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