= 1; | |
725d704e | 743 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 744 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
745 | area->nr_free++; |
746 | set_page_order(&page[size], high); | |
747 | } | |
1da177e4 LT |
748 | } |
749 | ||
1da177e4 LT |
750 | /* |
751 | * This page is about to be returned from the page allocator | |
752 | */ | |
2a7684a2 | 753 | static inline int check_new_page(struct page *page) |
1da177e4 | 754 | { |
92be2e33 NP |
755 | if (unlikely(page_mapcount(page) | |
756 | (page->mapping != NULL) | | |
a3af9c38 | 757 | (atomic_read(&page->_count) != 0) | |
8cc3b392 | 758 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
224abf92 | 759 | bad_page(page); |
689bcebf | 760 | return 1; |
8cc3b392 | 761 | } |
2a7684a2 WF |
762 | return 0; |
763 | } | |
764 | ||
765 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
766 | { | |
767 | int i; | |
768 | ||
769 | for (i = 0; i < (1 << order); i++) { | |
770 | struct page *p = page + i; | |
771 | if (unlikely(check_new_page(p))) | |
772 | return 1; | |
773 | } | |
689bcebf | 774 | |
4c21e2f2 | 775 | set_page_private(page, 0); |
7835e98b | 776 | set_page_refcounted(page); |
cc102509 NP |
777 | |
778 | arch_alloc_page(page, order); | |
1da177e4 | 779 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
780 | |
781 | if (gfp_flags & __GFP_ZERO) | |
782 | prep_zero_page(page, order, gfp_flags); | |
783 | ||
784 | if (order && (gfp_flags & __GFP_COMP)) | |
785 | prep_compound_page(page, order); | |
786 | ||
689bcebf | 787 | return 0; |
1da177e4 LT |
788 | } |
789 | ||
56fd56b8 MG |
790 | /* |
791 | * Go through the free lists for the given migratetype and remove | |
792 | * the smallest available page from the freelists | |
793 | */ | |
728ec980 MG |
794 | static inline |
795 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
796 | int migratetype) |
797 | { | |
798 | unsigned int current_order; | |
799 | struct free_area * area; | |
800 | struct page *page; | |
801 | ||
802 | /* Find a page of the appropriate size in the preferred list */ | |
803 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
804 | area = &(zone->free_area[current_order]); | |
805 | if (list_empty(&area->free_list[migratetype])) | |
806 | continue; | |
807 | ||
808 | page = list_entry(area->free_list[migratetype].next, | |
809 | struct page, lru); | |
810 | list_del(&page->lru); | |
811 | rmv_page_order(page); | |
812 | area->nr_free--; | |
56fd56b8 MG |
813 | expand(zone, page, order, current_order, area, migratetype); |
814 | return page; | |
815 | } | |
816 | ||
817 | return NULL; | |
818 | } | |
819 | ||
820 | ||
b2a0ac88 MG |
821 | /* |
822 | * This array describes the order lists are fallen back to when | |
823 | * the free lists for the desirable migrate type are depleted | |
824 | */ | |
825 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
826 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
827 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
828 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
829 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
830 | }; |
831 | ||
c361be55 MG |
832 | /* |
833 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 834 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
835 | * boundary. If alignment is required, use move_freepages_block() |
836 | */ | |
b69a7288 AB |
837 | static int move_freepages(struct zone *zone, |
838 | struct page *start_page, struct page *end_page, | |
839 | int migratetype) | |
c361be55 MG |
840 | { |
841 | struct page *page; | |
842 | unsigned long order; | |
d100313f | 843 | int pages_moved = 0; |
c361be55 MG |
844 | |
845 | #ifndef CONFIG_HOLES_IN_ZONE | |
846 | /* | |
847 | * page_zone is not safe to call in this context when | |
848 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
849 | * anyway as we check zone boundaries in move_freepages_block(). | |
850 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 851 | * grouping pages by mobility |
c361be55 MG |
852 | */ |
853 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
854 | #endif | |
855 | ||
856 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
857 | /* Make sure we are not inadvertently changing nodes */ |
858 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
859 | ||
c361be55 MG |
860 | if (!pfn_valid_within(page_to_pfn(page))) { |
861 | page++; | |
862 | continue; | |
863 | } | |
864 | ||
865 | if (!PageBuddy(page)) { | |
866 | page++; | |
867 | continue; | |
868 | } | |
869 | ||
870 | order = page_order(page); | |
871 | list_del(&page->lru); | |
872 | list_add(&page->lru, | |
873 | &zone->free_area[order].free_list[migratetype]); | |
874 | page += 1 << order; | |
d100313f | 875 | pages_moved += 1 << order; |
c361be55 MG |
876 | } |
877 | ||
d100313f | 878 | return pages_moved; |
c361be55 MG |
879 | } |
880 | ||
b69a7288 AB |
881 | static int move_freepages_block(struct zone *zone, struct page *page, |
882 | int migratetype) | |
c361be55 MG |
883 | { |
884 | unsigned long start_pfn, end_pfn; | |
885 | struct page *start_page, *end_page; | |
886 | ||
887 | start_pfn = page_to_pfn(page); | |
d9c23400 | 888 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 889 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
890 | end_page = start_page + pageblock_nr_pages - 1; |
891 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
892 | |
893 | /* Do not cross zone boundaries */ | |
894 | if (start_pfn < zone->zone_start_pfn) | |
895 | start_page = page; | |
896 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
897 | return 0; | |
898 | ||
899 | return move_freepages(zone, start_page, end_page, migratetype); | |
900 | } | |
901 | ||
2f66a68f MG |
902 | static void change_pageblock_range(struct page *pageblock_page, |
903 | int start_order, int migratetype) | |
904 | { | |
905 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
906 | ||
907 | while (nr_pageblocks--) { | |
908 | set_pageblock_migratetype(pageblock_page, migratetype); | |
909 | pageblock_page += pageblock_nr_pages; | |
910 | } | |
911 | } | |
912 | ||
b2a0ac88 | 913 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
914 | static inline struct page * |
915 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
916 | { |
917 | struct free_area * area; | |
918 | int current_order; | |
919 | struct page *page; | |
920 | int migratetype, i; | |
921 | ||
922 | /* Find the largest possible block of pages in the other list */ | |
923 | for (current_order = MAX_ORDER-1; current_order >= order; | |
924 | --current_order) { | |
925 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
926 | migratetype = fallbacks[start_migratetype][i]; | |
927 | ||
56fd56b8 MG |
928 | /* MIGRATE_RESERVE handled later if necessary */ |
929 | if (migratetype == MIGRATE_RESERVE) | |
930 | continue; | |
e010487d | 931 | |
b2a0ac88 MG |
932 | area = &(zone->free_area[current_order]); |
933 | if (list_empty(&area->free_list[migratetype])) | |
934 | continue; | |
935 | ||
936 | page = list_entry(area->free_list[migratetype].next, | |
937 | struct page, lru); | |
938 | area->nr_free--; | |
939 | ||
940 | /* | |
c361be55 | 941 | * If breaking a large block of pages, move all free |
46dafbca MG |
942 | * pages to the preferred allocation list. If falling |
943 | * back for a reclaimable kernel allocation, be more | |
944 | * agressive about taking ownership of free pages | |
b2a0ac88 | 945 | */ |
d9c23400 | 946 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
dd5d241e MG |
947 | start_migratetype == MIGRATE_RECLAIMABLE || |
948 | page_group_by_mobility_disabled) { | |
46dafbca MG |
949 | unsigned long pages; |
950 | pages = move_freepages_block(zone, page, | |
951 | start_migratetype); | |
952 | ||
953 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
954 | if (pages >= (1 << (pageblock_order-1)) || |
955 | page_group_by_mobility_disabled) | |
46dafbca MG |
956 | set_pageblock_migratetype(page, |
957 | start_migratetype); | |
958 | ||
b2a0ac88 | 959 | migratetype = start_migratetype; |
c361be55 | 960 | } |
b2a0ac88 MG |
961 | |
962 | /* Remove the page from the freelists */ | |
963 | list_del(&page->lru); | |
964 | rmv_page_order(page); | |
b2a0ac88 | 965 | |
2f66a68f MG |
966 | /* Take ownership for orders >= pageblock_order */ |
967 | if (current_order >= pageblock_order) | |
968 | change_pageblock_range(page, current_order, | |
b2a0ac88 MG |
969 | start_migratetype); |
970 | ||
971 | expand(zone, page, order, current_order, area, migratetype); | |
e0fff1bd MG |
972 | |
973 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
974 | start_migratetype, migratetype); | |
975 | ||
b2a0ac88 MG |
976 | return page; |
977 | } | |
978 | } | |
979 | ||
728ec980 | 980 | return NULL; |
b2a0ac88 MG |
981 | } |
982 | ||
56fd56b8 | 983 | /* |
1da177e4 LT |
984 | * Do the hard work of removing an element from the buddy allocator. |
985 | * Call me with the zone->lock already held. | |
986 | */ | |
b2a0ac88 MG |
987 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
988 | int migratetype) | |
1da177e4 | 989 | { |
1da177e4 LT |
990 | struct page *page; |
991 | ||
728ec980 | 992 | retry_reserve: |
56fd56b8 | 993 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 994 | |
728ec980 | 995 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 996 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 997 | |
728ec980 MG |
998 | /* |
999 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1000 | * is used because __rmqueue_smallest is an inline function | |
1001 | * and we want just one call site | |
1002 | */ | |
1003 | if (!page) { | |
1004 | migratetype = MIGRATE_RESERVE; | |
1005 | goto retry_reserve; | |
1006 | } | |
1007 | } | |
1008 | ||
0d3d062a | 1009 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1010 | return page; |
1da177e4 LT |
1011 | } |
1012 | ||
1013 | /* | |
1014 | * Obtain a specified number of elements from the buddy allocator, all under | |
1015 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1016 | * Returns the number of new pages which were placed at *list. | |
1017 | */ | |
1018 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 | 1019 | unsigned long count, struct list_head *list, |
e084b2d9 | 1020 | int migratetype, int cold) |
1da177e4 | 1021 | { |
1da177e4 | 1022 | int i; |
1da177e4 | 1023 | |
c54ad30c | 1024 | spin_lock(&zone->lock); |
1da177e4 | 1025 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1026 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1027 | if (unlikely(page == NULL)) |
1da177e4 | 1028 | break; |
81eabcbe MG |
1029 | |
1030 | /* | |
1031 | * Split buddy pages returned by expand() are received here | |
1032 | * in physical page order. The page is added to the callers and | |
1033 | * list and the list head then moves forward. From the callers | |
1034 | * perspective, the linked list is ordered by page number in | |
1035 | * some conditions. This is useful for IO devices that can | |
1036 | * merge IO requests if the physical pages are ordered | |
1037 | * properly. | |
1038 | */ | |
e084b2d9 MG |
1039 | if (likely(cold == 0)) |
1040 | list_add(&page->lru, list); | |
1041 | else | |
1042 | list_add_tail(&page->lru, list); | |
535131e6 | 1043 | set_page_private(page, migratetype); |
81eabcbe | 1044 | list = &page->lru; |
1da177e4 | 1045 | } |
f2260e6b | 1046 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1047 | spin_unlock(&zone->lock); |
085cc7d5 | 1048 | return i; |
1da177e4 LT |
1049 | } |
1050 | ||
4ae7c039 | 1051 | #ifdef CONFIG_NUMA |
8fce4d8e | 1052 | /* |
4037d452 CL |
1053 | * Called from the vmstat counter updater to drain pagesets of this |
1054 | * currently executing processor on remote nodes after they have | |
1055 | * expired. | |
1056 | * | |
879336c3 CL |
1057 | * Note that this function must be called with the thread pinned to |
1058 | * a single processor. | |
8fce4d8e | 1059 | */ |
4037d452 | 1060 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1061 | { |
4ae7c039 | 1062 | unsigned long flags; |
4037d452 | 1063 | int to_drain; |
4ae7c039 | 1064 | |
4037d452 CL |
1065 | local_irq_save(flags); |
1066 | if (pcp->count >= pcp->batch) | |
1067 | to_drain = pcp->batch; | |
1068 | else | |
1069 | to_drain = pcp->count; | |
5f8dcc21 | 1070 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 CL |
1071 | pcp->count -= to_drain; |
1072 | local_irq_restore(flags); | |
4ae7c039 CL |
1073 | } |
1074 | #endif | |
1075 | ||
9f8f2172 CL |
1076 | /* |
1077 | * Drain pages of the indicated processor. | |
1078 | * | |
1079 | * The processor must either be the current processor and the | |
1080 | * thread pinned to the current processor or a processor that | |
1081 | * is not online. | |
1082 | */ | |
1083 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1084 | { |
c54ad30c | 1085 | unsigned long flags; |
1da177e4 | 1086 | struct zone *zone; |
1da177e4 | 1087 | |
ee99c71c | 1088 | for_each_populated_zone(zone) { |
1da177e4 | 1089 | struct per_cpu_pageset *pset; |
3dfa5721 | 1090 | struct per_cpu_pages *pcp; |
1da177e4 | 1091 | |
99dcc3e5 CL |
1092 | local_irq_save(flags); |
1093 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1094 | |
1095 | pcp = &pset->pcp; | |
5f8dcc21 | 1096 | free_pcppages_bulk(zone, pcp->count, pcp); |
3dfa5721 CL |
1097 | pcp->count = 0; |
1098 | local_irq_restore(flags); | |
1da177e4 LT |
1099 | } |
1100 | } | |
1da177e4 | 1101 | |
9f8f2172 CL |
1102 | /* |
1103 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1104 | */ | |
1105 | void drain_local_pages(void *arg) | |
1106 | { | |
1107 | drain_pages(smp_processor_id()); | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1112 | */ | |
1113 | void drain_all_pages(void) | |
1114 | { | |
15c8b6c1 | 1115 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
1116 | } |
1117 | ||
296699de | 1118 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1119 | |
1120 | void mark_free_pages(struct zone *zone) | |
1121 | { | |
f623f0db RW |
1122 | unsigned long pfn, max_zone_pfn; |
1123 | unsigned long flags; | |
b2a0ac88 | 1124 | int order, t; |
1da177e4 LT |
1125 | struct list_head *curr; |
1126 | ||
1127 | if (!zone->spanned_pages) | |
1128 | return; | |
1129 | ||
1130 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1131 | |
1132 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1133 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1134 | if (pfn_valid(pfn)) { | |
1135 | struct page *page = pfn_to_page(pfn); | |
1136 | ||
7be98234 RW |
1137 | if (!swsusp_page_is_forbidden(page)) |
1138 | swsusp_unset_page_free(page); | |
f623f0db | 1139 | } |
1da177e4 | 1140 | |
b2a0ac88 MG |
1141 | for_each_migratetype_order(order, t) { |
1142 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1143 | unsigned long i; |
1da177e4 | 1144 | |
f623f0db RW |
1145 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1146 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1147 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1148 | } |
b2a0ac88 | 1149 | } |
1da177e4 LT |
1150 | spin_unlock_irqrestore(&zone->lock, flags); |
1151 | } | |
e2c55dc8 | 1152 | #endif /* CONFIG_PM */ |
1da177e4 | 1153 | |
1da177e4 LT |
1154 | /* |
1155 | * Free a 0-order page | |
fc91668e | 1156 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1157 | */ |
fc91668e | 1158 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1159 | { |
1160 | struct zone *zone = page_zone(page); | |
1161 | struct per_cpu_pages *pcp; | |
1162 | unsigned long flags; | |
5f8dcc21 | 1163 | int migratetype; |
451ea25d | 1164 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 1165 | |
ec95f53a | 1166 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1167 | return; |
1168 | ||
5f8dcc21 MG |
1169 | migratetype = get_pageblock_migratetype(page); |
1170 | set_page_private(page, migratetype); | |
1da177e4 | 1171 | local_irq_save(flags); |
c277331d | 1172 | if (unlikely(wasMlocked)) |
da456f14 | 1173 | free_page_mlock(page); |
f8891e5e | 1174 | __count_vm_event(PGFREE); |
da456f14 | 1175 | |
5f8dcc21 MG |
1176 | /* |
1177 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1178 | * Free ISOLATE pages back to the allocator because they are being | |
1179 | * offlined but treat RESERVE as movable pages so we can get those | |
1180 | * areas back if necessary. Otherwise, we may have to free | |
1181 | * excessively into the page allocator | |
1182 | */ | |
1183 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1184 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1185 | free_one_page(zone, page, 0, migratetype); | |
1186 | goto out; | |
1187 | } | |
1188 | migratetype = MIGRATE_MOVABLE; | |
1189 | } | |
1190 | ||
99dcc3e5 | 1191 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1192 | if (cold) |
5f8dcc21 | 1193 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1194 | else |
5f8dcc21 | 1195 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1196 | pcp->count++; |
48db57f8 | 1197 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1198 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1199 | pcp->count -= pcp->batch; |
1200 | } | |
5f8dcc21 MG |
1201 | |
1202 | out: | |
1da177e4 | 1203 | local_irq_restore(flags); |
1da177e4 LT |
1204 | } |
1205 | ||
8dfcc9ba NP |
1206 | /* |
1207 | * split_page takes a non-compound higher-order page, and splits it into | |
1208 | * n (1<<order) sub-pages: page[0..n] | |
1209 | * Each sub-page must be freed individually. | |
1210 | * | |
1211 | * Note: this is probably too low level an operation for use in drivers. | |
1212 | * Please consult with lkml before using this in your driver. | |
1213 | */ | |
1214 | void split_page(struct page *page, unsigned int order) | |
1215 | { | |
1216 | int i; | |
1217 | ||
725d704e NP |
1218 | VM_BUG_ON(PageCompound(page)); |
1219 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1220 | |
1221 | #ifdef CONFIG_KMEMCHECK | |
1222 | /* | |
1223 | * Split shadow pages too, because free(page[0]) would | |
1224 | * otherwise free the whole shadow. | |
1225 | */ | |
1226 | if (kmemcheck_page_is_tracked(page)) | |
1227 | split_page(virt_to_page(page[0].shadow), order); | |
1228 | #endif | |
1229 | ||
7835e98b NP |
1230 | for (i = 1; i < (1 << order); i++) |
1231 | set_page_refcounted(page + i); | |
8dfcc9ba | 1232 | } |
8dfcc9ba | 1233 | |
748446bb MG |
1234 | /* |
1235 | * Similar to split_page except the page is already free. As this is only | |
1236 | * being used for migration, the migratetype of the block also changes. | |
1237 | * As this is called with interrupts disabled, the caller is responsible | |
1238 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1239 | * are enabled. | |
1240 | * | |
1241 | * Note: this is probably too low level an operation for use in drivers. | |
1242 | * Please consult with lkml before using this in your driver. | |
1243 | */ | |
1244 | int split_free_page(struct page *page) | |
1245 | { | |
1246 | unsigned int order; | |
1247 | unsigned long watermark; | |
1248 | struct zone *zone; | |
1249 | ||
1250 | BUG_ON(!PageBuddy(page)); | |
1251 | ||
1252 | zone = page_zone(page); | |
1253 | order = page_order(page); | |
1254 | ||
1255 | /* Obey watermarks as if the page was being allocated */ | |
1256 | watermark = low_wmark_pages(zone) + (1 << order); | |
1257 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1258 | return 0; | |
1259 | ||
1260 | /* Remove page from free list */ | |
1261 | list_del(&page->lru); | |
1262 | zone->free_area[order].nr_free--; | |
1263 | rmv_page_order(page); | |
1264 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order)); | |
1265 | ||
1266 | /* Split into individual pages */ | |
1267 | set_page_refcounted(page); | |
1268 | split_page(page, order); | |
1269 | ||
1270 | if (order >= pageblock_order - 1) { | |
1271 | struct page *endpage = page + (1 << order) - 1; | |
1272 | for (; page < endpage; page += pageblock_nr_pages) | |
1273 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
1274 | } | |
1275 | ||
1276 | return 1 << order; | |
1277 | } | |
1278 | ||
1da177e4 LT |
1279 | /* |
1280 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1281 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1282 | * or two. | |
1283 | */ | |
0a15c3e9 MG |
1284 | static inline |
1285 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1286 | struct zone *zone, int order, gfp_t gfp_flags, |
1287 | int migratetype) | |
1da177e4 LT |
1288 | { |
1289 | unsigned long flags; | |
689bcebf | 1290 | struct page *page; |
1da177e4 LT |
1291 | int cold = !!(gfp_flags & __GFP_COLD); |
1292 | ||
689bcebf | 1293 | again: |
48db57f8 | 1294 | if (likely(order == 0)) { |
1da177e4 | 1295 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1296 | struct list_head *list; |
1da177e4 | 1297 | |
1da177e4 | 1298 | local_irq_save(flags); |
99dcc3e5 CL |
1299 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1300 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1301 | if (list_empty(list)) { |
535131e6 | 1302 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1303 | pcp->batch, list, |
e084b2d9 | 1304 | migratetype, cold); |
5f8dcc21 | 1305 | if (unlikely(list_empty(list))) |
6fb332fa | 1306 | goto failed; |
535131e6 | 1307 | } |
b92a6edd | 1308 | |
5f8dcc21 MG |
1309 | if (cold) |
1310 | page = list_entry(list->prev, struct page, lru); | |
1311 | else | |
1312 | page = list_entry(list->next, struct page, lru); | |
1313 | ||
b92a6edd MG |
1314 | list_del(&page->lru); |
1315 | pcp->count--; | |
7fb1d9fc | 1316 | } else { |
dab48dab AM |
1317 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1318 | /* | |
1319 | * __GFP_NOFAIL is not to be used in new code. | |
1320 | * | |
1321 | * All __GFP_NOFAIL callers should be fixed so that they | |
1322 | * properly detect and handle allocation failures. | |
1323 | * | |
1324 | * We most definitely don't want callers attempting to | |
4923abf9 | 1325 | * allocate greater than order-1 page units with |
dab48dab AM |
1326 | * __GFP_NOFAIL. |
1327 | */ | |
4923abf9 | 1328 | WARN_ON_ONCE(order > 1); |
dab48dab | 1329 | } |
1da177e4 | 1330 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1331 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1332 | spin_unlock(&zone->lock); |
1333 | if (!page) | |
1334 | goto failed; | |
6ccf80eb | 1335 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
1da177e4 LT |
1336 | } |
1337 | ||
f8891e5e | 1338 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
18ea7e71 | 1339 | zone_statistics(preferred_zone, zone); |
a74609fa | 1340 | local_irq_restore(flags); |
1da177e4 | 1341 | |
725d704e | 1342 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1343 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1344 | goto again; |
1da177e4 | 1345 | return page; |
a74609fa NP |
1346 | |
1347 | failed: | |
1348 | local_irq_restore(flags); | |
a74609fa | 1349 | return NULL; |
1da177e4 LT |
1350 | } |
1351 | ||
41858966 MG |
1352 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
1353 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1354 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1355 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1356 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1357 | ||
1358 | /* Mask to get the watermark bits */ | |
1359 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1360 | ||
3148890b NP |
1361 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
1362 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1363 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1364 | |
933e312e AM |
1365 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1366 | ||
1367 | static struct fail_page_alloc_attr { | |
1368 | struct fault_attr attr; | |
1369 | ||
1370 | u32 ignore_gfp_highmem; | |
1371 | u32 ignore_gfp_wait; | |
54114994 | 1372 | u32 min_order; |
933e312e AM |
1373 | |
1374 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1375 | ||
1376 | struct dentry *ignore_gfp_highmem_file; | |
1377 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1378 | struct dentry *min_order_file; |
933e312e AM |
1379 | |
1380 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1381 | ||
1382 | } fail_page_alloc = { | |
1383 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1384 | .ignore_gfp_wait = 1, |
1385 | .ignore_gfp_highmem = 1, | |
54114994 | 1386 | .min_order = 1, |
933e312e AM |
1387 | }; |
1388 | ||
1389 | static int __init setup_fail_page_alloc(char *str) | |
1390 | { | |
1391 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1392 | } | |
1393 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1394 | ||
1395 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1396 | { | |
54114994 AM |
1397 | if (order < fail_page_alloc.min_order) |
1398 | return 0; | |
933e312e AM |
1399 | if (gfp_mask & __GFP_NOFAIL) |
1400 | return 0; | |
1401 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1402 | return 0; | |
1403 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1404 | return 0; | |
1405 | ||
1406 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1407 | } | |
1408 | ||
1409 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1410 | ||
1411 | static int __init fail_page_alloc_debugfs(void) | |
1412 | { | |
1413 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1414 | struct dentry *dir; | |
1415 | int err; | |
1416 | ||
1417 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1418 | "fail_page_alloc"); | |
1419 | if (err) | |
1420 | return err; | |
1421 | dir = fail_page_alloc.attr.dentries.dir; | |
1422 | ||
1423 | fail_page_alloc.ignore_gfp_wait_file = | |
1424 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1425 | &fail_page_alloc.ignore_gfp_wait); | |
1426 | ||
1427 | fail_page_alloc.ignore_gfp_highmem_file = | |
1428 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1429 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1430 | fail_page_alloc.min_order_file = |
1431 | debugfs_create_u32("min-order", mode, dir, | |
1432 | &fail_page_alloc.min_order); | |
933e312e AM |
1433 | |
1434 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1435 | !fail_page_alloc.ignore_gfp_highmem_file || |
1436 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1437 | err = -ENOMEM; |
1438 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1439 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1440 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1441 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1442 | } | |
1443 | ||
1444 | return err; | |
1445 | } | |
1446 | ||
1447 | late_initcall(fail_page_alloc_debugfs); | |
1448 | ||
1449 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1450 | ||
1451 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1452 | ||
1453 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1454 | { | |
1455 | return 0; | |
1456 | } | |
1457 | ||
1458 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1459 | ||
1da177e4 | 1460 | /* |
88f5acf8 | 1461 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1462 | * of the allocation. |
1463 | */ | |
88f5acf8 MG |
1464 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1465 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1466 | { |
1467 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1468 | long min = mark; |
1da177e4 LT |
1469 | int o; |
1470 | ||
88f5acf8 | 1471 | free_pages -= (1 << order) + 1; |
7fb1d9fc | 1472 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1473 | min -= min / 2; |
7fb1d9fc | 1474 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1475 | min -= min / 4; |
1476 | ||
1477 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 1478 | return false; |
1da177e4 LT |
1479 | for (o = 0; o < order; o++) { |
1480 | /* At the next order, this order's pages become unavailable */ | |
1481 | free_pages -= z->free_area[o].nr_free << o; | |
1482 | ||
1483 | /* Require fewer higher order pages to be free */ | |
1484 | min >>= 1; | |
1485 | ||
1486 | if (free_pages <= min) | |
88f5acf8 | 1487 | return false; |
1da177e4 | 1488 | } |
88f5acf8 MG |
1489 | return true; |
1490 | } | |
1491 | ||
1492 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1493 | int classzone_idx, int alloc_flags) | |
1494 | { | |
1495 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1496 | zone_page_state(z, NR_FREE_PAGES)); | |
1497 | } | |
1498 | ||
1499 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1500 | int classzone_idx, int alloc_flags) | |
1501 | { | |
1502 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1503 | ||
1504 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1505 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1506 | ||
1507 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1508 | free_pages); | |
1da177e4 LT |
1509 | } |
1510 | ||
9276b1bc PJ |
1511 | #ifdef CONFIG_NUMA |
1512 | /* | |
1513 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1514 | * skip over zones that are not allowed by the cpuset, or that have | |
1515 | * been recently (in last second) found to be nearly full. See further | |
1516 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1517 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1518 | * |
1519 | * If the zonelist cache is present in the passed in zonelist, then | |
1520 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1521 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1522 | * |
1523 | * If the zonelist cache is not available for this zonelist, does | |
1524 | * nothing and returns NULL. | |
1525 | * | |
1526 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1527 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1528 | * | |
1529 | * We hold off even calling zlc_setup, until after we've checked the | |
1530 | * first zone in the zonelist, on the theory that most allocations will | |
1531 | * be satisfied from that first zone, so best to examine that zone as | |
1532 | * quickly as we can. | |
1533 | */ | |
1534 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1535 | { | |
1536 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1537 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1538 | ||
1539 | zlc = zonelist->zlcache_ptr; | |
1540 | if (!zlc) | |
1541 | return NULL; | |
1542 | ||
f05111f5 | 1543 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1544 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1545 | zlc->last_full_zap = jiffies; | |
1546 | } | |
1547 | ||
1548 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1549 | &cpuset_current_mems_allowed : | |
37b07e41 | 1550 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1551 | return allowednodes; |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1556 | * if it is worth looking at further for free memory: | |
1557 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1558 | * bit set in the zonelist_cache fullzones BITMAP). | |
1559 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1560 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1561 | * Return true (non-zero) if zone is worth looking at further, or | |
1562 | * else return false (zero) if it is not. | |
1563 | * | |
1564 | * This check -ignores- the distinction between various watermarks, | |
1565 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1566 | * found to be full for any variation of these watermarks, it will | |
1567 | * be considered full for up to one second by all requests, unless | |
1568 | * we are so low on memory on all allowed nodes that we are forced | |
1569 | * into the second scan of the zonelist. | |
1570 | * | |
1571 | * In the second scan we ignore this zonelist cache and exactly | |
1572 | * apply the watermarks to all zones, even it is slower to do so. | |
1573 | * We are low on memory in the second scan, and should leave no stone | |
1574 | * unturned looking for a free page. | |
1575 | */ | |
dd1a239f | 1576 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1577 | nodemask_t *allowednodes) |
1578 | { | |
1579 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1580 | int i; /* index of *z in zonelist zones */ | |
1581 | int n; /* node that zone *z is on */ | |
1582 | ||
1583 | zlc = zonelist->zlcache_ptr; | |
1584 | if (!zlc) | |
1585 | return 1; | |
1586 | ||
dd1a239f | 1587 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1588 | n = zlc->z_to_n[i]; |
1589 | ||
1590 | /* This zone is worth trying if it is allowed but not full */ | |
1591 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1596 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1597 | * from that zone don't waste time re-examining it. | |
1598 | */ | |
dd1a239f | 1599 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1600 | { |
1601 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1602 | int i; /* index of *z in zonelist zones */ | |
1603 | ||
1604 | zlc = zonelist->zlcache_ptr; | |
1605 | if (!zlc) | |
1606 | return; | |
1607 | ||
dd1a239f | 1608 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1609 | |
1610 | set_bit(i, zlc->fullzones); | |
1611 | } | |
1612 | ||
1613 | #else /* CONFIG_NUMA */ | |
1614 | ||
1615 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1616 | { | |
1617 | return NULL; | |
1618 | } | |
1619 | ||
dd1a239f | 1620 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1621 | nodemask_t *allowednodes) |
1622 | { | |
1623 | return 1; | |
1624 | } | |
1625 | ||
dd1a239f | 1626 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1627 | { |
1628 | } | |
1629 | #endif /* CONFIG_NUMA */ | |
1630 | ||
7fb1d9fc | 1631 | /* |
0798e519 | 1632 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1633 | * a page. |
1634 | */ | |
1635 | static struct page * | |
19770b32 | 1636 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1637 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1638 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1639 | { |
dd1a239f | 1640 | struct zoneref *z; |
7fb1d9fc | 1641 | struct page *page = NULL; |
54a6eb5c | 1642 | int classzone_idx; |
5117f45d | 1643 | struct zone *zone; |
9276b1bc PJ |
1644 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1645 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1646 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1647 | |
19770b32 | 1648 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1649 | zonelist_scan: |
7fb1d9fc | 1650 | /* |
9276b1bc | 1651 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1652 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1653 | */ | |
19770b32 MG |
1654 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1655 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1656 | if (NUMA_BUILD && zlc_active && |
1657 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1658 | continue; | |
7fb1d9fc | 1659 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1660 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1661 | goto try_next_zone; |
7fb1d9fc | 1662 | |
41858966 | 1663 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1664 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1665 | unsigned long mark; |
fa5e084e MG |
1666 | int ret; |
1667 | ||
41858966 | 1668 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1669 | if (zone_watermark_ok(zone, order, mark, |
1670 | classzone_idx, alloc_flags)) | |
1671 | goto try_this_zone; | |
1672 | ||
1673 | if (zone_reclaim_mode == 0) | |
1674 | goto this_zone_full; | |
1675 | ||
1676 | ret = zone_reclaim(zone, gfp_mask, order); | |
1677 | switch (ret) { | |
1678 | case ZONE_RECLAIM_NOSCAN: | |
1679 | /* did not scan */ | |
1680 | goto try_next_zone; | |
1681 | case ZONE_RECLAIM_FULL: | |
1682 | /* scanned but unreclaimable */ | |
1683 | goto this_zone_full; | |
1684 | default: | |
1685 | /* did we reclaim enough */ | |
1686 | if (!zone_watermark_ok(zone, order, mark, | |
1687 | classzone_idx, alloc_flags)) | |
9276b1bc | 1688 | goto this_zone_full; |
0798e519 | 1689 | } |
7fb1d9fc RS |
1690 | } |
1691 | ||
fa5e084e | 1692 | try_this_zone: |
3dd28266 MG |
1693 | page = buffered_rmqueue(preferred_zone, zone, order, |
1694 | gfp_mask, migratetype); | |
0798e519 | 1695 | if (page) |
7fb1d9fc | 1696 | break; |
9276b1bc PJ |
1697 | this_zone_full: |
1698 | if (NUMA_BUILD) | |
1699 | zlc_mark_zone_full(zonelist, z); | |
1700 | try_next_zone: | |
62bc62a8 | 1701 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
d395b734 MG |
1702 | /* |
1703 | * we do zlc_setup after the first zone is tried but only | |
1704 | * if there are multiple nodes make it worthwhile | |
1705 | */ | |
9276b1bc PJ |
1706 | allowednodes = zlc_setup(zonelist, alloc_flags); |
1707 | zlc_active = 1; | |
1708 | did_zlc_setup = 1; | |
1709 | } | |
54a6eb5c | 1710 | } |
9276b1bc PJ |
1711 | |
1712 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1713 | /* Disable zlc cache for second zonelist scan */ | |
1714 | zlc_active = 0; | |
1715 | goto zonelist_scan; | |
1716 | } | |
7fb1d9fc | 1717 | return page; |
753ee728 MH |
1718 | } |
1719 | ||
11e33f6a MG |
1720 | static inline int |
1721 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1722 | unsigned long pages_reclaimed) | |
1da177e4 | 1723 | { |
11e33f6a MG |
1724 | /* Do not loop if specifically requested */ |
1725 | if (gfp_mask & __GFP_NORETRY) | |
1726 | return 0; | |
1da177e4 | 1727 | |
11e33f6a MG |
1728 | /* |
1729 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1730 | * means __GFP_NOFAIL, but that may not be true in other | |
1731 | * implementations. | |
1732 | */ | |
1733 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1734 | return 1; | |
1735 | ||
1736 | /* | |
1737 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1738 | * specified, then we retry until we no longer reclaim any pages | |
1739 | * (above), or we've reclaimed an order of pages at least as | |
1740 | * large as the allocation's order. In both cases, if the | |
1741 | * allocation still fails, we stop retrying. | |
1742 | */ | |
1743 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1744 | return 1; | |
cf40bd16 | 1745 | |
11e33f6a MG |
1746 | /* |
1747 | * Don't let big-order allocations loop unless the caller | |
1748 | * explicitly requests that. | |
1749 | */ | |
1750 | if (gfp_mask & __GFP_NOFAIL) | |
1751 | return 1; | |
1da177e4 | 1752 | |
11e33f6a MG |
1753 | return 0; |
1754 | } | |
933e312e | 1755 | |
11e33f6a MG |
1756 | static inline struct page * |
1757 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1758 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1759 | nodemask_t *nodemask, struct zone *preferred_zone, |
1760 | int migratetype) | |
11e33f6a MG |
1761 | { |
1762 | struct page *page; | |
1763 | ||
1764 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 1765 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 1766 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
1767 | return NULL; |
1768 | } | |
6b1de916 | 1769 | |
11e33f6a MG |
1770 | /* |
1771 | * Go through the zonelist yet one more time, keep very high watermark | |
1772 | * here, this is only to catch a parallel oom killing, we must fail if | |
1773 | * we're still under heavy pressure. | |
1774 | */ | |
1775 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1776 | order, zonelist, high_zoneidx, | |
5117f45d | 1777 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 1778 | preferred_zone, migratetype); |
7fb1d9fc | 1779 | if (page) |
11e33f6a MG |
1780 | goto out; |
1781 | ||
4365a567 KH |
1782 | if (!(gfp_mask & __GFP_NOFAIL)) { |
1783 | /* The OOM killer will not help higher order allocs */ | |
1784 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1785 | goto out; | |
03668b3c DR |
1786 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
1787 | if (high_zoneidx < ZONE_NORMAL) | |
1788 | goto out; | |
4365a567 KH |
1789 | /* |
1790 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
1791 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
1792 | * The caller should handle page allocation failure by itself if | |
1793 | * it specifies __GFP_THISNODE. | |
1794 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
1795 | */ | |
1796 | if (gfp_mask & __GFP_THISNODE) | |
1797 | goto out; | |
1798 | } | |
11e33f6a | 1799 | /* Exhausted what can be done so it's blamo time */ |
4365a567 | 1800 | out_of_memory(zonelist, gfp_mask, order, nodemask); |
11e33f6a MG |
1801 | |
1802 | out: | |
1803 | clear_zonelist_oom(zonelist, gfp_mask); | |
1804 | return page; | |
1805 | } | |
1806 | ||
56de7263 MG |
1807 | #ifdef CONFIG_COMPACTION |
1808 | /* Try memory compaction for high-order allocations before reclaim */ | |
1809 | static struct page * | |
1810 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
1811 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1812 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
77f1fe6b MG |
1813 | int migratetype, unsigned long *did_some_progress, |
1814 | bool sync_migration) | |
56de7263 MG |
1815 | { |
1816 | struct page *page; | |
3e7d3449 | 1817 | struct task_struct *tsk = current; |
56de7263 | 1818 | |
4f92e258 | 1819 | if (!order || compaction_deferred(preferred_zone)) |
56de7263 MG |
1820 | return NULL; |
1821 | ||
3e7d3449 | 1822 | tsk->flags |= PF_MEMALLOC; |
56de7263 | 1823 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
77f1fe6b | 1824 | nodemask, sync_migration); |
3e7d3449 | 1825 | tsk->flags &= ~PF_MEMALLOC; |
56de7263 MG |
1826 | if (*did_some_progress != COMPACT_SKIPPED) { |
1827 | ||
1828 | /* Page migration frees to the PCP lists but we want merging */ | |
1829 | drain_pages(get_cpu()); | |
1830 | put_cpu(); | |
1831 | ||
1832 | page = get_page_from_freelist(gfp_mask, nodemask, | |
1833 | order, zonelist, high_zoneidx, | |
1834 | alloc_flags, preferred_zone, | |
1835 | migratetype); | |
1836 | if (page) { | |
4f92e258 MG |
1837 | preferred_zone->compact_considered = 0; |
1838 | preferred_zone->compact_defer_shift = 0; | |
56de7263 MG |
1839 | count_vm_event(COMPACTSUCCESS); |
1840 | return page; | |
1841 | } | |
1842 | ||
1843 | /* | |
1844 | * It's bad if compaction run occurs and fails. | |
1845 | * The most likely reason is that pages exist, | |
1846 | * but not enough to satisfy watermarks. | |
1847 | */ | |
1848 | count_vm_event(COMPACTFAIL); | |
4f92e258 | 1849 | defer_compaction(preferred_zone); |
56de7263 MG |
1850 | |
1851 | cond_resched(); | |
1852 | } | |
1853 | ||
1854 | return NULL; | |
1855 | } | |
1856 | #else | |
1857 | static inline struct page * | |
1858 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
1859 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1860 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
77f1fe6b MG |
1861 | int migratetype, unsigned long *did_some_progress, |
1862 | bool sync_migration) | |
56de7263 MG |
1863 | { |
1864 | return NULL; | |
1865 | } | |
1866 | #endif /* CONFIG_COMPACTION */ | |
1867 | ||
11e33f6a MG |
1868 | /* The really slow allocator path where we enter direct reclaim */ |
1869 | static inline struct page * | |
1870 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1871 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
5117f45d | 1872 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
3dd28266 | 1873 | int migratetype, unsigned long *did_some_progress) |
11e33f6a MG |
1874 | { |
1875 | struct page *page = NULL; | |
1876 | struct reclaim_state reclaim_state; | |
1877 | struct task_struct *p = current; | |
9ee493ce | 1878 | bool drained = false; |
11e33f6a MG |
1879 | |
1880 | cond_resched(); | |
1881 | ||
1882 | /* We now go into synchronous reclaim */ | |
1883 | cpuset_memory_pressure_bump(); | |
11e33f6a MG |
1884 | p->flags |= PF_MEMALLOC; |
1885 | lockdep_set_current_reclaim_state(gfp_mask); | |
1886 | reclaim_state.reclaimed_slab = 0; | |
1887 | p->reclaim_state = &reclaim_state; | |
1888 | ||
1889 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1890 | ||
1891 | p->reclaim_state = NULL; | |
1892 | lockdep_clear_current_reclaim_state(); | |
1893 | p->flags &= ~PF_MEMALLOC; | |
1894 | ||
1895 | cond_resched(); | |
1896 | ||
9ee493ce MG |
1897 | if (unlikely(!(*did_some_progress))) |
1898 | return NULL; | |
11e33f6a | 1899 | |
9ee493ce MG |
1900 | retry: |
1901 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1902 | zonelist, high_zoneidx, |
3dd28266 MG |
1903 | alloc_flags, preferred_zone, |
1904 | migratetype); | |
9ee493ce MG |
1905 | |
1906 | /* | |
1907 | * If an allocation failed after direct reclaim, it could be because | |
1908 | * pages are pinned on the per-cpu lists. Drain them and try again | |
1909 | */ | |
1910 | if (!page && !drained) { | |
1911 | drain_all_pages(); | |
1912 | drained = true; | |
1913 | goto retry; | |
1914 | } | |
1915 | ||
11e33f6a MG |
1916 | return page; |
1917 | } | |
1918 | ||
1da177e4 | 1919 | /* |
11e33f6a MG |
1920 | * This is called in the allocator slow-path if the allocation request is of |
1921 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 1922 | */ |
11e33f6a MG |
1923 | static inline struct page * |
1924 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1925 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1926 | nodemask_t *nodemask, struct zone *preferred_zone, |
1927 | int migratetype) | |
11e33f6a MG |
1928 | { |
1929 | struct page *page; | |
1930 | ||
1931 | do { | |
1932 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 1933 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 1934 | preferred_zone, migratetype); |
11e33f6a MG |
1935 | |
1936 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 1937 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
1938 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
1939 | ||
1940 | return page; | |
1941 | } | |
1942 | ||
1943 | static inline | |
1944 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
99504748 MG |
1945 | enum zone_type high_zoneidx, |
1946 | enum zone_type classzone_idx) | |
1da177e4 | 1947 | { |
dd1a239f MG |
1948 | struct zoneref *z; |
1949 | struct zone *zone; | |
1da177e4 | 1950 | |
11e33f6a | 1951 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
99504748 | 1952 | wakeup_kswapd(zone, order, classzone_idx); |
11e33f6a | 1953 | } |
cf40bd16 | 1954 | |
341ce06f PZ |
1955 | static inline int |
1956 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
1957 | { | |
1958 | struct task_struct *p = current; | |
1959 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; | |
1960 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 1961 | |
a56f57ff | 1962 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 1963 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 1964 | |
341ce06f PZ |
1965 | /* |
1966 | * The caller may dip into page reserves a bit more if the caller | |
1967 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1968 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1969 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1970 | */ | |
e6223a3b | 1971 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 1972 | |
341ce06f | 1973 | if (!wait) { |
5c3240d9 AA |
1974 | /* |
1975 | * Not worth trying to allocate harder for | |
1976 | * __GFP_NOMEMALLOC even if it can't schedule. | |
1977 | */ | |
1978 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
1979 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 1980 | /* |
341ce06f PZ |
1981 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
1982 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 1983 | */ |
341ce06f | 1984 | alloc_flags &= ~ALLOC_CPUSET; |
9d0ed60f | 1985 | } else if (unlikely(rt_task(p)) && !in_interrupt()) |
341ce06f PZ |
1986 | alloc_flags |= ALLOC_HARDER; |
1987 | ||
1988 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
1989 | if (!in_interrupt() && | |
1990 | ((p->flags & PF_MEMALLOC) || | |
1991 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
1992 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1da177e4 | 1993 | } |
6b1de916 | 1994 | |
341ce06f PZ |
1995 | return alloc_flags; |
1996 | } | |
1997 | ||
11e33f6a MG |
1998 | static inline struct page * |
1999 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2000 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2001 | nodemask_t *nodemask, struct zone *preferred_zone, |
2002 | int migratetype) | |
11e33f6a MG |
2003 | { |
2004 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2005 | struct page *page = NULL; | |
2006 | int alloc_flags; | |
2007 | unsigned long pages_reclaimed = 0; | |
2008 | unsigned long did_some_progress; | |
2009 | struct task_struct *p = current; | |
77f1fe6b | 2010 | bool sync_migration = false; |
1da177e4 | 2011 | |
72807a74 MG |
2012 | /* |
2013 | * In the slowpath, we sanity check order to avoid ever trying to | |
2014 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2015 | * be using allocators in order of preference for an area that is | |
2016 | * too large. | |
2017 | */ | |
1fc28b70 MG |
2018 | if (order >= MAX_ORDER) { |
2019 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2020 | return NULL; |
1fc28b70 | 2021 | } |
1da177e4 | 2022 | |
952f3b51 CL |
2023 | /* |
2024 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2025 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2026 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2027 | * using a larger set of nodes after it has established that the | |
2028 | * allowed per node queues are empty and that nodes are | |
2029 | * over allocated. | |
2030 | */ | |
2031 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
2032 | goto nopage; | |
2033 | ||
cc4a6851 | 2034 | restart: |
32dba98e AA |
2035 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2036 | wake_all_kswapd(order, zonelist, high_zoneidx, | |
99504748 | 2037 | zone_idx(preferred_zone)); |
1da177e4 | 2038 | |
9bf2229f | 2039 | /* |
7fb1d9fc RS |
2040 | * OK, we're below the kswapd watermark and have kicked background |
2041 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2042 | * to how we want to proceed. | |
9bf2229f | 2043 | */ |
341ce06f | 2044 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2045 | |
341ce06f | 2046 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2047 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2048 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2049 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2050 | if (page) |
2051 | goto got_pg; | |
1da177e4 | 2052 | |
b43a57bb | 2053 | rebalance: |
11e33f6a | 2054 | /* Allocate without watermarks if the context allows */ |
341ce06f PZ |
2055 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
2056 | page = __alloc_pages_high_priority(gfp_mask, order, | |
2057 | zonelist, high_zoneidx, nodemask, | |
2058 | preferred_zone, migratetype); | |
2059 | if (page) | |
2060 | goto got_pg; | |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | /* Atomic allocations - we can't balance anything */ | |
2064 | if (!wait) | |
2065 | goto nopage; | |
2066 | ||
341ce06f PZ |
2067 | /* Avoid recursion of direct reclaim */ |
2068 | if (p->flags & PF_MEMALLOC) | |
2069 | goto nopage; | |
2070 | ||
6583bb64 DR |
2071 | /* Avoid allocations with no watermarks from looping endlessly */ |
2072 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2073 | goto nopage; | |
2074 | ||
77f1fe6b MG |
2075 | /* |
2076 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2077 | * attempts after direct reclaim are synchronous | |
2078 | */ | |
56de7263 MG |
2079 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2080 | zonelist, high_zoneidx, | |
2081 | nodemask, | |
2082 | alloc_flags, preferred_zone, | |
77f1fe6b MG |
2083 | migratetype, &did_some_progress, |
2084 | sync_migration); | |
56de7263 MG |
2085 | if (page) |
2086 | goto got_pg; | |
77f1fe6b | 2087 | sync_migration = true; |
56de7263 | 2088 | |
11e33f6a MG |
2089 | /* Try direct reclaim and then allocating */ |
2090 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2091 | zonelist, high_zoneidx, | |
2092 | nodemask, | |
5117f45d | 2093 | alloc_flags, preferred_zone, |
3dd28266 | 2094 | migratetype, &did_some_progress); |
11e33f6a MG |
2095 | if (page) |
2096 | goto got_pg; | |
1da177e4 | 2097 | |
e33c3b5e | 2098 | /* |
11e33f6a MG |
2099 | * If we failed to make any progress reclaiming, then we are |
2100 | * running out of options and have to consider going OOM | |
e33c3b5e | 2101 | */ |
11e33f6a MG |
2102 | if (!did_some_progress) { |
2103 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
2104 | if (oom_killer_disabled) |
2105 | goto nopage; | |
11e33f6a MG |
2106 | page = __alloc_pages_may_oom(gfp_mask, order, |
2107 | zonelist, high_zoneidx, | |
3dd28266 MG |
2108 | nodemask, preferred_zone, |
2109 | migratetype); | |
11e33f6a MG |
2110 | if (page) |
2111 | goto got_pg; | |
1da177e4 | 2112 | |
03668b3c DR |
2113 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2114 | /* | |
2115 | * The oom killer is not called for high-order | |
2116 | * allocations that may fail, so if no progress | |
2117 | * is being made, there are no other options and | |
2118 | * retrying is unlikely to help. | |
2119 | */ | |
2120 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2121 | goto nopage; | |
2122 | /* | |
2123 | * The oom killer is not called for lowmem | |
2124 | * allocations to prevent needlessly killing | |
2125 | * innocent tasks. | |
2126 | */ | |
2127 | if (high_zoneidx < ZONE_NORMAL) | |
2128 | goto nopage; | |
2129 | } | |
e2c55dc8 | 2130 | |
ff0ceb9d DR |
2131 | goto restart; |
2132 | } | |
1da177e4 LT |
2133 | } |
2134 | ||
11e33f6a | 2135 | /* Check if we should retry the allocation */ |
a41f24ea | 2136 | pages_reclaimed += did_some_progress; |
11e33f6a MG |
2137 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
2138 | /* Wait for some write requests to complete then retry */ | |
0e093d99 | 2139 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2140 | goto rebalance; |
3e7d3449 MG |
2141 | } else { |
2142 | /* | |
2143 | * High-order allocations do not necessarily loop after | |
2144 | * direct reclaim and reclaim/compaction depends on compaction | |
2145 | * being called after reclaim so call directly if necessary | |
2146 | */ | |
2147 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2148 | zonelist, high_zoneidx, | |
2149 | nodemask, | |
2150 | alloc_flags, preferred_zone, | |
77f1fe6b MG |
2151 | migratetype, &did_some_progress, |
2152 | sync_migration); | |
3e7d3449 MG |
2153 | if (page) |
2154 | goto got_pg; | |
1da177e4 LT |
2155 | } |
2156 | ||
2157 | nopage: | |
2158 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
2159 | printk(KERN_WARNING "%s: page allocation failure." | |
2160 | " order:%d, mode:0x%x\n", | |
2161 | p->comm, order, gfp_mask); | |
2162 | dump_stack(); | |
578c2fd6 | 2163 | show_mem(); |
1da177e4 | 2164 | } |
b1eeab67 | 2165 | return page; |
1da177e4 | 2166 | got_pg: |
b1eeab67 VN |
2167 | if (kmemcheck_enabled) |
2168 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1da177e4 | 2169 | return page; |
11e33f6a | 2170 | |
1da177e4 | 2171 | } |
11e33f6a MG |
2172 | |
2173 | /* | |
2174 | * This is the 'heart' of the zoned buddy allocator. | |
2175 | */ | |
2176 | struct page * | |
2177 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2178 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2179 | { | |
2180 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2181 | struct zone *preferred_zone; |
11e33f6a | 2182 | struct page *page; |
3dd28266 | 2183 | int migratetype = allocflags_to_migratetype(gfp_mask); |
11e33f6a | 2184 | |
dcce284a BH |
2185 | gfp_mask &= gfp_allowed_mask; |
2186 | ||
11e33f6a MG |
2187 | lockdep_trace_alloc(gfp_mask); |
2188 | ||
2189 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2190 | ||
2191 | if (should_fail_alloc_page(gfp_mask, order)) | |
2192 | return NULL; | |
2193 | ||
2194 | /* | |
2195 | * Check the zones suitable for the gfp_mask contain at least one | |
2196 | * valid zone. It's possible to have an empty zonelist as a result | |
2197 | * of GFP_THISNODE and a memoryless node | |
2198 | */ | |
2199 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2200 | return NULL; | |
2201 | ||
c0ff7453 | 2202 | get_mems_allowed(); |
5117f45d MG |
2203 | /* The preferred zone is used for statistics later */ |
2204 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); | |
c0ff7453 MX |
2205 | if (!preferred_zone) { |
2206 | put_mems_allowed(); | |
5117f45d | 2207 | return NULL; |
c0ff7453 | 2208 | } |
5117f45d MG |
2209 | |
2210 | /* First allocation attempt */ | |
11e33f6a | 2211 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
5117f45d | 2212 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
3dd28266 | 2213 | preferred_zone, migratetype); |
11e33f6a MG |
2214 | if (unlikely(!page)) |
2215 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 2216 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2217 | preferred_zone, migratetype); |
c0ff7453 | 2218 | put_mems_allowed(); |
11e33f6a | 2219 | |
4b4f278c | 2220 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
11e33f6a | 2221 | return page; |
1da177e4 | 2222 | } |
d239171e | 2223 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2224 | |
2225 | /* | |
2226 | * Common helper functions. | |
2227 | */ | |
920c7a5d | 2228 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2229 | { |
945a1113 AM |
2230 | struct page *page; |
2231 | ||
2232 | /* | |
2233 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2234 | * a highmem page | |
2235 | */ | |
2236 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2237 | ||
1da177e4 LT |
2238 | page = alloc_pages(gfp_mask, order); |
2239 | if (!page) | |
2240 | return 0; | |
2241 | return (unsigned long) page_address(page); | |
2242 | } | |
1da177e4 LT |
2243 | EXPORT_SYMBOL(__get_free_pages); |
2244 | ||
920c7a5d | 2245 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2246 | { |
945a1113 | 2247 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2248 | } |
1da177e4 LT |
2249 | EXPORT_SYMBOL(get_zeroed_page); |
2250 | ||
2251 | void __pagevec_free(struct pagevec *pvec) | |
2252 | { | |
2253 | int i = pagevec_count(pvec); | |
2254 | ||
4b4f278c MG |
2255 | while (--i >= 0) { |
2256 | trace_mm_pagevec_free(pvec->pages[i], pvec->cold); | |
1da177e4 | 2257 | free_hot_cold_page(pvec->pages[i], pvec->cold); |
4b4f278c | 2258 | } |
1da177e4 LT |
2259 | } |
2260 | ||
920c7a5d | 2261 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2262 | { |
b5810039 | 2263 | if (put_page_testzero(page)) { |
1da177e4 | 2264 | if (order == 0) |
fc91668e | 2265 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2266 | else |
2267 | __free_pages_ok(page, order); | |
2268 | } | |
2269 | } | |
2270 | ||
2271 | EXPORT_SYMBOL(__free_pages); | |
2272 | ||
920c7a5d | 2273 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2274 | { |
2275 | if (addr != 0) { | |
725d704e | 2276 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2277 | __free_pages(virt_to_page((void *)addr), order); |
2278 | } | |
2279 | } | |
2280 | ||
2281 | EXPORT_SYMBOL(free_pages); | |
2282 | ||
2be0ffe2 TT |
2283 | /** |
2284 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2285 | * @size: the number of bytes to allocate | |
2286 | * @gfp_mask: GFP flags for the allocation | |
2287 | * | |
2288 | * This function is similar to alloc_pages(), except that it allocates the | |
2289 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2290 | * allocate memory in power-of-two pages. | |
2291 | * | |
2292 | * This function is also limited by MAX_ORDER. | |
2293 | * | |
2294 | * Memory allocated by this function must be released by free_pages_exact(). | |
2295 | */ | |
2296 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2297 | { | |
2298 | unsigned int order = get_order(size); | |
2299 | unsigned long addr; | |
2300 | ||
2301 | addr = __get_free_pages(gfp_mask, order); | |
2302 | if (addr) { | |
2303 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2304 | unsigned long used = addr + PAGE_ALIGN(size); | |
2305 | ||
5bfd7560 | 2306 | split_page(virt_to_page((void *)addr), order); |
2be0ffe2 TT |
2307 | while (used < alloc_end) { |
2308 | free_page(used); | |
2309 | used += PAGE_SIZE; | |
2310 | } | |
2311 | } | |
2312 | ||
2313 | return (void *)addr; | |
2314 | } | |
2315 | EXPORT_SYMBOL(alloc_pages_exact); | |
2316 | ||
2317 | /** | |
2318 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2319 | * @virt: the value returned by alloc_pages_exact. | |
2320 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2321 | * | |
2322 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2323 | */ | |
2324 | void free_pages_exact(void *virt, size_t size) | |
2325 | { | |
2326 | unsigned long addr = (unsigned long)virt; | |
2327 | unsigned long end = addr + PAGE_ALIGN(size); | |
2328 | ||
2329 | while (addr < end) { | |
2330 | free_page(addr); | |
2331 | addr += PAGE_SIZE; | |
2332 | } | |
2333 | } | |
2334 | EXPORT_SYMBOL(free_pages_exact); | |
2335 | ||
1da177e4 LT |
2336 | static unsigned int nr_free_zone_pages(int offset) |
2337 | { | |
dd1a239f | 2338 | struct zoneref *z; |
54a6eb5c MG |
2339 | struct zone *zone; |
2340 | ||
e310fd43 | 2341 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2342 | unsigned int sum = 0; |
2343 | ||
0e88460d | 2344 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2345 | |
54a6eb5c | 2346 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2347 | unsigned long size = zone->present_pages; |
41858966 | 2348 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2349 | if (size > high) |
2350 | sum += size - high; | |
1da177e4 LT |
2351 | } |
2352 | ||
2353 | return sum; | |
2354 | } | |
2355 | ||
2356 | /* | |
2357 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2358 | */ | |
2359 | unsigned int nr_free_buffer_pages(void) | |
2360 | { | |
af4ca457 | 2361 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2362 | } |
c2f1a551 | 2363 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2364 | |
2365 | /* | |
2366 | * Amount of free RAM allocatable within all zones | |
2367 | */ | |
2368 | unsigned int nr_free_pagecache_pages(void) | |
2369 | { | |
2a1e274a | 2370 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2371 | } |
08e0f6a9 CL |
2372 | |
2373 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2374 | { |
08e0f6a9 | 2375 | if (NUMA_BUILD) |
25ba77c1 | 2376 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2377 | } |
1da177e4 | 2378 | |
1da177e4 LT |
2379 | void si_meminfo(struct sysinfo *val) |
2380 | { | |
2381 | val->totalram = totalram_pages; | |
2382 | val->sharedram = 0; | |
d23ad423 | 2383 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2384 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2385 | val->totalhigh = totalhigh_pages; |
2386 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2387 | val->mem_unit = PAGE_SIZE; |
2388 | } | |
2389 | ||
2390 | EXPORT_SYMBOL(si_meminfo); | |
2391 | ||
2392 | #ifdef CONFIG_NUMA | |
2393 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2394 | { | |
2395 | pg_data_t *pgdat = NODE_DATA(nid); | |
2396 | ||
2397 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2398 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2399 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2400 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2401 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2402 | NR_FREE_PAGES); | |
98d2b0eb CL |
2403 | #else |
2404 | val->totalhigh = 0; | |
2405 | val->freehigh = 0; | |
2406 | #endif | |
1da177e4 LT |
2407 | val->mem_unit = PAGE_SIZE; |
2408 | } | |
2409 | #endif | |
2410 | ||
2411 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
2412 | ||
2413 | /* | |
2414 | * Show free area list (used inside shift_scroll-lock stuff) | |
2415 | * We also calculate the percentage fragmentation. We do this by counting the | |
2416 | * memory on each free list with the exception of the first item on the list. | |
2417 | */ | |
2418 | void show_free_areas(void) | |
2419 | { | |
c7241913 | 2420 | int cpu; |
1da177e4 LT |
2421 | struct zone *zone; |
2422 | ||
ee99c71c | 2423 | for_each_populated_zone(zone) { |
c7241913 JS |
2424 | show_node(zone); |
2425 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2426 | |
6b482c67 | 2427 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2428 | struct per_cpu_pageset *pageset; |
2429 | ||
99dcc3e5 | 2430 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 2431 | |
3dfa5721 CL |
2432 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2433 | cpu, pageset->pcp.high, | |
2434 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2435 | } |
2436 | } | |
2437 | ||
a731286d KM |
2438 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2439 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2440 | " unevictable:%lu" |
b76146ed | 2441 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2442 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
4b02108a | 2443 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe | 2444 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2445 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2446 | global_page_state(NR_ISOLATED_ANON), |
2447 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2448 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2449 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2450 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2451 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2452 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2453 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2454 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2455 | global_page_state(NR_SLAB_RECLAIMABLE), |
2456 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2457 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2458 | global_page_state(NR_SHMEM), |
a25700a5 AM |
2459 | global_page_state(NR_PAGETABLE), |
2460 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2461 | |
ee99c71c | 2462 | for_each_populated_zone(zone) { |
1da177e4 LT |
2463 | int i; |
2464 | ||
2465 | show_node(zone); | |
2466 | printk("%s" | |
2467 | " free:%lukB" | |
2468 | " min:%lukB" | |
2469 | " low:%lukB" | |
2470 | " high:%lukB" | |
4f98a2fe RR |
2471 | " active_anon:%lukB" |
2472 | " inactive_anon:%lukB" | |
2473 | " active_file:%lukB" | |
2474 | " inactive_file:%lukB" | |
7b854121 | 2475 | " unevictable:%lukB" |
a731286d KM |
2476 | " isolated(anon):%lukB" |
2477 | " isolated(file):%lukB" | |
1da177e4 | 2478 | " present:%lukB" |
4a0aa73f KM |
2479 | " mlocked:%lukB" |
2480 | " dirty:%lukB" | |
2481 | " writeback:%lukB" | |
2482 | " mapped:%lukB" | |
4b02108a | 2483 | " shmem:%lukB" |
4a0aa73f KM |
2484 | " slab_reclaimable:%lukB" |
2485 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 2486 | " kernel_stack:%lukB" |
4a0aa73f KM |
2487 | " pagetables:%lukB" |
2488 | " unstable:%lukB" | |
2489 | " bounce:%lukB" | |
2490 | " writeback_tmp:%lukB" | |
1da177e4 LT |
2491 | " pages_scanned:%lu" |
2492 | " all_unreclaimable? %s" | |
2493 | "\n", | |
2494 | zone->name, | |
88f5acf8 | 2495 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
2496 | K(min_wmark_pages(zone)), |
2497 | K(low_wmark_pages(zone)), | |
2498 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
2499 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2500 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2501 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2502 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 2503 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
2504 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
2505 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 2506 | K(zone->present_pages), |
4a0aa73f KM |
2507 | K(zone_page_state(zone, NR_MLOCK)), |
2508 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
2509 | K(zone_page_state(zone, NR_WRITEBACK)), | |
2510 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 2511 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
2512 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
2513 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
2514 | zone_page_state(zone, NR_KERNEL_STACK) * |
2515 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
2516 | K(zone_page_state(zone, NR_PAGETABLE)), |
2517 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
2518 | K(zone_page_state(zone, NR_BOUNCE)), | |
2519 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), | |
1da177e4 | 2520 | zone->pages_scanned, |
93e4a89a | 2521 | (zone->all_unreclaimable ? "yes" : "no") |
1da177e4 LT |
2522 | ); |
2523 | printk("lowmem_reserve[]:"); | |
2524 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2525 | printk(" %lu", zone->lowmem_reserve[i]); | |
2526 | printk("\n"); | |
2527 | } | |
2528 | ||
ee99c71c | 2529 | for_each_populated_zone(zone) { |
8f9de51a | 2530 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 LT |
2531 | |
2532 | show_node(zone); | |
2533 | printk("%s: ", zone->name); | |
1da177e4 LT |
2534 | |
2535 | spin_lock_irqsave(&zone->lock, flags); | |
2536 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2537 | nr[order] = zone->free_area[order].nr_free; |
2538 | total += nr[order] << order; | |
1da177e4 LT |
2539 | } |
2540 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2541 | for (order = 0; order < MAX_ORDER; order++) |
2542 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2543 | printk("= %lukB\n", K(total)); |
2544 | } | |
2545 | ||
e6f3602d LW |
2546 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2547 | ||
1da177e4 LT |
2548 | show_swap_cache_info(); |
2549 | } | |
2550 | ||
19770b32 MG |
2551 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2552 | { | |
2553 | zoneref->zone = zone; | |
2554 | zoneref->zone_idx = zone_idx(zone); | |
2555 | } | |
2556 | ||
1da177e4 LT |
2557 | /* |
2558 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2559 | * |
2560 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2561 | */ |
f0c0b2b8 KH |
2562 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2563 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2564 | { |
1a93205b CL |
2565 | struct zone *zone; |
2566 | ||
98d2b0eb | 2567 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2568 | zone_type++; |
02a68a5e CL |
2569 | |
2570 | do { | |
2f6726e5 | 2571 | zone_type--; |
070f8032 | 2572 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2573 | if (populated_zone(zone)) { |
dd1a239f MG |
2574 | zoneref_set_zone(zone, |
2575 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2576 | check_highest_zone(zone_type); |
1da177e4 | 2577 | } |
02a68a5e | 2578 | |
2f6726e5 | 2579 | } while (zone_type); |
070f8032 | 2580 | return nr_zones; |
1da177e4 LT |
2581 | } |
2582 | ||
f0c0b2b8 KH |
2583 | |
2584 | /* | |
2585 | * zonelist_order: | |
2586 | * 0 = automatic detection of better ordering. | |
2587 | * 1 = order by ([node] distance, -zonetype) | |
2588 | * 2 = order by (-zonetype, [node] distance) | |
2589 | * | |
2590 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2591 | * the same zonelist. So only NUMA can configure this param. | |
2592 | */ | |
2593 | #define ZONELIST_ORDER_DEFAULT 0 | |
2594 | #define ZONELIST_ORDER_NODE 1 | |
2595 | #define ZONELIST_ORDER_ZONE 2 | |
2596 | ||
2597 | /* zonelist order in the kernel. | |
2598 | * set_zonelist_order() will set this to NODE or ZONE. | |
2599 | */ | |
2600 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2601 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2602 | ||
2603 | ||
1da177e4 | 2604 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2605 | /* The value user specified ....changed by config */ |
2606 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2607 | /* string for sysctl */ | |
2608 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2609 | char numa_zonelist_order[16] = "default"; | |
2610 | ||
2611 | /* | |
2612 | * interface for configure zonelist ordering. | |
2613 | * command line option "numa_zonelist_order" | |
2614 | * = "[dD]efault - default, automatic configuration. | |
2615 | * = "[nN]ode - order by node locality, then by zone within node | |
2616 | * = "[zZ]one - order by zone, then by locality within zone | |
2617 | */ | |
2618 | ||
2619 | static int __parse_numa_zonelist_order(char *s) | |
2620 | { | |
2621 | if (*s == 'd' || *s == 'D') { | |
2622 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2623 | } else if (*s == 'n' || *s == 'N') { | |
2624 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2625 | } else if (*s == 'z' || *s == 'Z') { | |
2626 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2627 | } else { | |
2628 | printk(KERN_WARNING | |
2629 | "Ignoring invalid numa_zonelist_order value: " | |
2630 | "%s\n", s); | |
2631 | return -EINVAL; | |
2632 | } | |
2633 | return 0; | |
2634 | } | |
2635 | ||
2636 | static __init int setup_numa_zonelist_order(char *s) | |
2637 | { | |
ecb256f8 VL |
2638 | int ret; |
2639 | ||
2640 | if (!s) | |
2641 | return 0; | |
2642 | ||
2643 | ret = __parse_numa_zonelist_order(s); | |
2644 | if (ret == 0) | |
2645 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
2646 | ||
2647 | return ret; | |
f0c0b2b8 KH |
2648 | } |
2649 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2650 | ||
2651 | /* | |
2652 | * sysctl handler for numa_zonelist_order | |
2653 | */ | |
2654 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 2655 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
2656 | loff_t *ppos) |
2657 | { | |
2658 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2659 | int ret; | |
443c6f14 | 2660 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 2661 | |
443c6f14 | 2662 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 2663 | if (write) |
443c6f14 | 2664 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 2665 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 2666 | if (ret) |
443c6f14 | 2667 | goto out; |
f0c0b2b8 KH |
2668 | if (write) { |
2669 | int oldval = user_zonelist_order; | |
2670 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2671 | /* | |
2672 | * bogus value. restore saved string | |
2673 | */ | |
2674 | strncpy((char*)table->data, saved_string, | |
2675 | NUMA_ZONELIST_ORDER_LEN); | |
2676 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
2677 | } else if (oldval != user_zonelist_order) { |
2678 | mutex_lock(&zonelists_mutex); | |
1f522509 | 2679 | build_all_zonelists(NULL); |
4eaf3f64 HL |
2680 | mutex_unlock(&zonelists_mutex); |
2681 | } | |
f0c0b2b8 | 2682 | } |
443c6f14 AK |
2683 | out: |
2684 | mutex_unlock(&zl_order_mutex); | |
2685 | return ret; | |
f0c0b2b8 KH |
2686 | } |
2687 | ||
2688 | ||
62bc62a8 | 2689 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
2690 | static int node_load[MAX_NUMNODES]; |
2691 | ||
1da177e4 | 2692 | /** |
4dc3b16b | 2693 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2694 | * @node: node whose fallback list we're appending |
2695 | * @used_node_mask: nodemask_t of already used nodes | |
2696 | * | |
2697 | * We use a number of factors to determine which is the next node that should | |
2698 | * appear on a given node's fallback list. The node should not have appeared | |
2699 | * already in @node's fallback list, and it should be the next closest node | |
2700 | * according to the distance array (which contains arbitrary distance values | |
2701 | * from each node to each node in the system), and should also prefer nodes | |
2702 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2703 | * on them otherwise. | |
2704 | * It returns -1 if no node is found. | |
2705 | */ | |
f0c0b2b8 | 2706 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2707 | { |
4cf808eb | 2708 | int n, val; |
1da177e4 LT |
2709 | int min_val = INT_MAX; |
2710 | int best_node = -1; | |
a70f7302 | 2711 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2712 | |
4cf808eb LT |
2713 | /* Use the local node if we haven't already */ |
2714 | if (!node_isset(node, *used_node_mask)) { | |
2715 | node_set(node, *used_node_mask); | |
2716 | return node; | |
2717 | } | |
1da177e4 | 2718 | |
37b07e41 | 2719 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2720 | |
2721 | /* Don't want a node to appear more than once */ | |
2722 | if (node_isset(n, *used_node_mask)) | |
2723 | continue; | |
2724 | ||
1da177e4 LT |
2725 | /* Use the distance array to find the distance */ |
2726 | val = node_distance(node, n); | |
2727 | ||
4cf808eb LT |
2728 | /* Penalize nodes under us ("prefer the next node") */ |
2729 | val += (n < node); | |
2730 | ||
1da177e4 | 2731 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2732 | tmp = cpumask_of_node(n); |
2733 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2734 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2735 | ||
2736 | /* Slight preference for less loaded node */ | |
2737 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2738 | val += node_load[n]; | |
2739 | ||
2740 | if (val < min_val) { | |
2741 | min_val = val; | |
2742 | best_node = n; | |
2743 | } | |
2744 | } | |
2745 | ||
2746 | if (best_node >= 0) | |
2747 | node_set(best_node, *used_node_mask); | |
2748 | ||
2749 | return best_node; | |
2750 | } | |
2751 | ||
f0c0b2b8 KH |
2752 | |
2753 | /* | |
2754 | * Build zonelists ordered by node and zones within node. | |
2755 | * This results in maximum locality--normal zone overflows into local | |
2756 | * DMA zone, if any--but risks exhausting DMA zone. | |
2757 | */ | |
2758 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2759 | { |
f0c0b2b8 | 2760 | int j; |
1da177e4 | 2761 | struct zonelist *zonelist; |
f0c0b2b8 | 2762 | |
54a6eb5c | 2763 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2764 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2765 | ; |
2766 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2767 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2768 | zonelist->_zonerefs[j].zone = NULL; |
2769 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2770 | } |
2771 | ||
523b9458 CL |
2772 | /* |
2773 | * Build gfp_thisnode zonelists | |
2774 | */ | |
2775 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2776 | { | |
523b9458 CL |
2777 | int j; |
2778 | struct zonelist *zonelist; | |
2779 | ||
54a6eb5c MG |
2780 | zonelist = &pgdat->node_zonelists[1]; |
2781 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
2782 | zonelist->_zonerefs[j].zone = NULL; |
2783 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
2784 | } |
2785 | ||
f0c0b2b8 KH |
2786 | /* |
2787 | * Build zonelists ordered by zone and nodes within zones. | |
2788 | * This results in conserving DMA zone[s] until all Normal memory is | |
2789 | * exhausted, but results in overflowing to remote node while memory | |
2790 | * may still exist in local DMA zone. | |
2791 | */ | |
2792 | static int node_order[MAX_NUMNODES]; | |
2793 | ||
2794 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2795 | { | |
f0c0b2b8 KH |
2796 | int pos, j, node; |
2797 | int zone_type; /* needs to be signed */ | |
2798 | struct zone *z; | |
2799 | struct zonelist *zonelist; | |
2800 | ||
54a6eb5c MG |
2801 | zonelist = &pgdat->node_zonelists[0]; |
2802 | pos = 0; | |
2803 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2804 | for (j = 0; j < nr_nodes; j++) { | |
2805 | node = node_order[j]; | |
2806 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2807 | if (populated_zone(z)) { | |
dd1a239f MG |
2808 | zoneref_set_zone(z, |
2809 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 2810 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
2811 | } |
2812 | } | |
f0c0b2b8 | 2813 | } |
dd1a239f MG |
2814 | zonelist->_zonerefs[pos].zone = NULL; |
2815 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
2816 | } |
2817 | ||
2818 | static int default_zonelist_order(void) | |
2819 | { | |
2820 | int nid, zone_type; | |
2821 | unsigned long low_kmem_size,total_size; | |
2822 | struct zone *z; | |
2823 | int average_size; | |
2824 | /* | |
88393161 | 2825 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
2826 | * If they are really small and used heavily, the system can fall |
2827 | * into OOM very easily. | |
e325c90f | 2828 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
2829 | */ |
2830 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2831 | low_kmem_size = 0; | |
2832 | total_size = 0; | |
2833 | for_each_online_node(nid) { | |
2834 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2835 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2836 | if (populated_zone(z)) { | |
2837 | if (zone_type < ZONE_NORMAL) | |
2838 | low_kmem_size += z->present_pages; | |
2839 | total_size += z->present_pages; | |
e325c90f DR |
2840 | } else if (zone_type == ZONE_NORMAL) { |
2841 | /* | |
2842 | * If any node has only lowmem, then node order | |
2843 | * is preferred to allow kernel allocations | |
2844 | * locally; otherwise, they can easily infringe | |
2845 | * on other nodes when there is an abundance of | |
2846 | * lowmem available to allocate from. | |
2847 | */ | |
2848 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
2849 | } |
2850 | } | |
2851 | } | |
2852 | if (!low_kmem_size || /* there are no DMA area. */ | |
2853 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2854 | return ZONELIST_ORDER_NODE; | |
2855 | /* | |
2856 | * look into each node's config. | |
2857 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2858 | * local memory, NODE_ORDER may be suitable. | |
2859 | */ | |
37b07e41 LS |
2860 | average_size = total_size / |
2861 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2862 | for_each_online_node(nid) { |
2863 | low_kmem_size = 0; | |
2864 | total_size = 0; | |
2865 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2866 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2867 | if (populated_zone(z)) { | |
2868 | if (zone_type < ZONE_NORMAL) | |
2869 | low_kmem_size += z->present_pages; | |
2870 | total_size += z->present_pages; | |
2871 | } | |
2872 | } | |
2873 | if (low_kmem_size && | |
2874 | total_size > average_size && /* ignore small node */ | |
2875 | low_kmem_size > total_size * 70/100) | |
2876 | return ZONELIST_ORDER_NODE; | |
2877 | } | |
2878 | return ZONELIST_ORDER_ZONE; | |
2879 | } | |
2880 | ||
2881 | static void set_zonelist_order(void) | |
2882 | { | |
2883 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2884 | current_zonelist_order = default_zonelist_order(); | |
2885 | else | |
2886 | current_zonelist_order = user_zonelist_order; | |
2887 | } | |
2888 | ||
2889 | static void build_zonelists(pg_data_t *pgdat) | |
2890 | { | |
2891 | int j, node, load; | |
2892 | enum zone_type i; | |
1da177e4 | 2893 | nodemask_t used_mask; |
f0c0b2b8 KH |
2894 | int local_node, prev_node; |
2895 | struct zonelist *zonelist; | |
2896 | int order = current_zonelist_order; | |
1da177e4 LT |
2897 | |
2898 | /* initialize zonelists */ | |
523b9458 | 2899 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 2900 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
2901 | zonelist->_zonerefs[0].zone = NULL; |
2902 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
2903 | } |
2904 | ||
2905 | /* NUMA-aware ordering of nodes */ | |
2906 | local_node = pgdat->node_id; | |
62bc62a8 | 2907 | load = nr_online_nodes; |
1da177e4 LT |
2908 | prev_node = local_node; |
2909 | nodes_clear(used_mask); | |
f0c0b2b8 | 2910 | |
f0c0b2b8 KH |
2911 | memset(node_order, 0, sizeof(node_order)); |
2912 | j = 0; | |
2913 | ||
1da177e4 | 2914 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2915 | int distance = node_distance(local_node, node); |
2916 | ||
2917 | /* | |
2918 | * If another node is sufficiently far away then it is better | |
2919 | * to reclaim pages in a zone before going off node. | |
2920 | */ | |
2921 | if (distance > RECLAIM_DISTANCE) | |
2922 | zone_reclaim_mode = 1; | |
2923 | ||
1da177e4 LT |
2924 | /* |
2925 | * We don't want to pressure a particular node. | |
2926 | * So adding penalty to the first node in same | |
2927 | * distance group to make it round-robin. | |
2928 | */ | |
9eeff239 | 2929 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2930 | node_load[node] = load; |
2931 | ||
1da177e4 LT |
2932 | prev_node = node; |
2933 | load--; | |
f0c0b2b8 KH |
2934 | if (order == ZONELIST_ORDER_NODE) |
2935 | build_zonelists_in_node_order(pgdat, node); | |
2936 | else | |
2937 | node_order[j++] = node; /* remember order */ | |
2938 | } | |
1da177e4 | 2939 | |
f0c0b2b8 KH |
2940 | if (order == ZONELIST_ORDER_ZONE) { |
2941 | /* calculate node order -- i.e., DMA last! */ | |
2942 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2943 | } |
523b9458 CL |
2944 | |
2945 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2946 | } |
2947 | ||
9276b1bc | 2948 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2949 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 2950 | { |
54a6eb5c MG |
2951 | struct zonelist *zonelist; |
2952 | struct zonelist_cache *zlc; | |
dd1a239f | 2953 | struct zoneref *z; |
9276b1bc | 2954 | |
54a6eb5c MG |
2955 | zonelist = &pgdat->node_zonelists[0]; |
2956 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2957 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
2958 | for (z = zonelist->_zonerefs; z->zone; z++) |
2959 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
2960 | } |
2961 | ||
7aac7898 LS |
2962 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
2963 | /* | |
2964 | * Return node id of node used for "local" allocations. | |
2965 | * I.e., first node id of first zone in arg node's generic zonelist. | |
2966 | * Used for initializing percpu 'numa_mem', which is used primarily | |
2967 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
2968 | */ | |
2969 | int local_memory_node(int node) | |
2970 | { | |
2971 | struct zone *zone; | |
2972 | ||
2973 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
2974 | gfp_zone(GFP_KERNEL), | |
2975 | NULL, | |
2976 | &zone); | |
2977 | return zone->node; | |
2978 | } | |
2979 | #endif | |
f0c0b2b8 | 2980 | |
1da177e4 LT |
2981 | #else /* CONFIG_NUMA */ |
2982 | ||
f0c0b2b8 KH |
2983 | static void set_zonelist_order(void) |
2984 | { | |
2985 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2986 | } | |
2987 | ||
2988 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2989 | { |
19655d34 | 2990 | int node, local_node; |
54a6eb5c MG |
2991 | enum zone_type j; |
2992 | struct zonelist *zonelist; | |
1da177e4 LT |
2993 | |
2994 | local_node = pgdat->node_id; | |
1da177e4 | 2995 | |
54a6eb5c MG |
2996 | zonelist = &pgdat->node_zonelists[0]; |
2997 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 2998 | |
54a6eb5c MG |
2999 | /* |
3000 | * Now we build the zonelist so that it contains the zones | |
3001 | * of all the other nodes. | |
3002 | * We don't want to pressure a particular node, so when | |
3003 | * building the zones for node N, we make sure that the | |
3004 | * zones coming right after the local ones are those from | |
3005 | * node N+1 (modulo N) | |
3006 | */ | |
3007 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3008 | if (!node_online(node)) | |
3009 | continue; | |
3010 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3011 | MAX_NR_ZONES - 1); | |
1da177e4 | 3012 | } |
54a6eb5c MG |
3013 | for (node = 0; node < local_node; node++) { |
3014 | if (!node_online(node)) | |
3015 | continue; | |
3016 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3017 | MAX_NR_ZONES - 1); | |
3018 | } | |
3019 | ||
dd1a239f MG |
3020 | zonelist->_zonerefs[j].zone = NULL; |
3021 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3022 | } |
3023 | ||
9276b1bc | 3024 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3025 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3026 | { |
54a6eb5c | 3027 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3028 | } |
3029 | ||
1da177e4 LT |
3030 | #endif /* CONFIG_NUMA */ |
3031 | ||
99dcc3e5 CL |
3032 | /* |
3033 | * Boot pageset table. One per cpu which is going to be used for all | |
3034 | * zones and all nodes. The parameters will be set in such a way | |
3035 | * that an item put on a list will immediately be handed over to | |
3036 | * the buddy list. This is safe since pageset manipulation is done | |
3037 | * with interrupts disabled. | |
3038 | * | |
3039 | * The boot_pagesets must be kept even after bootup is complete for | |
3040 | * unused processors and/or zones. They do play a role for bootstrapping | |
3041 | * hotplugged processors. | |
3042 | * | |
3043 | * zoneinfo_show() and maybe other functions do | |
3044 | * not check if the processor is online before following the pageset pointer. | |
3045 | * Other parts of the kernel may not check if the zone is available. | |
3046 | */ | |
3047 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3048 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3049 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3050 | |
4eaf3f64 HL |
3051 | /* |
3052 | * Global mutex to protect against size modification of zonelists | |
3053 | * as well as to serialize pageset setup for the new populated zone. | |
3054 | */ | |
3055 | DEFINE_MUTEX(zonelists_mutex); | |
3056 | ||
9b1a4d38 | 3057 | /* return values int ....just for stop_machine() */ |
1f522509 | 3058 | static __init_refok int __build_all_zonelists(void *data) |
1da177e4 | 3059 | { |
6811378e | 3060 | int nid; |
99dcc3e5 | 3061 | int cpu; |
9276b1bc | 3062 | |
7f9cfb31 BL |
3063 | #ifdef CONFIG_NUMA |
3064 | memset(node_load, 0, sizeof(node_load)); | |
3065 | #endif | |
9276b1bc | 3066 | for_each_online_node(nid) { |
7ea1530a CL |
3067 | pg_data_t *pgdat = NODE_DATA(nid); |
3068 | ||
3069 | build_zonelists(pgdat); | |
3070 | build_zonelist_cache(pgdat); | |
9276b1bc | 3071 | } |
99dcc3e5 CL |
3072 | |
3073 | /* | |
3074 | * Initialize the boot_pagesets that are going to be used | |
3075 | * for bootstrapping processors. The real pagesets for | |
3076 | * each zone will be allocated later when the per cpu | |
3077 | * allocator is available. | |
3078 | * | |
3079 | * boot_pagesets are used also for bootstrapping offline | |
3080 | * cpus if the system is already booted because the pagesets | |
3081 | * are needed to initialize allocators on a specific cpu too. | |
3082 | * F.e. the percpu allocator needs the page allocator which | |
3083 | * needs the percpu allocator in order to allocate its pagesets | |
3084 | * (a chicken-egg dilemma). | |
3085 | */ | |
7aac7898 | 3086 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3087 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3088 | ||
7aac7898 LS |
3089 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3090 | /* | |
3091 | * We now know the "local memory node" for each node-- | |
3092 | * i.e., the node of the first zone in the generic zonelist. | |
3093 | * Set up numa_mem percpu variable for on-line cpus. During | |
3094 | * boot, only the boot cpu should be on-line; we'll init the | |
3095 | * secondary cpus' numa_mem as they come on-line. During | |
3096 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3097 | */ | |
3098 | if (cpu_online(cpu)) | |
3099 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3100 | #endif | |
3101 | } | |
3102 | ||
6811378e YG |
3103 | return 0; |
3104 | } | |
3105 | ||
4eaf3f64 HL |
3106 | /* |
3107 | * Called with zonelists_mutex held always | |
3108 | * unless system_state == SYSTEM_BOOTING. | |
3109 | */ | |
1f522509 | 3110 | void build_all_zonelists(void *data) |
6811378e | 3111 | { |
f0c0b2b8 KH |
3112 | set_zonelist_order(); |
3113 | ||
6811378e | 3114 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3115 | __build_all_zonelists(NULL); |
68ad8df4 | 3116 | mminit_verify_zonelist(); |
6811378e YG |
3117 | cpuset_init_current_mems_allowed(); |
3118 | } else { | |
183ff22b | 3119 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 3120 | of zonelist */ |
e9959f0f KH |
3121 | #ifdef CONFIG_MEMORY_HOTPLUG |
3122 | if (data) | |
3123 | setup_zone_pageset((struct zone *)data); | |
3124 | #endif | |
3125 | stop_machine(__build_all_zonelists, NULL, NULL); | |
6811378e YG |
3126 | /* cpuset refresh routine should be here */ |
3127 | } | |
bd1e22b8 | 3128 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3129 | /* |
3130 | * Disable grouping by mobility if the number of pages in the | |
3131 | * system is too low to allow the mechanism to work. It would be | |
3132 | * more accurate, but expensive to check per-zone. This check is | |
3133 | * made on memory-hotadd so a system can start with mobility | |
3134 | * disabled and enable it later | |
3135 | */ | |
d9c23400 | 3136 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3137 | page_group_by_mobility_disabled = 1; |
3138 | else | |
3139 | page_group_by_mobility_disabled = 0; | |
3140 | ||
3141 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3142 | "Total pages: %ld\n", | |
62bc62a8 | 3143 | nr_online_nodes, |
f0c0b2b8 | 3144 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3145 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3146 | vm_total_pages); |
3147 | #ifdef CONFIG_NUMA | |
3148 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3149 | #endif | |
1da177e4 LT |
3150 | } |
3151 | ||
3152 | /* | |
3153 | * Helper functions to size the waitqueue hash table. | |
3154 | * Essentially these want to choose hash table sizes sufficiently | |
3155 | * large so that collisions trying to wait on pages are rare. | |
3156 | * But in fact, the number of active page waitqueues on typical | |
3157 | * systems is ridiculously low, less than 200. So this is even | |
3158 | * conservative, even though it seems large. | |
3159 | * | |
3160 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3161 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3162 | */ | |
3163 | #define PAGES_PER_WAITQUEUE 256 | |
3164 | ||
cca448fe | 3165 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3166 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3167 | { |
3168 | unsigned long size = 1; | |
3169 | ||
3170 | pages /= PAGES_PER_WAITQUEUE; | |
3171 | ||
3172 | while (size < pages) | |
3173 | size <<= 1; | |
3174 | ||
3175 | /* | |
3176 | * Once we have dozens or even hundreds of threads sleeping | |
3177 | * on IO we've got bigger problems than wait queue collision. | |
3178 | * Limit the size of the wait table to a reasonable size. | |
3179 | */ | |
3180 | size = min(size, 4096UL); | |
3181 | ||
3182 | return max(size, 4UL); | |
3183 | } | |
cca448fe YG |
3184 | #else |
3185 | /* | |
3186 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3187 | * a suitable size for its wait_table. So we use the maximum size now. | |
3188 | * | |
3189 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3190 | * | |
3191 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3192 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3193 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3194 | * | |
3195 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3196 | * or more by the traditional way. (See above). It equals: | |
3197 | * | |
3198 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3199 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3200 | * powerpc (64K page size) : = (32G +16M)byte. | |
3201 | */ | |
3202 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3203 | { | |
3204 | return 4096UL; | |
3205 | } | |
3206 | #endif | |
1da177e4 LT |
3207 | |
3208 | /* | |
3209 | * This is an integer logarithm so that shifts can be used later | |
3210 | * to extract the more random high bits from the multiplicative | |
3211 | * hash function before the remainder is taken. | |
3212 | */ | |
3213 | static inline unsigned long wait_table_bits(unsigned long size) | |
3214 | { | |
3215 | return ffz(~size); | |
3216 | } | |
3217 | ||
3218 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
3219 | ||
56fd56b8 | 3220 | /* |
d9c23400 | 3221 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3222 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3223 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3224 | * higher will lead to a bigger reserve which will get freed as contiguous |
3225 | * blocks as reclaim kicks in | |
3226 | */ | |
3227 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3228 | { | |
3229 | unsigned long start_pfn, pfn, end_pfn; | |
3230 | struct page *page; | |
78986a67 MG |
3231 | unsigned long block_migratetype; |
3232 | int reserve; | |
56fd56b8 MG |
3233 | |
3234 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
3235 | start_pfn = zone->zone_start_pfn; | |
3236 | end_pfn = start_pfn + zone->spanned_pages; | |
41858966 | 3237 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3238 | pageblock_order; |
56fd56b8 | 3239 | |
78986a67 MG |
3240 | /* |
3241 | * Reserve blocks are generally in place to help high-order atomic | |
3242 | * allocations that are short-lived. A min_free_kbytes value that | |
3243 | * would result in more than 2 reserve blocks for atomic allocations | |
3244 | * is assumed to be in place to help anti-fragmentation for the | |
3245 | * future allocation of hugepages at runtime. | |
3246 | */ | |
3247 | reserve = min(2, reserve); | |
3248 | ||
d9c23400 | 3249 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3250 | if (!pfn_valid(pfn)) |
3251 | continue; | |
3252 | page = pfn_to_page(pfn); | |
3253 | ||
344c790e AL |
3254 | /* Watch out for overlapping nodes */ |
3255 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3256 | continue; | |
3257 | ||
56fd56b8 MG |
3258 | /* Blocks with reserved pages will never free, skip them. */ |
3259 | if (PageReserved(page)) | |
3260 | continue; | |
3261 | ||
3262 | block_migratetype = get_pageblock_migratetype(page); | |
3263 | ||
3264 | /* If this block is reserved, account for it */ | |
3265 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
3266 | reserve--; | |
3267 | continue; | |
3268 | } | |
3269 | ||
3270 | /* Suitable for reserving if this block is movable */ | |
3271 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
3272 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
3273 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
3274 | reserve--; | |
3275 | continue; | |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * If the reserve is met and this is a previous reserved block, | |
3280 | * take it back | |
3281 | */ | |
3282 | if (block_migratetype == MIGRATE_RESERVE) { | |
3283 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3284 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3285 | } | |
3286 | } | |
3287 | } | |
ac0e5b7a | 3288 | |
1da177e4 LT |
3289 | /* |
3290 | * Initially all pages are reserved - free ones are freed | |
3291 | * up by free_all_bootmem() once the early boot process is | |
3292 | * done. Non-atomic initialization, single-pass. | |
3293 | */ | |
c09b4240 | 3294 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3295 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3296 | { |
1da177e4 | 3297 | struct page *page; |
29751f69 AW |
3298 | unsigned long end_pfn = start_pfn + size; |
3299 | unsigned long pfn; | |
86051ca5 | 3300 | struct zone *z; |
1da177e4 | 3301 | |
22b31eec HD |
3302 | if (highest_memmap_pfn < end_pfn - 1) |
3303 | highest_memmap_pfn = end_pfn - 1; | |
3304 | ||
86051ca5 | 3305 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3306 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3307 | /* |
3308 | * There can be holes in boot-time mem_map[]s | |
3309 | * handed to this function. They do not | |
3310 | * exist on hotplugged memory. | |
3311 | */ | |
3312 | if (context == MEMMAP_EARLY) { | |
3313 | if (!early_pfn_valid(pfn)) | |
3314 | continue; | |
3315 | if (!early_pfn_in_nid(pfn, nid)) | |
3316 | continue; | |
3317 | } | |
d41dee36 AW |
3318 | page = pfn_to_page(pfn); |
3319 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 3320 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 3321 | init_page_count(page); |
1da177e4 LT |
3322 | reset_page_mapcount(page); |
3323 | SetPageReserved(page); | |
b2a0ac88 MG |
3324 | /* |
3325 | * Mark the block movable so that blocks are reserved for | |
3326 | * movable at startup. This will force kernel allocations | |
3327 | * to reserve their blocks rather than leaking throughout | |
3328 | * the address space during boot when many long-lived | |
56fd56b8 MG |
3329 | * kernel allocations are made. Later some blocks near |
3330 | * the start are marked MIGRATE_RESERVE by | |
3331 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
3332 | * |
3333 | * bitmap is created for zone's valid pfn range. but memmap | |
3334 | * can be created for invalid pages (for alignment) | |
3335 | * check here not to call set_pageblock_migratetype() against | |
3336 | * pfn out of zone. | |
b2a0ac88 | 3337 | */ |
86051ca5 KH |
3338 | if ((z->zone_start_pfn <= pfn) |
3339 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
3340 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 3341 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 3342 | |
1da177e4 LT |
3343 | INIT_LIST_HEAD(&page->lru); |
3344 | #ifdef WANT_PAGE_VIRTUAL | |
3345 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3346 | if (!is_highmem_idx(zone)) | |
3212c6be | 3347 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 3348 | #endif |
1da177e4 LT |
3349 | } |
3350 | } | |
3351 | ||
1e548deb | 3352 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3353 | { |
b2a0ac88 MG |
3354 | int order, t; |
3355 | for_each_migratetype_order(order, t) { | |
3356 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3357 | zone->free_area[order].nr_free = 0; |
3358 | } | |
3359 | } | |
3360 | ||
3361 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3362 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3363 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3364 | #endif |
3365 | ||
1d6f4e60 | 3366 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3367 | { |
3a6be87f | 3368 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3369 | int batch; |
3370 | ||
3371 | /* | |
3372 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3373 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3374 | * |
3375 | * OK, so we don't know how big the cache is. So guess. | |
3376 | */ | |
3377 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3378 | if (batch * PAGE_SIZE > 512 * 1024) |
3379 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3380 | batch /= 4; /* We effectively *= 4 below */ |
3381 | if (batch < 1) | |
3382 | batch = 1; | |
3383 | ||
3384 | /* | |
0ceaacc9 NP |
3385 | * Clamp the batch to a 2^n - 1 value. Having a power |
3386 | * of 2 value was found to be more likely to have | |
3387 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3388 | * |
0ceaacc9 NP |
3389 | * For example if 2 tasks are alternately allocating |
3390 | * batches of pages, one task can end up with a lot | |
3391 | * of pages of one half of the possible page colors | |
3392 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3393 | */ |
9155203a | 3394 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3395 | |
e7c8d5c9 | 3396 | return batch; |
3a6be87f DH |
3397 | |
3398 | #else | |
3399 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3400 | * conditions. | |
3401 | * | |
3402 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3403 | * of contiguous memory as there's no hardware page translation to | |
3404 | * assemble apparent contiguous memory from discontiguous pages. | |
3405 | * | |
3406 | * Queueing large contiguous runs of pages for batching, however, | |
3407 | * causes the pages to actually be freed in smaller chunks. As there | |
3408 | * can be a significant delay between the individual batches being | |
3409 | * recycled, this leads to the once large chunks of space being | |
3410 | * fragmented and becoming unavailable for high-order allocations. | |
3411 | */ | |
3412 | return 0; | |
3413 | #endif | |
e7c8d5c9 CL |
3414 | } |
3415 | ||
b69a7288 | 3416 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3417 | { |
3418 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3419 | int migratetype; |
2caaad41 | 3420 | |
1c6fe946 MD |
3421 | memset(p, 0, sizeof(*p)); |
3422 | ||
3dfa5721 | 3423 | pcp = &p->pcp; |
2caaad41 | 3424 | pcp->count = 0; |
2caaad41 CL |
3425 | pcp->high = 6 * batch; |
3426 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3427 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3428 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3429 | } |
3430 | ||
8ad4b1fb RS |
3431 | /* |
3432 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3433 | * to the value high for the pageset p. | |
3434 | */ | |
3435 | ||
3436 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3437 | unsigned long high) | |
3438 | { | |
3439 | struct per_cpu_pages *pcp; | |
3440 | ||
3dfa5721 | 3441 | pcp = &p->pcp; |
8ad4b1fb RS |
3442 | pcp->high = high; |
3443 | pcp->batch = max(1UL, high/4); | |
3444 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3445 | pcp->batch = PAGE_SHIFT * 8; | |
3446 | } | |
3447 | ||
319774e2 WF |
3448 | static __meminit void setup_zone_pageset(struct zone *zone) |
3449 | { | |
3450 | int cpu; | |
3451 | ||
3452 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
3453 | ||
3454 | for_each_possible_cpu(cpu) { | |
3455 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
3456 | ||
3457 | setup_pageset(pcp, zone_batchsize(zone)); | |
3458 | ||
3459 | if (percpu_pagelist_fraction) | |
3460 | setup_pagelist_highmark(pcp, | |
3461 | (zone->present_pages / | |
3462 | percpu_pagelist_fraction)); | |
3463 | } | |
3464 | } | |
3465 | ||
2caaad41 | 3466 | /* |
99dcc3e5 CL |
3467 | * Allocate per cpu pagesets and initialize them. |
3468 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 3469 | */ |
99dcc3e5 | 3470 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 3471 | { |
99dcc3e5 | 3472 | struct zone *zone; |
e7c8d5c9 | 3473 | |
319774e2 WF |
3474 | for_each_populated_zone(zone) |
3475 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
3476 | } |
3477 | ||
577a32f6 | 3478 | static noinline __init_refok |
cca448fe | 3479 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
3480 | { |
3481 | int i; | |
3482 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 3483 | size_t alloc_size; |
ed8ece2e DH |
3484 | |
3485 | /* | |
3486 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3487 | * per zone. | |
3488 | */ | |
02b694de YG |
3489 | zone->wait_table_hash_nr_entries = |
3490 | wait_table_hash_nr_entries(zone_size_pages); | |
3491 | zone->wait_table_bits = | |
3492 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
3493 | alloc_size = zone->wait_table_hash_nr_entries |
3494 | * sizeof(wait_queue_head_t); | |
3495 | ||
cd94b9db | 3496 | if (!slab_is_available()) { |
cca448fe YG |
3497 | zone->wait_table = (wait_queue_head_t *) |
3498 | alloc_bootmem_node(pgdat, alloc_size); | |
3499 | } else { | |
3500 | /* | |
3501 | * This case means that a zone whose size was 0 gets new memory | |
3502 | * via memory hot-add. | |
3503 | * But it may be the case that a new node was hot-added. In | |
3504 | * this case vmalloc() will not be able to use this new node's | |
3505 | * memory - this wait_table must be initialized to use this new | |
3506 | * node itself as well. | |
3507 | * To use this new node's memory, further consideration will be | |
3508 | * necessary. | |
3509 | */ | |
8691f3a7 | 3510 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3511 | } |
3512 | if (!zone->wait_table) | |
3513 | return -ENOMEM; | |
ed8ece2e | 3514 | |
02b694de | 3515 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3516 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3517 | |
3518 | return 0; | |
ed8ece2e DH |
3519 | } |
3520 | ||
112067f0 SL |
3521 | static int __zone_pcp_update(void *data) |
3522 | { | |
3523 | struct zone *zone = data; | |
3524 | int cpu; | |
3525 | unsigned long batch = zone_batchsize(zone), flags; | |
3526 | ||
2d30a1f6 | 3527 | for_each_possible_cpu(cpu) { |
112067f0 SL |
3528 | struct per_cpu_pageset *pset; |
3529 | struct per_cpu_pages *pcp; | |
3530 | ||
99dcc3e5 | 3531 | pset = per_cpu_ptr(zone->pageset, cpu); |
112067f0 SL |
3532 | pcp = &pset->pcp; |
3533 | ||
3534 | local_irq_save(flags); | |
5f8dcc21 | 3535 | free_pcppages_bulk(zone, pcp->count, pcp); |
112067f0 SL |
3536 | setup_pageset(pset, batch); |
3537 | local_irq_restore(flags); | |
3538 | } | |
3539 | return 0; | |
3540 | } | |
3541 | ||
3542 | void zone_pcp_update(struct zone *zone) | |
3543 | { | |
3544 | stop_machine(__zone_pcp_update, zone, NULL); | |
3545 | } | |
3546 | ||
c09b4240 | 3547 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 3548 | { |
99dcc3e5 CL |
3549 | /* |
3550 | * per cpu subsystem is not up at this point. The following code | |
3551 | * relies on the ability of the linker to provide the | |
3552 | * offset of a (static) per cpu variable into the per cpu area. | |
3553 | */ | |
3554 | zone->pageset = &boot_pageset; | |
ed8ece2e | 3555 | |
f5335c0f | 3556 | if (zone->present_pages) |
99dcc3e5 CL |
3557 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
3558 | zone->name, zone->present_pages, | |
3559 | zone_batchsize(zone)); | |
ed8ece2e DH |
3560 | } |
3561 | ||
718127cc YG |
3562 | __meminit int init_currently_empty_zone(struct zone *zone, |
3563 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3564 | unsigned long size, |
3565 | enum memmap_context context) | |
ed8ece2e DH |
3566 | { |
3567 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3568 | int ret; |
3569 | ret = zone_wait_table_init(zone, size); | |
3570 | if (ret) | |
3571 | return ret; | |
ed8ece2e DH |
3572 | pgdat->nr_zones = zone_idx(zone) + 1; |
3573 | ||
ed8ece2e DH |
3574 | zone->zone_start_pfn = zone_start_pfn; |
3575 | ||
708614e6 MG |
3576 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3577 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3578 | pgdat->node_id, | |
3579 | (unsigned long)zone_idx(zone), | |
3580 | zone_start_pfn, (zone_start_pfn + size)); | |
3581 | ||
1e548deb | 3582 | zone_init_free_lists(zone); |
718127cc YG |
3583 | |
3584 | return 0; | |
ed8ece2e DH |
3585 | } |
3586 | ||
c713216d MG |
3587 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3588 | /* | |
3589 | * Basic iterator support. Return the first range of PFNs for a node | |
3590 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3591 | */ | |
a3142c8e | 3592 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
3593 | { |
3594 | int i; | |
3595 | ||
3596 | for (i = 0; i < nr_nodemap_entries; i++) | |
3597 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3598 | return i; | |
3599 | ||
3600 | return -1; | |
3601 | } | |
3602 | ||
3603 | /* | |
3604 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 3605 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 3606 | */ |
a3142c8e | 3607 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
3608 | { |
3609 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3610 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3611 | return index; | |
3612 | ||
3613 | return -1; | |
3614 | } | |
3615 | ||
3616 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3617 | /* | |
3618 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3619 | * Architectures may implement their own version but if add_active_range() | |
3620 | * was used and there are no special requirements, this is a convenient | |
3621 | * alternative | |
3622 | */ | |
f2dbcfa7 | 3623 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
3624 | { |
3625 | int i; | |
3626 | ||
3627 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3628 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3629 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3630 | ||
3631 | if (start_pfn <= pfn && pfn < end_pfn) | |
3632 | return early_node_map[i].nid; | |
3633 | } | |
cc2559bc KH |
3634 | /* This is a memory hole */ |
3635 | return -1; | |
c713216d MG |
3636 | } |
3637 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3638 | ||
f2dbcfa7 KH |
3639 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3640 | { | |
cc2559bc KH |
3641 | int nid; |
3642 | ||
3643 | nid = __early_pfn_to_nid(pfn); | |
3644 | if (nid >= 0) | |
3645 | return nid; | |
3646 | /* just returns 0 */ | |
3647 | return 0; | |
f2dbcfa7 KH |
3648 | } |
3649 | ||
cc2559bc KH |
3650 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3651 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3652 | { | |
3653 | int nid; | |
3654 | ||
3655 | nid = __early_pfn_to_nid(pfn); | |
3656 | if (nid >= 0 && nid != node) | |
3657 | return false; | |
3658 | return true; | |
3659 | } | |
3660 | #endif | |
f2dbcfa7 | 3661 | |
c713216d MG |
3662 | /* Basic iterator support to walk early_node_map[] */ |
3663 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3664 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3665 | i = next_active_region_index_in_nid(i, nid)) | |
3666 | ||
3667 | /** | |
3668 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3669 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3670 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3671 | * |
3672 | * If an architecture guarantees that all ranges registered with | |
3673 | * add_active_ranges() contain no holes and may be freed, this | |
3674 | * this function may be used instead of calling free_bootmem() manually. | |
3675 | */ | |
3676 | void __init free_bootmem_with_active_regions(int nid, | |
3677 | unsigned long max_low_pfn) | |
3678 | { | |
3679 | int i; | |
3680 | ||
3681 | for_each_active_range_index_in_nid(i, nid) { | |
3682 | unsigned long size_pages = 0; | |
3683 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3684 | ||
3685 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3686 | continue; | |
3687 | ||
3688 | if (end_pfn > max_low_pfn) | |
3689 | end_pfn = max_low_pfn; | |
3690 | ||
3691 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3692 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3693 | PFN_PHYS(early_node_map[i].start_pfn), | |
3694 | size_pages << PAGE_SHIFT); | |
3695 | } | |
3696 | } | |
3697 | ||
edbe7d23 YL |
3698 | #ifdef CONFIG_HAVE_MEMBLOCK |
3699 | u64 __init find_memory_core_early(int nid, u64 size, u64 align, | |
3700 | u64 goal, u64 limit) | |
3701 | { | |
3702 | int i; | |
3703 | ||
3704 | /* Need to go over early_node_map to find out good range for node */ | |
3705 | for_each_active_range_index_in_nid(i, nid) { | |
3706 | u64 addr; | |
3707 | u64 ei_start, ei_last; | |
3708 | u64 final_start, final_end; | |
3709 | ||
3710 | ei_last = early_node_map[i].end_pfn; | |
3711 | ei_last <<= PAGE_SHIFT; | |
3712 | ei_start = early_node_map[i].start_pfn; | |
3713 | ei_start <<= PAGE_SHIFT; | |
3714 | ||
3715 | final_start = max(ei_start, goal); | |
3716 | final_end = min(ei_last, limit); | |
3717 | ||
3718 | if (final_start >= final_end) | |
3719 | continue; | |
3720 | ||
3721 | addr = memblock_find_in_range(final_start, final_end, size, align); | |
3722 | ||
3723 | if (addr == MEMBLOCK_ERROR) | |
3724 | continue; | |
3725 | ||
3726 | return addr; | |
3727 | } | |
3728 | ||
3729 | return MEMBLOCK_ERROR; | |
3730 | } | |
3731 | #endif | |
3732 | ||
08677214 YL |
3733 | int __init add_from_early_node_map(struct range *range, int az, |
3734 | int nr_range, int nid) | |
3735 | { | |
3736 | int i; | |
3737 | u64 start, end; | |
3738 | ||
3739 | /* need to go over early_node_map to find out good range for node */ | |
3740 | for_each_active_range_index_in_nid(i, nid) { | |
3741 | start = early_node_map[i].start_pfn; | |
3742 | end = early_node_map[i].end_pfn; | |
3743 | nr_range = add_range(range, az, nr_range, start, end); | |
3744 | } | |
3745 | return nr_range; | |
3746 | } | |
3747 | ||
2ee78f7b | 3748 | #ifdef CONFIG_NO_BOOTMEM |
08677214 YL |
3749 | void * __init __alloc_memory_core_early(int nid, u64 size, u64 align, |
3750 | u64 goal, u64 limit) | |
3751 | { | |
08677214 | 3752 | void *ptr; |
72d7c3b3 | 3753 | u64 addr; |
08677214 | 3754 | |
72d7c3b3 YL |
3755 | if (limit > memblock.current_limit) |
3756 | limit = memblock.current_limit; | |
b8ab9f82 | 3757 | |
72d7c3b3 | 3758 | addr = find_memory_core_early(nid, size, align, goal, limit); |
08677214 | 3759 | |
72d7c3b3 YL |
3760 | if (addr == MEMBLOCK_ERROR) |
3761 | return NULL; | |
08677214 | 3762 | |
72d7c3b3 YL |
3763 | ptr = phys_to_virt(addr); |
3764 | memset(ptr, 0, size); | |
3765 | memblock_x86_reserve_range(addr, addr + size, "BOOTMEM"); | |
3766 | /* | |
3767 | * The min_count is set to 0 so that bootmem allocated blocks | |
3768 | * are never reported as leaks. | |
3769 | */ | |
3770 | kmemleak_alloc(ptr, size, 0, 0); | |
3771 | return ptr; | |
08677214 | 3772 | } |
2ee78f7b | 3773 | #endif |
08677214 YL |
3774 | |
3775 | ||
b5bc6c0e YL |
3776 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
3777 | { | |
3778 | int i; | |
d52d53b8 | 3779 | int ret; |
b5bc6c0e | 3780 | |
d52d53b8 YL |
3781 | for_each_active_range_index_in_nid(i, nid) { |
3782 | ret = work_fn(early_node_map[i].start_pfn, | |
3783 | early_node_map[i].end_pfn, data); | |
3784 | if (ret) | |
3785 | break; | |
3786 | } | |
b5bc6c0e | 3787 | } |
c713216d MG |
3788 | /** |
3789 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3790 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3791 | * |
3792 | * If an architecture guarantees that all ranges registered with | |
3793 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3794 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3795 | */ |
3796 | void __init sparse_memory_present_with_active_regions(int nid) | |
3797 | { | |
3798 | int i; | |
3799 | ||
3800 | for_each_active_range_index_in_nid(i, nid) | |
3801 | memory_present(early_node_map[i].nid, | |
3802 | early_node_map[i].start_pfn, | |
3803 | early_node_map[i].end_pfn); | |
3804 | } | |
3805 | ||
3806 | /** | |
3807 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3808 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3809 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3810 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3811 | * |
3812 | * It returns the start and end page frame of a node based on information | |
3813 | * provided by an arch calling add_active_range(). If called for a node | |
3814 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3815 | * PFNs will be 0. |
c713216d | 3816 | */ |
a3142c8e | 3817 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3818 | unsigned long *start_pfn, unsigned long *end_pfn) |
3819 | { | |
3820 | int i; | |
3821 | *start_pfn = -1UL; | |
3822 | *end_pfn = 0; | |
3823 | ||
3824 | for_each_active_range_index_in_nid(i, nid) { | |
3825 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3826 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3827 | } | |
3828 | ||
633c0666 | 3829 | if (*start_pfn == -1UL) |
c713216d | 3830 | *start_pfn = 0; |
c713216d MG |
3831 | } |
3832 | ||
2a1e274a MG |
3833 | /* |
3834 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3835 | * assumption is made that zones within a node are ordered in monotonic | |
3836 | * increasing memory addresses so that the "highest" populated zone is used | |
3837 | */ | |
b69a7288 | 3838 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3839 | { |
3840 | int zone_index; | |
3841 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3842 | if (zone_index == ZONE_MOVABLE) | |
3843 | continue; | |
3844 | ||
3845 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3846 | arch_zone_lowest_possible_pfn[zone_index]) | |
3847 | break; | |
3848 | } | |
3849 | ||
3850 | VM_BUG_ON(zone_index == -1); | |
3851 | movable_zone = zone_index; | |
3852 | } | |
3853 | ||
3854 | /* | |
3855 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3856 | * because it is sized independant of architecture. Unlike the other zones, | |
3857 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3858 | * in each node depending on the size of each node and how evenly kernelcore | |
3859 | * is distributed. This helper function adjusts the zone ranges | |
3860 | * provided by the architecture for a given node by using the end of the | |
3861 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3862 | * zones within a node are in order of monotonic increases memory addresses | |
3863 | */ | |
b69a7288 | 3864 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
3865 | unsigned long zone_type, |
3866 | unsigned long node_start_pfn, | |
3867 | unsigned long node_end_pfn, | |
3868 | unsigned long *zone_start_pfn, | |
3869 | unsigned long *zone_end_pfn) | |
3870 | { | |
3871 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3872 | if (zone_movable_pfn[nid]) { | |
3873 | /* Size ZONE_MOVABLE */ | |
3874 | if (zone_type == ZONE_MOVABLE) { | |
3875 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3876 | *zone_end_pfn = min(node_end_pfn, | |
3877 | arch_zone_highest_possible_pfn[movable_zone]); | |
3878 | ||
3879 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3880 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3881 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3882 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3883 | ||
3884 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3885 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3886 | *zone_start_pfn = *zone_end_pfn; | |
3887 | } | |
3888 | } | |
3889 | ||
c713216d MG |
3890 | /* |
3891 | * Return the number of pages a zone spans in a node, including holes | |
3892 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3893 | */ | |
6ea6e688 | 3894 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3895 | unsigned long zone_type, |
3896 | unsigned long *ignored) | |
3897 | { | |
3898 | unsigned long node_start_pfn, node_end_pfn; | |
3899 | unsigned long zone_start_pfn, zone_end_pfn; | |
3900 | ||
3901 | /* Get the start and end of the node and zone */ | |
3902 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3903 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3904 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3905 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3906 | node_start_pfn, node_end_pfn, | |
3907 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3908 | |
3909 | /* Check that this node has pages within the zone's required range */ | |
3910 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3911 | return 0; | |
3912 | ||
3913 | /* Move the zone boundaries inside the node if necessary */ | |
3914 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3915 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3916 | ||
3917 | /* Return the spanned pages */ | |
3918 | return zone_end_pfn - zone_start_pfn; | |
3919 | } | |
3920 | ||
3921 | /* | |
3922 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3923 | * then all holes in the requested range will be accounted for. |
c713216d | 3924 | */ |
32996250 | 3925 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3926 | unsigned long range_start_pfn, |
3927 | unsigned long range_end_pfn) | |
3928 | { | |
3929 | int i = 0; | |
3930 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3931 | unsigned long start_pfn; | |
3932 | ||
3933 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3934 | i = first_active_region_index_in_nid(nid); | |
3935 | if (i == -1) | |
3936 | return 0; | |
3937 | ||
b5445f95 MG |
3938 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3939 | ||
9c7cd687 MG |
3940 | /* Account for ranges before physical memory on this node */ |
3941 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3942 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3943 | |
3944 | /* Find all holes for the zone within the node */ | |
3945 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3946 | ||
3947 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3948 | if (prev_end_pfn >= range_end_pfn) | |
3949 | break; | |
3950 | ||
3951 | /* Make sure the end of the zone is not within the hole */ | |
3952 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3953 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3954 | ||
3955 | /* Update the hole size cound and move on */ | |
3956 | if (start_pfn > range_start_pfn) { | |
3957 | BUG_ON(prev_end_pfn > start_pfn); | |
3958 | hole_pages += start_pfn - prev_end_pfn; | |
3959 | } | |
3960 | prev_end_pfn = early_node_map[i].end_pfn; | |
3961 | } | |
3962 | ||
9c7cd687 MG |
3963 | /* Account for ranges past physical memory on this node */ |
3964 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3965 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3966 | max(range_start_pfn, prev_end_pfn); |
3967 | ||
c713216d MG |
3968 | return hole_pages; |
3969 | } | |
3970 | ||
3971 | /** | |
3972 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3973 | * @start_pfn: The start PFN to start searching for holes | |
3974 | * @end_pfn: The end PFN to stop searching for holes | |
3975 | * | |
88ca3b94 | 3976 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3977 | */ |
3978 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3979 | unsigned long end_pfn) | |
3980 | { | |
3981 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3982 | } | |
3983 | ||
3984 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3985 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3986 | unsigned long zone_type, |
3987 | unsigned long *ignored) | |
3988 | { | |
9c7cd687 MG |
3989 | unsigned long node_start_pfn, node_end_pfn; |
3990 | unsigned long zone_start_pfn, zone_end_pfn; | |
3991 | ||
3992 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3993 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3994 | node_start_pfn); | |
3995 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3996 | node_end_pfn); | |
3997 | ||
2a1e274a MG |
3998 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3999 | node_start_pfn, node_end_pfn, | |
4000 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4001 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4002 | } |
0e0b864e | 4003 | |
c713216d | 4004 | #else |
6ea6e688 | 4005 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4006 | unsigned long zone_type, |
4007 | unsigned long *zones_size) | |
4008 | { | |
4009 | return zones_size[zone_type]; | |
4010 | } | |
4011 | ||
6ea6e688 | 4012 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4013 | unsigned long zone_type, |
4014 | unsigned long *zholes_size) | |
4015 | { | |
4016 | if (!zholes_size) | |
4017 | return 0; | |
4018 | ||
4019 | return zholes_size[zone_type]; | |
4020 | } | |
0e0b864e | 4021 | |
c713216d MG |
4022 | #endif |
4023 | ||
a3142c8e | 4024 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
4025 | unsigned long *zones_size, unsigned long *zholes_size) |
4026 | { | |
4027 | unsigned long realtotalpages, totalpages = 0; | |
4028 | enum zone_type i; | |
4029 | ||
4030 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4031 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
4032 | zones_size); | |
4033 | pgdat->node_spanned_pages = totalpages; | |
4034 | ||
4035 | realtotalpages = totalpages; | |
4036 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4037 | realtotalpages -= | |
4038 | zone_absent_pages_in_node(pgdat->node_id, i, | |
4039 | zholes_size); | |
4040 | pgdat->node_present_pages = realtotalpages; | |
4041 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4042 | realtotalpages); | |
4043 | } | |
4044 | ||
835c134e MG |
4045 | #ifndef CONFIG_SPARSEMEM |
4046 | /* | |
4047 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4048 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4049 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4050 | * round what is now in bits to nearest long in bits, then return it in |
4051 | * bytes. | |
4052 | */ | |
4053 | static unsigned long __init usemap_size(unsigned long zonesize) | |
4054 | { | |
4055 | unsigned long usemapsize; | |
4056 | ||
d9c23400 MG |
4057 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4058 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4059 | usemapsize *= NR_PAGEBLOCK_BITS; |
4060 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4061 | ||
4062 | return usemapsize / 8; | |
4063 | } | |
4064 | ||
4065 | static void __init setup_usemap(struct pglist_data *pgdat, | |
4066 | struct zone *zone, unsigned long zonesize) | |
4067 | { | |
4068 | unsigned long usemapsize = usemap_size(zonesize); | |
4069 | zone->pageblock_flags = NULL; | |
58a01a45 | 4070 | if (usemapsize) |
835c134e | 4071 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
835c134e MG |
4072 | } |
4073 | #else | |
fa9f90be | 4074 | static inline void setup_usemap(struct pglist_data *pgdat, |
835c134e MG |
4075 | struct zone *zone, unsigned long zonesize) {} |
4076 | #endif /* CONFIG_SPARSEMEM */ | |
4077 | ||
d9c23400 | 4078 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
4079 | |
4080 | /* Return a sensible default order for the pageblock size. */ | |
4081 | static inline int pageblock_default_order(void) | |
4082 | { | |
4083 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
4084 | return HUGETLB_PAGE_ORDER; | |
4085 | ||
4086 | return MAX_ORDER-1; | |
4087 | } | |
4088 | ||
d9c23400 MG |
4089 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
4090 | static inline void __init set_pageblock_order(unsigned int order) | |
4091 | { | |
4092 | /* Check that pageblock_nr_pages has not already been setup */ | |
4093 | if (pageblock_order) | |
4094 | return; | |
4095 | ||
4096 | /* | |
4097 | * Assume the largest contiguous order of interest is a huge page. | |
4098 | * This value may be variable depending on boot parameters on IA64 | |
4099 | */ | |
4100 | pageblock_order = order; | |
4101 | } | |
4102 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4103 | ||
ba72cb8c MG |
4104 | /* |
4105 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
4106 | * and pageblock_default_order() are unused as pageblock_order is set | |
4107 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
4108 | * pageblock_order based on the kernel config | |
4109 | */ | |
4110 | static inline int pageblock_default_order(unsigned int order) | |
4111 | { | |
4112 | return MAX_ORDER-1; | |
4113 | } | |
d9c23400 MG |
4114 | #define set_pageblock_order(x) do {} while (0) |
4115 | ||
4116 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4117 | ||
1da177e4 LT |
4118 | /* |
4119 | * Set up the zone data structures: | |
4120 | * - mark all pages reserved | |
4121 | * - mark all memory queues empty | |
4122 | * - clear the memory bitmaps | |
4123 | */ | |
b5a0e011 | 4124 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
4125 | unsigned long *zones_size, unsigned long *zholes_size) |
4126 | { | |
2f1b6248 | 4127 | enum zone_type j; |
ed8ece2e | 4128 | int nid = pgdat->node_id; |
1da177e4 | 4129 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4130 | int ret; |
1da177e4 | 4131 | |
208d54e5 | 4132 | pgdat_resize_init(pgdat); |
1da177e4 LT |
4133 | pgdat->nr_zones = 0; |
4134 | init_waitqueue_head(&pgdat->kswapd_wait); | |
4135 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 4136 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
4137 | |
4138 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
4139 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 4140 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 4141 | enum lru_list l; |
1da177e4 | 4142 | |
c713216d MG |
4143 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
4144 | realsize = size - zone_absent_pages_in_node(nid, j, | |
4145 | zholes_size); | |
1da177e4 | 4146 | |
0e0b864e MG |
4147 | /* |
4148 | * Adjust realsize so that it accounts for how much memory | |
4149 | * is used by this zone for memmap. This affects the watermark | |
4150 | * and per-cpu initialisations | |
4151 | */ | |
f7232154 JW |
4152 | memmap_pages = |
4153 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
4154 | if (realsize >= memmap_pages) { |
4155 | realsize -= memmap_pages; | |
5594c8c8 YL |
4156 | if (memmap_pages) |
4157 | printk(KERN_DEBUG | |
4158 | " %s zone: %lu pages used for memmap\n", | |
4159 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4160 | } else |
4161 | printk(KERN_WARNING | |
4162 | " %s zone: %lu pages exceeds realsize %lu\n", | |
4163 | zone_names[j], memmap_pages, realsize); | |
4164 | ||
6267276f CL |
4165 | /* Account for reserved pages */ |
4166 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 4167 | realsize -= dma_reserve; |
d903ef9f | 4168 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4169 | zone_names[0], dma_reserve); |
0e0b864e MG |
4170 | } |
4171 | ||
98d2b0eb | 4172 | if (!is_highmem_idx(j)) |
1da177e4 LT |
4173 | nr_kernel_pages += realsize; |
4174 | nr_all_pages += realsize; | |
4175 | ||
4176 | zone->spanned_pages = size; | |
4177 | zone->present_pages = realsize; | |
9614634f | 4178 | #ifdef CONFIG_NUMA |
d5f541ed | 4179 | zone->node = nid; |
8417bba4 | 4180 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 4181 | / 100; |
0ff38490 | 4182 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 4183 | #endif |
1da177e4 LT |
4184 | zone->name = zone_names[j]; |
4185 | spin_lock_init(&zone->lock); | |
4186 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4187 | zone_seqlock_init(zone); |
1da177e4 | 4188 | zone->zone_pgdat = pgdat; |
1da177e4 | 4189 | |
ed8ece2e | 4190 | zone_pcp_init(zone); |
b69408e8 CL |
4191 | for_each_lru(l) { |
4192 | INIT_LIST_HEAD(&zone->lru[l].list); | |
f8629631 | 4193 | zone->reclaim_stat.nr_saved_scan[l] = 0; |
b69408e8 | 4194 | } |
6e901571 KM |
4195 | zone->reclaim_stat.recent_rotated[0] = 0; |
4196 | zone->reclaim_stat.recent_rotated[1] = 0; | |
4197 | zone->reclaim_stat.recent_scanned[0] = 0; | |
4198 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 4199 | zap_zone_vm_stats(zone); |
e815af95 | 4200 | zone->flags = 0; |
1da177e4 LT |
4201 | if (!size) |
4202 | continue; | |
4203 | ||
ba72cb8c | 4204 | set_pageblock_order(pageblock_default_order()); |
835c134e | 4205 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
4206 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4207 | size, MEMMAP_EARLY); | |
718127cc | 4208 | BUG_ON(ret); |
76cdd58e | 4209 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4210 | zone_start_pfn += size; |
1da177e4 LT |
4211 | } |
4212 | } | |
4213 | ||
577a32f6 | 4214 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4215 | { |
1da177e4 LT |
4216 | /* Skip empty nodes */ |
4217 | if (!pgdat->node_spanned_pages) | |
4218 | return; | |
4219 | ||
d41dee36 | 4220 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4221 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4222 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4223 | unsigned long size, start, end; |
d41dee36 AW |
4224 | struct page *map; |
4225 | ||
e984bb43 BP |
4226 | /* |
4227 | * The zone's endpoints aren't required to be MAX_ORDER | |
4228 | * aligned but the node_mem_map endpoints must be in order | |
4229 | * for the buddy allocator to function correctly. | |
4230 | */ | |
4231 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
4232 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
4233 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
4234 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4235 | map = alloc_remap(pgdat->node_id, size); |
4236 | if (!map) | |
4237 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 4238 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4239 | } |
12d810c1 | 4240 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4241 | /* |
4242 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4243 | */ | |
c713216d | 4244 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4245 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
4246 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
4247 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 4248 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
4249 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4250 | } | |
1da177e4 | 4251 | #endif |
d41dee36 | 4252 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4253 | } |
4254 | ||
9109fb7b JW |
4255 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4256 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4257 | { |
9109fb7b JW |
4258 | pg_data_t *pgdat = NODE_DATA(nid); |
4259 | ||
1da177e4 LT |
4260 | pgdat->node_id = nid; |
4261 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 4262 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
4263 | |
4264 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4265 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4266 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4267 | nid, (unsigned long)pgdat, | |
4268 | (unsigned long)pgdat->node_mem_map); | |
4269 | #endif | |
1da177e4 LT |
4270 | |
4271 | free_area_init_core(pgdat, zones_size, zholes_size); | |
4272 | } | |
4273 | ||
c713216d | 4274 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
4275 | |
4276 | #if MAX_NUMNODES > 1 | |
4277 | /* | |
4278 | * Figure out the number of possible node ids. | |
4279 | */ | |
4280 | static void __init setup_nr_node_ids(void) | |
4281 | { | |
4282 | unsigned int node; | |
4283 | unsigned int highest = 0; | |
4284 | ||
4285 | for_each_node_mask(node, node_possible_map) | |
4286 | highest = node; | |
4287 | nr_node_ids = highest + 1; | |
4288 | } | |
4289 | #else | |
4290 | static inline void setup_nr_node_ids(void) | |
4291 | { | |
4292 | } | |
4293 | #endif | |
4294 | ||
c713216d MG |
4295 | /** |
4296 | * add_active_range - Register a range of PFNs backed by physical memory | |
4297 | * @nid: The node ID the range resides on | |
4298 | * @start_pfn: The start PFN of the available physical memory | |
4299 | * @end_pfn: The end PFN of the available physical memory | |
4300 | * | |
4301 | * These ranges are stored in an early_node_map[] and later used by | |
4302 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
4303 | * range spans a memory hole, it is up to the architecture to ensure | |
4304 | * the memory is not freed by the bootmem allocator. If possible | |
4305 | * the range being registered will be merged with existing ranges. | |
4306 | */ | |
4307 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
4308 | unsigned long end_pfn) | |
4309 | { | |
4310 | int i; | |
4311 | ||
6b74ab97 MG |
4312 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
4313 | "Entering add_active_range(%d, %#lx, %#lx) " | |
4314 | "%d entries of %d used\n", | |
4315 | nid, start_pfn, end_pfn, | |
4316 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
c713216d | 4317 | |
2dbb51c4 MG |
4318 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
4319 | ||
c713216d MG |
4320 | /* Merge with existing active regions if possible */ |
4321 | for (i = 0; i < nr_nodemap_entries; i++) { | |
4322 | if (early_node_map[i].nid != nid) | |
4323 | continue; | |
4324 | ||
4325 | /* Skip if an existing region covers this new one */ | |
4326 | if (start_pfn >= early_node_map[i].start_pfn && | |
4327 | end_pfn <= early_node_map[i].end_pfn) | |
4328 | return; | |
4329 | ||
4330 | /* Merge forward if suitable */ | |
4331 | if (start_pfn <= early_node_map[i].end_pfn && | |
4332 | end_pfn > early_node_map[i].end_pfn) { | |
4333 | early_node_map[i].end_pfn = end_pfn; | |
4334 | return; | |
4335 | } | |
4336 | ||
4337 | /* Merge backward if suitable */ | |
d2dbe08d | 4338 | if (start_pfn < early_node_map[i].start_pfn && |
c713216d MG |
4339 | end_pfn >= early_node_map[i].start_pfn) { |
4340 | early_node_map[i].start_pfn = start_pfn; | |
4341 | return; | |
4342 | } | |
4343 | } | |
4344 | ||
4345 | /* Check that early_node_map is large enough */ | |
4346 | if (i >= MAX_ACTIVE_REGIONS) { | |
4347 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
4348 | MAX_ACTIVE_REGIONS); | |
4349 | return; | |
4350 | } | |
4351 | ||
4352 | early_node_map[i].nid = nid; | |
4353 | early_node_map[i].start_pfn = start_pfn; | |
4354 | early_node_map[i].end_pfn = end_pfn; | |
4355 | nr_nodemap_entries = i + 1; | |
4356 | } | |
4357 | ||
4358 | /** | |
cc1050ba | 4359 | * remove_active_range - Shrink an existing registered range of PFNs |
c713216d | 4360 | * @nid: The node id the range is on that should be shrunk |
cc1050ba YL |
4361 | * @start_pfn: The new PFN of the range |
4362 | * @end_pfn: The new PFN of the range | |
c713216d MG |
4363 | * |
4364 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
cc1a9d86 YL |
4365 | * The map is kept near the end physical page range that has already been |
4366 | * registered. This function allows an arch to shrink an existing registered | |
4367 | * range. | |
c713216d | 4368 | */ |
cc1050ba YL |
4369 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
4370 | unsigned long end_pfn) | |
c713216d | 4371 | { |
cc1a9d86 YL |
4372 | int i, j; |
4373 | int removed = 0; | |
c713216d | 4374 | |
cc1050ba YL |
4375 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
4376 | nid, start_pfn, end_pfn); | |
4377 | ||
c713216d | 4378 | /* Find the old active region end and shrink */ |
cc1a9d86 | 4379 | for_each_active_range_index_in_nid(i, nid) { |
cc1050ba YL |
4380 | if (early_node_map[i].start_pfn >= start_pfn && |
4381 | early_node_map[i].end_pfn <= end_pfn) { | |
cc1a9d86 | 4382 | /* clear it */ |
cc1050ba | 4383 | early_node_map[i].start_pfn = 0; |
cc1a9d86 YL |
4384 | early_node_map[i].end_pfn = 0; |
4385 | removed = 1; | |
4386 | continue; | |
4387 | } | |
cc1050ba YL |
4388 | if (early_node_map[i].start_pfn < start_pfn && |
4389 | early_node_map[i].end_pfn > start_pfn) { | |
4390 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
4391 | early_node_map[i].end_pfn = start_pfn; | |
4392 | if (temp_end_pfn > end_pfn) | |
4393 | add_active_range(nid, end_pfn, temp_end_pfn); | |
4394 | continue; | |
4395 | } | |
4396 | if (early_node_map[i].start_pfn >= start_pfn && | |
4397 | early_node_map[i].end_pfn > end_pfn && | |
4398 | early_node_map[i].start_pfn < end_pfn) { | |
4399 | early_node_map[i].start_pfn = end_pfn; | |
cc1a9d86 | 4400 | continue; |
c713216d | 4401 | } |
cc1a9d86 YL |
4402 | } |
4403 | ||
4404 | if (!removed) | |
4405 | return; | |
4406 | ||
4407 | /* remove the blank ones */ | |
4408 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
4409 | if (early_node_map[i].nid != nid) | |
4410 | continue; | |
4411 | if (early_node_map[i].end_pfn) | |
4412 | continue; | |
4413 | /* we found it, get rid of it */ | |
4414 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
4415 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
4416 | sizeof(early_node_map[j])); | |
4417 | j = nr_nodemap_entries - 1; | |
4418 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
4419 | nr_nodemap_entries--; | |
4420 | } | |
c713216d MG |
4421 | } |
4422 | ||
4423 | /** | |
4424 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 4425 | * |
c713216d MG |
4426 | * During discovery, it may be found that a table like SRAT is invalid |
4427 | * and an alternative discovery method must be used. This function removes | |
4428 | * all currently registered regions. | |
4429 | */ | |
88ca3b94 | 4430 | void __init remove_all_active_ranges(void) |
c713216d MG |
4431 | { |
4432 | memset(early_node_map, 0, sizeof(early_node_map)); | |
4433 | nr_nodemap_entries = 0; | |
4434 | } | |
4435 | ||
4436 | /* Compare two active node_active_regions */ | |
4437 | static int __init cmp_node_active_region(const void *a, const void *b) | |
4438 | { | |
4439 | struct node_active_region *arange = (struct node_active_region *)a; | |
4440 | struct node_active_region *brange = (struct node_active_region *)b; | |
4441 | ||
4442 | /* Done this way to avoid overflows */ | |
4443 | if (arange->start_pfn > brange->start_pfn) | |
4444 | return 1; | |
4445 | if (arange->start_pfn < brange->start_pfn) | |
4446 | return -1; | |
4447 | ||
4448 | return 0; | |
4449 | } | |
4450 | ||
4451 | /* sort the node_map by start_pfn */ | |
32996250 | 4452 | void __init sort_node_map(void) |
c713216d MG |
4453 | { |
4454 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
4455 | sizeof(struct node_active_region), | |
4456 | cmp_node_active_region, NULL); | |
4457 | } | |
4458 | ||
a6af2bc3 | 4459 | /* Find the lowest pfn for a node */ |
b69a7288 | 4460 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d MG |
4461 | { |
4462 | int i; | |
a6af2bc3 | 4463 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 4464 | |
c713216d MG |
4465 | /* Assuming a sorted map, the first range found has the starting pfn */ |
4466 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 4467 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 4468 | |
a6af2bc3 MG |
4469 | if (min_pfn == ULONG_MAX) { |
4470 | printk(KERN_WARNING | |
2bc0d261 | 4471 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4472 | return 0; |
4473 | } | |
4474 | ||
4475 | return min_pfn; | |
c713216d MG |
4476 | } |
4477 | ||
4478 | /** | |
4479 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4480 | * | |
4481 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4482 | * add_active_range(). |
c713216d MG |
4483 | */ |
4484 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4485 | { | |
4486 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4487 | } | |
4488 | ||
37b07e41 LS |
4489 | /* |
4490 | * early_calculate_totalpages() | |
4491 | * Sum pages in active regions for movable zone. | |
4492 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4493 | */ | |
484f51f8 | 4494 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
4495 | { |
4496 | int i; | |
4497 | unsigned long totalpages = 0; | |
4498 | ||
37b07e41 LS |
4499 | for (i = 0; i < nr_nodemap_entries; i++) { |
4500 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 4501 | early_node_map[i].start_pfn; |
37b07e41 LS |
4502 | totalpages += pages; |
4503 | if (pages) | |
4504 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
4505 | } | |
4506 | return totalpages; | |
7e63efef MG |
4507 | } |
4508 | ||
2a1e274a MG |
4509 | /* |
4510 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4511 | * is spread evenly between nodes as long as the nodes have enough | |
4512 | * memory. When they don't, some nodes will have more kernelcore than | |
4513 | * others | |
4514 | */ | |
b69a7288 | 4515 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
4516 | { |
4517 | int i, nid; | |
4518 | unsigned long usable_startpfn; | |
4519 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4520 | /* save the state before borrow the nodemask */ |
4521 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4522 | unsigned long totalpages = early_calculate_totalpages(); |
4523 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4524 | |
7e63efef MG |
4525 | /* |
4526 | * If movablecore was specified, calculate what size of | |
4527 | * kernelcore that corresponds so that memory usable for | |
4528 | * any allocation type is evenly spread. If both kernelcore | |
4529 | * and movablecore are specified, then the value of kernelcore | |
4530 | * will be used for required_kernelcore if it's greater than | |
4531 | * what movablecore would have allowed. | |
4532 | */ | |
4533 | if (required_movablecore) { | |
7e63efef MG |
4534 | unsigned long corepages; |
4535 | ||
4536 | /* | |
4537 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4538 | * was requested by the user | |
4539 | */ | |
4540 | required_movablecore = | |
4541 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4542 | corepages = totalpages - required_movablecore; | |
4543 | ||
4544 | required_kernelcore = max(required_kernelcore, corepages); | |
4545 | } | |
4546 | ||
2a1e274a MG |
4547 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4548 | if (!required_kernelcore) | |
66918dcd | 4549 | goto out; |
2a1e274a MG |
4550 | |
4551 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4552 | find_usable_zone_for_movable(); | |
4553 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4554 | ||
4555 | restart: | |
4556 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4557 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4558 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
4559 | /* |
4560 | * Recalculate kernelcore_node if the division per node | |
4561 | * now exceeds what is necessary to satisfy the requested | |
4562 | * amount of memory for the kernel | |
4563 | */ | |
4564 | if (required_kernelcore < kernelcore_node) | |
4565 | kernelcore_node = required_kernelcore / usable_nodes; | |
4566 | ||
4567 | /* | |
4568 | * As the map is walked, we track how much memory is usable | |
4569 | * by the kernel using kernelcore_remaining. When it is | |
4570 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4571 | */ | |
4572 | kernelcore_remaining = kernelcore_node; | |
4573 | ||
4574 | /* Go through each range of PFNs within this node */ | |
4575 | for_each_active_range_index_in_nid(i, nid) { | |
4576 | unsigned long start_pfn, end_pfn; | |
4577 | unsigned long size_pages; | |
4578 | ||
4579 | start_pfn = max(early_node_map[i].start_pfn, | |
4580 | zone_movable_pfn[nid]); | |
4581 | end_pfn = early_node_map[i].end_pfn; | |
4582 | if (start_pfn >= end_pfn) | |
4583 | continue; | |
4584 | ||
4585 | /* Account for what is only usable for kernelcore */ | |
4586 | if (start_pfn < usable_startpfn) { | |
4587 | unsigned long kernel_pages; | |
4588 | kernel_pages = min(end_pfn, usable_startpfn) | |
4589 | - start_pfn; | |
4590 | ||
4591 | kernelcore_remaining -= min(kernel_pages, | |
4592 | kernelcore_remaining); | |
4593 | required_kernelcore -= min(kernel_pages, | |
4594 | required_kernelcore); | |
4595 | ||
4596 | /* Continue if range is now fully accounted */ | |
4597 | if (end_pfn <= usable_startpfn) { | |
4598 | ||
4599 | /* | |
4600 | * Push zone_movable_pfn to the end so | |
4601 | * that if we have to rebalance | |
4602 | * kernelcore across nodes, we will | |
4603 | * not double account here | |
4604 | */ | |
4605 | zone_movable_pfn[nid] = end_pfn; | |
4606 | continue; | |
4607 | } | |
4608 | start_pfn = usable_startpfn; | |
4609 | } | |
4610 | ||
4611 | /* | |
4612 | * The usable PFN range for ZONE_MOVABLE is from | |
4613 | * start_pfn->end_pfn. Calculate size_pages as the | |
4614 | * number of pages used as kernelcore | |
4615 | */ | |
4616 | size_pages = end_pfn - start_pfn; | |
4617 | if (size_pages > kernelcore_remaining) | |
4618 | size_pages = kernelcore_remaining; | |
4619 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4620 | ||
4621 | /* | |
4622 | * Some kernelcore has been met, update counts and | |
4623 | * break if the kernelcore for this node has been | |
4624 | * satisified | |
4625 | */ | |
4626 | required_kernelcore -= min(required_kernelcore, | |
4627 | size_pages); | |
4628 | kernelcore_remaining -= size_pages; | |
4629 | if (!kernelcore_remaining) | |
4630 | break; | |
4631 | } | |
4632 | } | |
4633 | ||
4634 | /* | |
4635 | * If there is still required_kernelcore, we do another pass with one | |
4636 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4637 | * along on the nodes that still have memory until kernelcore is | |
4638 | * satisified | |
4639 | */ | |
4640 | usable_nodes--; | |
4641 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4642 | goto restart; | |
4643 | ||
4644 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4645 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4646 | zone_movable_pfn[nid] = | |
4647 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4648 | |
4649 | out: | |
4650 | /* restore the node_state */ | |
4651 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4652 | } |
4653 | ||
37b07e41 LS |
4654 | /* Any regular memory on that node ? */ |
4655 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4656 | { | |
4657 | #ifdef CONFIG_HIGHMEM | |
4658 | enum zone_type zone_type; | |
4659 | ||
4660 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4661 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4662 | if (zone->present_pages) | |
4663 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4664 | } | |
4665 | #endif | |
4666 | } | |
4667 | ||
c713216d MG |
4668 | /** |
4669 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4670 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4671 | * |
4672 | * This will call free_area_init_node() for each active node in the system. | |
4673 | * Using the page ranges provided by add_active_range(), the size of each | |
4674 | * zone in each node and their holes is calculated. If the maximum PFN | |
4675 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4676 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4677 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4678 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4679 | * at arch_max_dma_pfn. | |
4680 | */ | |
4681 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4682 | { | |
4683 | unsigned long nid; | |
db99100d | 4684 | int i; |
c713216d | 4685 | |
a6af2bc3 MG |
4686 | /* Sort early_node_map as initialisation assumes it is sorted */ |
4687 | sort_node_map(); | |
4688 | ||
c713216d MG |
4689 | /* Record where the zone boundaries are */ |
4690 | memset(arch_zone_lowest_possible_pfn, 0, | |
4691 | sizeof(arch_zone_lowest_possible_pfn)); | |
4692 | memset(arch_zone_highest_possible_pfn, 0, | |
4693 | sizeof(arch_zone_highest_possible_pfn)); | |
4694 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4695 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4696 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4697 | if (i == ZONE_MOVABLE) |
4698 | continue; | |
c713216d MG |
4699 | arch_zone_lowest_possible_pfn[i] = |
4700 | arch_zone_highest_possible_pfn[i-1]; | |
4701 | arch_zone_highest_possible_pfn[i] = | |
4702 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4703 | } | |
2a1e274a MG |
4704 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4705 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4706 | ||
4707 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4708 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4709 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4710 | |
c713216d MG |
4711 | /* Print out the zone ranges */ |
4712 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4713 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4714 | if (i == ZONE_MOVABLE) | |
4715 | continue; | |
72f0ba02 DR |
4716 | printk(" %-8s ", zone_names[i]); |
4717 | if (arch_zone_lowest_possible_pfn[i] == | |
4718 | arch_zone_highest_possible_pfn[i]) | |
4719 | printk("empty\n"); | |
4720 | else | |
4721 | printk("%0#10lx -> %0#10lx\n", | |
c713216d MG |
4722 | arch_zone_lowest_possible_pfn[i], |
4723 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4724 | } |
4725 | ||
4726 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4727 | printk("Movable zone start PFN for each node\n"); | |
4728 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4729 | if (zone_movable_pfn[i]) | |
4730 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4731 | } | |
c713216d MG |
4732 | |
4733 | /* Print out the early_node_map[] */ | |
4734 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4735 | for (i = 0; i < nr_nodemap_entries; i++) | |
5dab8ec1 | 4736 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
c713216d MG |
4737 | early_node_map[i].start_pfn, |
4738 | early_node_map[i].end_pfn); | |
4739 | ||
4740 | /* Initialise every node */ | |
708614e6 | 4741 | mminit_verify_pageflags_layout(); |
8ef82866 | 4742 | setup_nr_node_ids(); |
c713216d MG |
4743 | for_each_online_node(nid) { |
4744 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4745 | free_area_init_node(nid, NULL, |
c713216d | 4746 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4747 | |
4748 | /* Any memory on that node */ | |
4749 | if (pgdat->node_present_pages) | |
4750 | node_set_state(nid, N_HIGH_MEMORY); | |
4751 | check_for_regular_memory(pgdat); | |
c713216d MG |
4752 | } |
4753 | } | |
2a1e274a | 4754 | |
7e63efef | 4755 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4756 | { |
4757 | unsigned long long coremem; | |
4758 | if (!p) | |
4759 | return -EINVAL; | |
4760 | ||
4761 | coremem = memparse(p, &p); | |
7e63efef | 4762 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4763 | |
7e63efef | 4764 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4765 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4766 | ||
4767 | return 0; | |
4768 | } | |
ed7ed365 | 4769 | |
7e63efef MG |
4770 | /* |
4771 | * kernelcore=size sets the amount of memory for use for allocations that | |
4772 | * cannot be reclaimed or migrated. | |
4773 | */ | |
4774 | static int __init cmdline_parse_kernelcore(char *p) | |
4775 | { | |
4776 | return cmdline_parse_core(p, &required_kernelcore); | |
4777 | } | |
4778 | ||
4779 | /* | |
4780 | * movablecore=size sets the amount of memory for use for allocations that | |
4781 | * can be reclaimed or migrated. | |
4782 | */ | |
4783 | static int __init cmdline_parse_movablecore(char *p) | |
4784 | { | |
4785 | return cmdline_parse_core(p, &required_movablecore); | |
4786 | } | |
4787 | ||
ed7ed365 | 4788 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4789 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4790 | |
c713216d MG |
4791 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
4792 | ||
0e0b864e | 4793 | /** |
88ca3b94 RD |
4794 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4795 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4796 | * |
4797 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4798 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4799 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4800 | * function may optionally be used to account for unfreeable pages in the |
4801 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4802 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4803 | */ |
4804 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4805 | { | |
4806 | dma_reserve = new_dma_reserve; | |
4807 | } | |
4808 | ||
93b7504e | 4809 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
08677214 YL |
4810 | struct pglist_data __refdata contig_page_data = { |
4811 | #ifndef CONFIG_NO_BOOTMEM | |
4812 | .bdata = &bootmem_node_data[0] | |
4813 | #endif | |
4814 | }; | |
1da177e4 | 4815 | EXPORT_SYMBOL(contig_page_data); |
93b7504e | 4816 | #endif |
1da177e4 LT |
4817 | |
4818 | void __init free_area_init(unsigned long *zones_size) | |
4819 | { | |
9109fb7b | 4820 | free_area_init_node(0, zones_size, |
1da177e4 LT |
4821 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
4822 | } | |
1da177e4 | 4823 | |
1da177e4 LT |
4824 | static int page_alloc_cpu_notify(struct notifier_block *self, |
4825 | unsigned long action, void *hcpu) | |
4826 | { | |
4827 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 4828 | |
8bb78442 | 4829 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
4830 | drain_pages(cpu); |
4831 | ||
4832 | /* | |
4833 | * Spill the event counters of the dead processor | |
4834 | * into the current processors event counters. | |
4835 | * This artificially elevates the count of the current | |
4836 | * processor. | |
4837 | */ | |
f8891e5e | 4838 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
4839 | |
4840 | /* | |
4841 | * Zero the differential counters of the dead processor | |
4842 | * so that the vm statistics are consistent. | |
4843 | * | |
4844 | * This is only okay since the processor is dead and cannot | |
4845 | * race with what we are doing. | |
4846 | */ | |
2244b95a | 4847 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
4848 | } |
4849 | return NOTIFY_OK; | |
4850 | } | |
1da177e4 LT |
4851 | |
4852 | void __init page_alloc_init(void) | |
4853 | { | |
4854 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4855 | } | |
4856 | ||
cb45b0e9 HA |
4857 | /* |
4858 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4859 | * or min_free_kbytes changes. | |
4860 | */ | |
4861 | static void calculate_totalreserve_pages(void) | |
4862 | { | |
4863 | struct pglist_data *pgdat; | |
4864 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4865 | enum zone_type i, j; |
cb45b0e9 HA |
4866 | |
4867 | for_each_online_pgdat(pgdat) { | |
4868 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4869 | struct zone *zone = pgdat->node_zones + i; | |
4870 | unsigned long max = 0; | |
4871 | ||
4872 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4873 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4874 | if (zone->lowmem_reserve[j] > max) | |
4875 | max = zone->lowmem_reserve[j]; | |
4876 | } | |
4877 | ||
41858966 MG |
4878 | /* we treat the high watermark as reserved pages. */ |
4879 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
4880 | |
4881 | if (max > zone->present_pages) | |
4882 | max = zone->present_pages; | |
4883 | reserve_pages += max; | |
4884 | } | |
4885 | } | |
4886 | totalreserve_pages = reserve_pages; | |
4887 | } | |
4888 | ||
1da177e4 LT |
4889 | /* |
4890 | * setup_per_zone_lowmem_reserve - called whenever | |
4891 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4892 | * has a correct pages reserved value, so an adequate number of | |
4893 | * pages are left in the zone after a successful __alloc_pages(). | |
4894 | */ | |
4895 | static void setup_per_zone_lowmem_reserve(void) | |
4896 | { | |
4897 | struct pglist_data *pgdat; | |
2f6726e5 | 4898 | enum zone_type j, idx; |
1da177e4 | 4899 | |
ec936fc5 | 4900 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4901 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4902 | struct zone *zone = pgdat->node_zones + j; | |
4903 | unsigned long present_pages = zone->present_pages; | |
4904 | ||
4905 | zone->lowmem_reserve[j] = 0; | |
4906 | ||
2f6726e5 CL |
4907 | idx = j; |
4908 | while (idx) { | |
1da177e4 LT |
4909 | struct zone *lower_zone; |
4910 | ||
2f6726e5 CL |
4911 | idx--; |
4912 | ||
1da177e4 LT |
4913 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4914 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4915 | ||
4916 | lower_zone = pgdat->node_zones + idx; | |
4917 | lower_zone->lowmem_reserve[j] = present_pages / | |
4918 | sysctl_lowmem_reserve_ratio[idx]; | |
4919 | present_pages += lower_zone->present_pages; | |
4920 | } | |
4921 | } | |
4922 | } | |
cb45b0e9 HA |
4923 | |
4924 | /* update totalreserve_pages */ | |
4925 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4926 | } |
4927 | ||
88ca3b94 | 4928 | /** |
bc75d33f | 4929 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
bce7394a | 4930 | * or when memory is hot-{added|removed} |
88ca3b94 | 4931 | * |
bc75d33f MK |
4932 | * Ensures that the watermark[min,low,high] values for each zone are set |
4933 | * correctly with respect to min_free_kbytes. | |
1da177e4 | 4934 | */ |
bc75d33f | 4935 | void setup_per_zone_wmarks(void) |
1da177e4 LT |
4936 | { |
4937 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4938 | unsigned long lowmem_pages = 0; | |
4939 | struct zone *zone; | |
4940 | unsigned long flags; | |
4941 | ||
4942 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4943 | for_each_zone(zone) { | |
4944 | if (!is_highmem(zone)) | |
4945 | lowmem_pages += zone->present_pages; | |
4946 | } | |
4947 | ||
4948 | for_each_zone(zone) { | |
ac924c60 AM |
4949 | u64 tmp; |
4950 | ||
1125b4e3 | 4951 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4952 | tmp = (u64)pages_min * zone->present_pages; |
4953 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4954 | if (is_highmem(zone)) { |
4955 | /* | |
669ed175 NP |
4956 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4957 | * need highmem pages, so cap pages_min to a small | |
4958 | * value here. | |
4959 | * | |
41858966 | 4960 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
4961 | * deltas controls asynch page reclaim, and so should |
4962 | * not be capped for highmem. | |
1da177e4 LT |
4963 | */ |
4964 | int min_pages; | |
4965 | ||
4966 | min_pages = zone->present_pages / 1024; | |
4967 | if (min_pages < SWAP_CLUSTER_MAX) | |
4968 | min_pages = SWAP_CLUSTER_MAX; | |
4969 | if (min_pages > 128) | |
4970 | min_pages = 128; | |
41858966 | 4971 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 4972 | } else { |
669ed175 NP |
4973 | /* |
4974 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4975 | * proportionate to the zone's size. |
4976 | */ | |
41858966 | 4977 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
4978 | } |
4979 | ||
41858966 MG |
4980 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
4981 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
56fd56b8 | 4982 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4983 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4984 | } |
cb45b0e9 HA |
4985 | |
4986 | /* update totalreserve_pages */ | |
4987 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4988 | } |
4989 | ||
55a4462a | 4990 | /* |
556adecb RR |
4991 | * The inactive anon list should be small enough that the VM never has to |
4992 | * do too much work, but large enough that each inactive page has a chance | |
4993 | * to be referenced again before it is swapped out. | |
4994 | * | |
4995 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4996 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4997 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4998 | * the anonymous pages are kept on the inactive list. | |
4999 | * | |
5000 | * total target max | |
5001 | * memory ratio inactive anon | |
5002 | * ------------------------------------- | |
5003 | * 10MB 1 5MB | |
5004 | * 100MB 1 50MB | |
5005 | * 1GB 3 250MB | |
5006 | * 10GB 10 0.9GB | |
5007 | * 100GB 31 3GB | |
5008 | * 1TB 101 10GB | |
5009 | * 10TB 320 32GB | |
5010 | */ | |
96cb4df5 | 5011 | void calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5012 | { |
96cb4df5 | 5013 | unsigned int gb, ratio; |
556adecb | 5014 | |
96cb4df5 MK |
5015 | /* Zone size in gigabytes */ |
5016 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
5017 | if (gb) | |
556adecb | 5018 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5019 | else |
5020 | ratio = 1; | |
556adecb | 5021 | |
96cb4df5 MK |
5022 | zone->inactive_ratio = ratio; |
5023 | } | |
556adecb | 5024 | |
96cb4df5 MK |
5025 | static void __init setup_per_zone_inactive_ratio(void) |
5026 | { | |
5027 | struct zone *zone; | |
5028 | ||
5029 | for_each_zone(zone) | |
5030 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5031 | } |
5032 | ||
1da177e4 LT |
5033 | /* |
5034 | * Initialise min_free_kbytes. | |
5035 | * | |
5036 | * For small machines we want it small (128k min). For large machines | |
5037 | * we want it large (64MB max). But it is not linear, because network | |
5038 | * bandwidth does not increase linearly with machine size. We use | |
5039 | * | |
5040 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
5041 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
5042 | * | |
5043 | * which yields | |
5044 | * | |
5045 | * 16MB: 512k | |
5046 | * 32MB: 724k | |
5047 | * 64MB: 1024k | |
5048 | * 128MB: 1448k | |
5049 | * 256MB: 2048k | |
5050 | * 512MB: 2896k | |
5051 | * 1024MB: 4096k | |
5052 | * 2048MB: 5792k | |
5053 | * 4096MB: 8192k | |
5054 | * 8192MB: 11584k | |
5055 | * 16384MB: 16384k | |
5056 | */ | |
bc75d33f | 5057 | static int __init init_per_zone_wmark_min(void) |
1da177e4 LT |
5058 | { |
5059 | unsigned long lowmem_kbytes; | |
5060 | ||
5061 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5062 | ||
5063 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
5064 | if (min_free_kbytes < 128) | |
5065 | min_free_kbytes = 128; | |
5066 | if (min_free_kbytes > 65536) | |
5067 | min_free_kbytes = 65536; | |
bc75d33f | 5068 | setup_per_zone_wmarks(); |
1da177e4 | 5069 | setup_per_zone_lowmem_reserve(); |
556adecb | 5070 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5071 | return 0; |
5072 | } | |
bc75d33f | 5073 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5074 | |
5075 | /* | |
5076 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
5077 | * that we can call two helper functions whenever min_free_kbytes | |
5078 | * changes. | |
5079 | */ | |
5080 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5081 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5082 | { |
8d65af78 | 5083 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 5084 | if (write) |
bc75d33f | 5085 | setup_per_zone_wmarks(); |
1da177e4 LT |
5086 | return 0; |
5087 | } | |
5088 | ||
9614634f CL |
5089 | #ifdef CONFIG_NUMA |
5090 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5091 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5092 | { |
5093 | struct zone *zone; | |
5094 | int rc; | |
5095 | ||
8d65af78 | 5096 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5097 | if (rc) |
5098 | return rc; | |
5099 | ||
5100 | for_each_zone(zone) | |
8417bba4 | 5101 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
5102 | sysctl_min_unmapped_ratio) / 100; |
5103 | return 0; | |
5104 | } | |
0ff38490 CL |
5105 | |
5106 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5107 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5108 | { |
5109 | struct zone *zone; | |
5110 | int rc; | |
5111 | ||
8d65af78 | 5112 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5113 | if (rc) |
5114 | return rc; | |
5115 | ||
5116 | for_each_zone(zone) | |
5117 | zone->min_slab_pages = (zone->present_pages * | |
5118 | sysctl_min_slab_ratio) / 100; | |
5119 | return 0; | |
5120 | } | |
9614634f CL |
5121 | #endif |
5122 | ||
1da177e4 LT |
5123 | /* |
5124 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5125 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5126 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5127 | * | |
5128 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5129 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5130 | * if in function of the boot time zone sizes. |
5131 | */ | |
5132 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5133 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5134 | { |
8d65af78 | 5135 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5136 | setup_per_zone_lowmem_reserve(); |
5137 | return 0; | |
5138 | } | |
5139 | ||
8ad4b1fb RS |
5140 | /* |
5141 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
5142 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
5143 | * can have before it gets flushed back to buddy allocator. | |
5144 | */ | |
5145 | ||
5146 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5147 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5148 | { |
5149 | struct zone *zone; | |
5150 | unsigned int cpu; | |
5151 | int ret; | |
5152 | ||
8d65af78 | 5153 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
8ad4b1fb RS |
5154 | if (!write || (ret == -EINVAL)) |
5155 | return ret; | |
364df0eb | 5156 | for_each_populated_zone(zone) { |
99dcc3e5 | 5157 | for_each_possible_cpu(cpu) { |
8ad4b1fb RS |
5158 | unsigned long high; |
5159 | high = zone->present_pages / percpu_pagelist_fraction; | |
99dcc3e5 CL |
5160 | setup_pagelist_highmark( |
5161 | per_cpu_ptr(zone->pageset, cpu), high); | |
8ad4b1fb RS |
5162 | } |
5163 | } | |
5164 | return 0; | |
5165 | } | |
5166 | ||
f034b5d4 | 5167 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5168 | |
5169 | #ifdef CONFIG_NUMA | |
5170 | static int __init set_hashdist(char *str) | |
5171 | { | |
5172 | if (!str) | |
5173 | return 0; | |
5174 | hashdist = simple_strtoul(str, &str, 0); | |
5175 | return 1; | |
5176 | } | |
5177 | __setup("hashdist=", set_hashdist); | |
5178 | #endif | |
5179 | ||
5180 | /* | |
5181 | * allocate a large system hash table from bootmem | |
5182 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5183 | * quantity of entries | |
5184 | * - limit is the number of hash buckets, not the total allocation size | |
5185 | */ | |
5186 | void *__init alloc_large_system_hash(const char *tablename, | |
5187 | unsigned long bucketsize, | |
5188 | unsigned long numentries, | |
5189 | int scale, | |
5190 | int flags, | |
5191 | unsigned int *_hash_shift, | |
5192 | unsigned int *_hash_mask, | |
5193 | unsigned long limit) | |
5194 | { | |
5195 | unsigned long long max = limit; | |
5196 | unsigned long log2qty, size; | |
5197 | void *table = NULL; | |
5198 | ||
5199 | /* allow the kernel cmdline to have a say */ | |
5200 | if (!numentries) { | |
5201 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5202 | numentries = nr_kernel_pages; |
1da177e4 LT |
5203 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
5204 | numentries >>= 20 - PAGE_SHIFT; | |
5205 | numentries <<= 20 - PAGE_SHIFT; | |
5206 | ||
5207 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5208 | if (scale > PAGE_SHIFT) | |
5209 | numentries >>= (scale - PAGE_SHIFT); | |
5210 | else | |
5211 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5212 | |
5213 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5214 | if (unlikely(flags & HASH_SMALL)) { |
5215 | /* Makes no sense without HASH_EARLY */ | |
5216 | WARN_ON(!(flags & HASH_EARLY)); | |
5217 | if (!(numentries >> *_hash_shift)) { | |
5218 | numentries = 1UL << *_hash_shift; | |
5219 | BUG_ON(!numentries); | |
5220 | } | |
5221 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5222 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5223 | } |
6e692ed3 | 5224 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5225 | |
5226 | /* limit allocation size to 1/16 total memory by default */ | |
5227 | if (max == 0) { | |
5228 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5229 | do_div(max, bucketsize); | |
5230 | } | |
5231 | ||
5232 | if (numentries > max) | |
5233 | numentries = max; | |
5234 | ||
f0d1b0b3 | 5235 | log2qty = ilog2(numentries); |
1da177e4 LT |
5236 | |
5237 | do { | |
5238 | size = bucketsize << log2qty; | |
5239 | if (flags & HASH_EARLY) | |
74768ed8 | 5240 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5241 | else if (hashdist) |
5242 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5243 | else { | |
1037b83b ED |
5244 | /* |
5245 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5246 | * some pages at the end of hash table which |
5247 | * alloc_pages_exact() automatically does | |
1037b83b | 5248 | */ |
264ef8a9 | 5249 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5250 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5251 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5252 | } | |
1da177e4 LT |
5253 | } |
5254 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5255 | ||
5256 | if (!table) | |
5257 | panic("Failed to allocate %s hash table\n", tablename); | |
5258 | ||
f241e660 | 5259 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5260 | tablename, |
f241e660 | 5261 | (1UL << log2qty), |
f0d1b0b3 | 5262 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5263 | size); |
5264 | ||
5265 | if (_hash_shift) | |
5266 | *_hash_shift = log2qty; | |
5267 | if (_hash_mask) | |
5268 | *_hash_mask = (1 << log2qty) - 1; | |
5269 | ||
5270 | return table; | |
5271 | } | |
a117e66e | 5272 | |
835c134e MG |
5273 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5274 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5275 | unsigned long pfn) | |
5276 | { | |
5277 | #ifdef CONFIG_SPARSEMEM | |
5278 | return __pfn_to_section(pfn)->pageblock_flags; | |
5279 | #else | |
5280 | return zone->pageblock_flags; | |
5281 | #endif /* CONFIG_SPARSEMEM */ | |
5282 | } | |
5283 | ||
5284 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5285 | { | |
5286 | #ifdef CONFIG_SPARSEMEM | |
5287 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5288 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5289 | #else |
5290 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 5291 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5292 | #endif /* CONFIG_SPARSEMEM */ |
5293 | } | |
5294 | ||
5295 | /** | |
d9c23400 | 5296 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5297 | * @page: The page within the block of interest |
5298 | * @start_bitidx: The first bit of interest to retrieve | |
5299 | * @end_bitidx: The last bit of interest | |
5300 | * returns pageblock_bits flags | |
5301 | */ | |
5302 | unsigned long get_pageblock_flags_group(struct page *page, | |
5303 | int start_bitidx, int end_bitidx) | |
5304 | { | |
5305 | struct zone *zone; | |
5306 | unsigned long *bitmap; | |
5307 | unsigned long pfn, bitidx; | |
5308 | unsigned long flags = 0; | |
5309 | unsigned long value = 1; | |
5310 | ||
5311 | zone = page_zone(page); | |
5312 | pfn = page_to_pfn(page); | |
5313 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5314 | bitidx = pfn_to_bitidx(zone, pfn); | |
5315 | ||
5316 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5317 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5318 | flags |= value; | |
6220ec78 | 5319 | |
835c134e MG |
5320 | return flags; |
5321 | } | |
5322 | ||
5323 | /** | |
d9c23400 | 5324 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5325 | * @page: The page within the block of interest |
5326 | * @start_bitidx: The first bit of interest | |
5327 | * @end_bitidx: The last bit of interest | |
5328 | * @flags: The flags to set | |
5329 | */ | |
5330 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5331 | int start_bitidx, int end_bitidx) | |
5332 | { | |
5333 | struct zone *zone; | |
5334 | unsigned long *bitmap; | |
5335 | unsigned long pfn, bitidx; | |
5336 | unsigned long value = 1; | |
5337 | ||
5338 | zone = page_zone(page); | |
5339 | pfn = page_to_pfn(page); | |
5340 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5341 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
5342 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
5343 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
5344 | |
5345 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5346 | if (flags & value) | |
5347 | __set_bit(bitidx + start_bitidx, bitmap); | |
5348 | else | |
5349 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5350 | } | |
a5d76b54 KH |
5351 | |
5352 | /* | |
5353 | * This is designed as sub function...plz see page_isolation.c also. | |
5354 | * set/clear page block's type to be ISOLATE. | |
5355 | * page allocater never alloc memory from ISOLATE block. | |
5356 | */ | |
5357 | ||
49ac8255 KH |
5358 | static int |
5359 | __count_immobile_pages(struct zone *zone, struct page *page, int count) | |
5360 | { | |
5361 | unsigned long pfn, iter, found; | |
5362 | /* | |
5363 | * For avoiding noise data, lru_add_drain_all() should be called | |
5364 | * If ZONE_MOVABLE, the zone never contains immobile pages | |
5365 | */ | |
5366 | if (zone_idx(zone) == ZONE_MOVABLE) | |
5367 | return true; | |
5368 | ||
5369 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE) | |
5370 | return true; | |
5371 | ||
5372 | pfn = page_to_pfn(page); | |
5373 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5374 | unsigned long check = pfn + iter; | |
5375 | ||
5376 | if (!pfn_valid_within(check)) { | |
5377 | iter++; | |
5378 | continue; | |
5379 | } | |
5380 | page = pfn_to_page(check); | |
5381 | if (!page_count(page)) { | |
5382 | if (PageBuddy(page)) | |
5383 | iter += (1 << page_order(page)) - 1; | |
5384 | continue; | |
5385 | } | |
5386 | if (!PageLRU(page)) | |
5387 | found++; | |
5388 | /* | |
5389 | * If there are RECLAIMABLE pages, we need to check it. | |
5390 | * But now, memory offline itself doesn't call shrink_slab() | |
5391 | * and it still to be fixed. | |
5392 | */ | |
5393 | /* | |
5394 | * If the page is not RAM, page_count()should be 0. | |
5395 | * we don't need more check. This is an _used_ not-movable page. | |
5396 | * | |
5397 | * The problematic thing here is PG_reserved pages. PG_reserved | |
5398 | * is set to both of a memory hole page and a _used_ kernel | |
5399 | * page at boot. | |
5400 | */ | |
5401 | if (found > count) | |
5402 | return false; | |
5403 | } | |
5404 | return true; | |
5405 | } | |
5406 | ||
5407 | bool is_pageblock_removable_nolock(struct page *page) | |
5408 | { | |
5409 | struct zone *zone = page_zone(page); | |
5410 | return __count_immobile_pages(zone, page, 0); | |
5411 | } | |
5412 | ||
a5d76b54 KH |
5413 | int set_migratetype_isolate(struct page *page) |
5414 | { | |
5415 | struct zone *zone; | |
49ac8255 | 5416 | unsigned long flags, pfn; |
925cc71e RJ |
5417 | struct memory_isolate_notify arg; |
5418 | int notifier_ret; | |
a5d76b54 | 5419 | int ret = -EBUSY; |
8e7e40d9 | 5420 | int zone_idx; |
a5d76b54 KH |
5421 | |
5422 | zone = page_zone(page); | |
8e7e40d9 | 5423 | zone_idx = zone_idx(zone); |
925cc71e | 5424 | |
a5d76b54 | 5425 | spin_lock_irqsave(&zone->lock, flags); |
925cc71e RJ |
5426 | |
5427 | pfn = page_to_pfn(page); | |
5428 | arg.start_pfn = pfn; | |
5429 | arg.nr_pages = pageblock_nr_pages; | |
5430 | arg.pages_found = 0; | |
5431 | ||
a5d76b54 | 5432 | /* |
925cc71e RJ |
5433 | * It may be possible to isolate a pageblock even if the |
5434 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
5435 | * notifier chain is used by balloon drivers to return the | |
5436 | * number of pages in a range that are held by the balloon | |
5437 | * driver to shrink memory. If all the pages are accounted for | |
5438 | * by balloons, are free, or on the LRU, isolation can continue. | |
5439 | * Later, for example, when memory hotplug notifier runs, these | |
5440 | * pages reported as "can be isolated" should be isolated(freed) | |
5441 | * by the balloon driver through the memory notifier chain. | |
a5d76b54 | 5442 | */ |
925cc71e RJ |
5443 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); |
5444 | notifier_ret = notifier_to_errno(notifier_ret); | |
4b20477f | 5445 | if (notifier_ret) |
a5d76b54 | 5446 | goto out; |
49ac8255 KH |
5447 | /* |
5448 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
5449 | * We just check MOVABLE pages. | |
5450 | */ | |
5451 | if (__count_immobile_pages(zone, page, arg.pages_found)) | |
925cc71e RJ |
5452 | ret = 0; |
5453 | ||
49ac8255 KH |
5454 | /* |
5455 | * immobile means "not-on-lru" paes. If immobile is larger than | |
5456 | * removable-by-driver pages reported by notifier, we'll fail. | |
5457 | */ | |
5458 | ||
a5d76b54 | 5459 | out: |
925cc71e RJ |
5460 | if (!ret) { |
5461 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
5462 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
5463 | } | |
5464 | ||
a5d76b54 KH |
5465 | spin_unlock_irqrestore(&zone->lock, flags); |
5466 | if (!ret) | |
9f8f2172 | 5467 | drain_all_pages(); |
a5d76b54 KH |
5468 | return ret; |
5469 | } | |
5470 | ||
5471 | void unset_migratetype_isolate(struct page *page) | |
5472 | { | |
5473 | struct zone *zone; | |
5474 | unsigned long flags; | |
5475 | zone = page_zone(page); | |
5476 | spin_lock_irqsave(&zone->lock, flags); | |
5477 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
5478 | goto out; | |
5479 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5480 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
5481 | out: | |
5482 | spin_unlock_irqrestore(&zone->lock, flags); | |
5483 | } | |
0c0e6195 KH |
5484 | |
5485 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5486 | /* | |
5487 | * All pages in the range must be isolated before calling this. | |
5488 | */ | |
5489 | void | |
5490 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5491 | { | |
5492 | struct page *page; | |
5493 | struct zone *zone; | |
5494 | int order, i; | |
5495 | unsigned long pfn; | |
5496 | unsigned long flags; | |
5497 | /* find the first valid pfn */ | |
5498 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5499 | if (pfn_valid(pfn)) | |
5500 | break; | |
5501 | if (pfn == end_pfn) | |
5502 | return; | |
5503 | zone = page_zone(pfn_to_page(pfn)); | |
5504 | spin_lock_irqsave(&zone->lock, flags); | |
5505 | pfn = start_pfn; | |
5506 | while (pfn < end_pfn) { | |
5507 | if (!pfn_valid(pfn)) { | |
5508 | pfn++; | |
5509 | continue; | |
5510 | } | |
5511 | page = pfn_to_page(pfn); | |
5512 | BUG_ON(page_count(page)); | |
5513 | BUG_ON(!PageBuddy(page)); | |
5514 | order = page_order(page); | |
5515 | #ifdef CONFIG_DEBUG_VM | |
5516 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5517 | pfn, 1 << order, end_pfn); | |
5518 | #endif | |
5519 | list_del(&page->lru); | |
5520 | rmv_page_order(page); | |
5521 | zone->free_area[order].nr_free--; | |
5522 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
5523 | - (1UL << order)); | |
5524 | for (i = 0; i < (1 << order); i++) | |
5525 | SetPageReserved((page+i)); | |
5526 | pfn += (1 << order); | |
5527 | } | |
5528 | spin_unlock_irqrestore(&zone->lock, flags); | |
5529 | } | |
5530 | #endif | |
8d22ba1b WF |
5531 | |
5532 | #ifdef CONFIG_MEMORY_FAILURE | |
5533 | bool is_free_buddy_page(struct page *page) | |
5534 | { | |
5535 | struct zone *zone = page_zone(page); | |
5536 | unsigned long pfn = page_to_pfn(page); | |
5537 | unsigned long flags; | |
5538 | int order; | |
5539 | ||
5540 | spin_lock_irqsave(&zone->lock, flags); | |
5541 | for (order = 0; order < MAX_ORDER; order++) { | |
5542 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
5543 | ||
5544 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
5545 | break; | |
5546 | } | |
5547 | spin_unlock_irqrestore(&zone->lock, flags); | |
5548 | ||
5549 | return order < MAX_ORDER; | |
5550 | } | |
5551 | #endif | |
718a3821 WF |
5552 | |
5553 | static struct trace_print_flags pageflag_names[] = { | |
5554 | {1UL << PG_locked, "locked" }, | |
5555 | {1UL << PG_error, "error" }, | |
5556 | {1UL << PG_referenced, "referenced" }, | |
5557 | {1UL << PG_uptodate, "uptodate" }, | |
5558 | {1UL << PG_dirty, "dirty" }, | |
5559 | {1UL << PG_lru, "lru" }, | |
5560 | {1UL << PG_active, "active" }, | |
5561 | {1UL << PG_slab, "slab" }, | |
5562 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
5563 | {1UL << PG_arch_1, "arch_1" }, | |
5564 | {1UL << PG_reserved, "reserved" }, | |
5565 | {1UL << PG_private, "private" }, | |
5566 | {1UL << PG_private_2, "private_2" }, | |
5567 | {1UL << PG_writeback, "writeback" }, | |
5568 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
5569 | {1UL << PG_head, "head" }, | |
5570 | {1UL << PG_tail, "tail" }, | |
5571 | #else | |
5572 | {1UL << PG_compound, "compound" }, | |
5573 | #endif | |
5574 | {1UL << PG_swapcache, "swapcache" }, | |
5575 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
5576 | {1UL << PG_reclaim, "reclaim" }, | |
5577 | {1UL << PG_buddy, "buddy" }, | |
5578 | {1UL << PG_swapbacked, "swapbacked" }, | |
5579 | {1UL << PG_unevictable, "unevictable" }, | |
5580 | #ifdef CONFIG_MMU | |
5581 | {1UL << PG_mlocked, "mlocked" }, | |
5582 | #endif | |
5583 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
5584 | {1UL << PG_uncached, "uncached" }, | |
5585 | #endif | |
5586 | #ifdef CONFIG_MEMORY_FAILURE | |
5587 | {1UL << PG_hwpoison, "hwpoison" }, | |
5588 | #endif | |
5589 | {-1UL, NULL }, | |
5590 | }; | |
5591 | ||
5592 | static void dump_page_flags(unsigned long flags) | |
5593 | { | |
5594 | const char *delim = ""; | |
5595 | unsigned long mask; | |
5596 | int i; | |
5597 | ||
5598 | printk(KERN_ALERT "page flags: %#lx(", flags); | |
5599 | ||
5600 | /* remove zone id */ | |
5601 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
5602 | ||
5603 | for (i = 0; pageflag_names[i].name && flags; i++) { | |
5604 | ||
5605 | mask = pageflag_names[i].mask; | |
5606 | if ((flags & mask) != mask) | |
5607 | continue; | |
5608 | ||
5609 | flags &= ~mask; | |
5610 | printk("%s%s", delim, pageflag_names[i].name); | |
5611 | delim = "|"; | |
5612 | } | |
5613 | ||
5614 | /* check for left over flags */ | |
5615 | if (flags) | |
5616 | printk("%s%#lx", delim, flags); | |
5617 | ||
5618 | printk(")\n"); | |
5619 | } | |
5620 | ||
5621 | void dump_page(struct page *page) | |
5622 | { | |
5623 | printk(KERN_ALERT | |
5624 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 5625 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
5626 | page->mapping, page->index); |
5627 | dump_page_flags(page->flags); | |
5628 | } |