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> | |
a238ab5b | 33 | #include <linux/ratelimit.h> |
5a3135c2 | 34 | #include <linux/oom.h> |
1da177e4 LT |
35 | #include <linux/notifier.h> |
36 | #include <linux/topology.h> | |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
6811378e | 45 | #include <linux/stop_machine.h> |
c713216d MG |
46 | #include <linux/sort.h> |
47 | #include <linux/pfn.h> | |
3fcfab16 | 48 | #include <linux/backing-dev.h> |
933e312e | 49 | #include <linux/fault-inject.h> |
a5d76b54 | 50 | #include <linux/page-isolation.h> |
52d4b9ac | 51 | #include <linux/page_cgroup.h> |
3ac7fe5a | 52 | #include <linux/debugobjects.h> |
dbb1f81c | 53 | #include <linux/kmemleak.h> |
925cc71e | 54 | #include <linux/memory.h> |
56de7263 | 55 | #include <linux/compaction.h> |
0d3d062a | 56 | #include <trace/events/kmem.h> |
718a3821 | 57 | #include <linux/ftrace_event.h> |
f212ad7c | 58 | #include <linux/memcontrol.h> |
268bb0ce | 59 | #include <linux/prefetch.h> |
c0a32fc5 | 60 | #include <linux/page-debug-flags.h> |
1da177e4 LT |
61 | |
62 | #include <asm/tlbflush.h> | |
ac924c60 | 63 | #include <asm/div64.h> |
1da177e4 LT |
64 | #include "internal.h" |
65 | ||
72812019 LS |
66 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
67 | DEFINE_PER_CPU(int, numa_node); | |
68 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
69 | #endif | |
70 | ||
7aac7898 LS |
71 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
72 | /* | |
73 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
74 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
75 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
76 | * defined in <linux/topology.h>. | |
77 | */ | |
78 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
79 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
80 | #endif | |
81 | ||
1da177e4 | 82 | /* |
13808910 | 83 | * Array of node states. |
1da177e4 | 84 | */ |
13808910 CL |
85 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
86 | [N_POSSIBLE] = NODE_MASK_ALL, | |
87 | [N_ONLINE] = { { [0] = 1UL } }, | |
88 | #ifndef CONFIG_NUMA | |
89 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
90 | #ifdef CONFIG_HIGHMEM | |
91 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
92 | #endif | |
93 | [N_CPU] = { { [0] = 1UL } }, | |
94 | #endif /* NUMA */ | |
95 | }; | |
96 | EXPORT_SYMBOL(node_states); | |
97 | ||
6c231b7b | 98 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 99 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
100 | /* |
101 | * When calculating the number of globally allowed dirty pages, there | |
102 | * is a certain number of per-zone reserves that should not be | |
103 | * considered dirtyable memory. This is the sum of those reserves | |
104 | * over all existing zones that contribute dirtyable memory. | |
105 | */ | |
106 | unsigned long dirty_balance_reserve __read_mostly; | |
107 | ||
8ad4b1fb | 108 | int percpu_pagelist_fraction; |
dcce284a | 109 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 110 | |
452aa699 RW |
111 | #ifdef CONFIG_PM_SLEEP |
112 | /* | |
113 | * The following functions are used by the suspend/hibernate code to temporarily | |
114 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
115 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
116 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
117 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
118 | * guaranteed not to run in parallel with that modification). | |
119 | */ | |
c9e664f1 RW |
120 | |
121 | static gfp_t saved_gfp_mask; | |
122 | ||
123 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
124 | { |
125 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
126 | if (saved_gfp_mask) { |
127 | gfp_allowed_mask = saved_gfp_mask; | |
128 | saved_gfp_mask = 0; | |
129 | } | |
452aa699 RW |
130 | } |
131 | ||
c9e664f1 | 132 | void pm_restrict_gfp_mask(void) |
452aa699 | 133 | { |
452aa699 | 134 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
135 | WARN_ON(saved_gfp_mask); |
136 | saved_gfp_mask = gfp_allowed_mask; | |
137 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 138 | } |
f90ac398 MG |
139 | |
140 | bool pm_suspended_storage(void) | |
141 | { | |
142 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
143 | return false; | |
144 | return true; | |
145 | } | |
452aa699 RW |
146 | #endif /* CONFIG_PM_SLEEP */ |
147 | ||
d9c23400 MG |
148 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
149 | int pageblock_order __read_mostly; | |
150 | #endif | |
151 | ||
d98c7a09 | 152 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 153 | |
1da177e4 LT |
154 | /* |
155 | * results with 256, 32 in the lowmem_reserve sysctl: | |
156 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
157 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
158 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
159 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
160 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
161 | * |
162 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
163 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 164 | */ |
2f1b6248 | 165 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 166 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 167 | 256, |
4b51d669 | 168 | #endif |
fb0e7942 | 169 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 170 | 256, |
fb0e7942 | 171 | #endif |
e53ef38d | 172 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 173 | 32, |
e53ef38d | 174 | #endif |
2a1e274a | 175 | 32, |
2f1b6248 | 176 | }; |
1da177e4 LT |
177 | |
178 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 179 | |
15ad7cdc | 180 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 181 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 182 | "DMA", |
4b51d669 | 183 | #endif |
fb0e7942 | 184 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 185 | "DMA32", |
fb0e7942 | 186 | #endif |
2f1b6248 | 187 | "Normal", |
e53ef38d | 188 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 189 | "HighMem", |
e53ef38d | 190 | #endif |
2a1e274a | 191 | "Movable", |
2f1b6248 CL |
192 | }; |
193 | ||
1da177e4 LT |
194 | int min_free_kbytes = 1024; |
195 | ||
2c85f51d JB |
196 | static unsigned long __meminitdata nr_kernel_pages; |
197 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 198 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 199 | |
0ee332c1 TH |
200 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
201 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
202 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
203 | static unsigned long __initdata required_kernelcore; | |
204 | static unsigned long __initdata required_movablecore; | |
205 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
206 | ||
207 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
208 | int movable_zone; | |
209 | EXPORT_SYMBOL(movable_zone); | |
210 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 211 | |
418508c1 MS |
212 | #if MAX_NUMNODES > 1 |
213 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 214 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 215 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 216 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
217 | #endif |
218 | ||
9ef9acb0 MG |
219 | int page_group_by_mobility_disabled __read_mostly; |
220 | ||
b2a0ac88 MG |
221 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
222 | { | |
49255c61 MG |
223 | |
224 | if (unlikely(page_group_by_mobility_disabled)) | |
225 | migratetype = MIGRATE_UNMOVABLE; | |
226 | ||
b2a0ac88 MG |
227 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
228 | PB_migrate, PB_migrate_end); | |
229 | } | |
230 | ||
7f33d49a RW |
231 | bool oom_killer_disabled __read_mostly; |
232 | ||
13e7444b | 233 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 234 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 235 | { |
bdc8cb98 DH |
236 | int ret = 0; |
237 | unsigned seq; | |
238 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 239 | |
bdc8cb98 DH |
240 | do { |
241 | seq = zone_span_seqbegin(zone); | |
242 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
243 | ret = 1; | |
244 | else if (pfn < zone->zone_start_pfn) | |
245 | ret = 1; | |
246 | } while (zone_span_seqretry(zone, seq)); | |
247 | ||
248 | return ret; | |
c6a57e19 DH |
249 | } |
250 | ||
251 | static int page_is_consistent(struct zone *zone, struct page *page) | |
252 | { | |
14e07298 | 253 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 254 | return 0; |
1da177e4 | 255 | if (zone != page_zone(page)) |
c6a57e19 DH |
256 | return 0; |
257 | ||
258 | return 1; | |
259 | } | |
260 | /* | |
261 | * Temporary debugging check for pages not lying within a given zone. | |
262 | */ | |
263 | static int bad_range(struct zone *zone, struct page *page) | |
264 | { | |
265 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 266 | return 1; |
c6a57e19 DH |
267 | if (!page_is_consistent(zone, page)) |
268 | return 1; | |
269 | ||
1da177e4 LT |
270 | return 0; |
271 | } | |
13e7444b NP |
272 | #else |
273 | static inline int bad_range(struct zone *zone, struct page *page) | |
274 | { | |
275 | return 0; | |
276 | } | |
277 | #endif | |
278 | ||
224abf92 | 279 | static void bad_page(struct page *page) |
1da177e4 | 280 | { |
d936cf9b HD |
281 | static unsigned long resume; |
282 | static unsigned long nr_shown; | |
283 | static unsigned long nr_unshown; | |
284 | ||
2a7684a2 WF |
285 | /* Don't complain about poisoned pages */ |
286 | if (PageHWPoison(page)) { | |
ef2b4b95 | 287 | reset_page_mapcount(page); /* remove PageBuddy */ |
2a7684a2 WF |
288 | return; |
289 | } | |
290 | ||
d936cf9b HD |
291 | /* |
292 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
293 | * or allow a steady drip of one report per second. | |
294 | */ | |
295 | if (nr_shown == 60) { | |
296 | if (time_before(jiffies, resume)) { | |
297 | nr_unshown++; | |
298 | goto out; | |
299 | } | |
300 | if (nr_unshown) { | |
1e9e6365 HD |
301 | printk(KERN_ALERT |
302 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
303 | nr_unshown); |
304 | nr_unshown = 0; | |
305 | } | |
306 | nr_shown = 0; | |
307 | } | |
308 | if (nr_shown++ == 0) | |
309 | resume = jiffies + 60 * HZ; | |
310 | ||
1e9e6365 | 311 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 312 | current->comm, page_to_pfn(page)); |
718a3821 | 313 | dump_page(page); |
3dc14741 | 314 | |
4f31888c | 315 | print_modules(); |
1da177e4 | 316 | dump_stack(); |
d936cf9b | 317 | out: |
8cc3b392 | 318 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
ef2b4b95 | 319 | reset_page_mapcount(page); /* remove PageBuddy */ |
9f158333 | 320 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
321 | } |
322 | ||
1da177e4 LT |
323 | /* |
324 | * Higher-order pages are called "compound pages". They are structured thusly: | |
325 | * | |
326 | * The first PAGE_SIZE page is called the "head page". | |
327 | * | |
328 | * The remaining PAGE_SIZE pages are called "tail pages". | |
329 | * | |
6416b9fa WSH |
330 | * All pages have PG_compound set. All tail pages have their ->first_page |
331 | * pointing at the head page. | |
1da177e4 | 332 | * |
41d78ba5 HD |
333 | * The first tail page's ->lru.next holds the address of the compound page's |
334 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
335 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 336 | */ |
d98c7a09 HD |
337 | |
338 | static void free_compound_page(struct page *page) | |
339 | { | |
d85f3385 | 340 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
341 | } |
342 | ||
01ad1c08 | 343 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
344 | { |
345 | int i; | |
346 | int nr_pages = 1 << order; | |
347 | ||
348 | set_compound_page_dtor(page, free_compound_page); | |
349 | set_compound_order(page, order); | |
350 | __SetPageHead(page); | |
351 | for (i = 1; i < nr_pages; i++) { | |
352 | struct page *p = page + i; | |
18229df5 | 353 | __SetPageTail(p); |
58a84aa9 | 354 | set_page_count(p, 0); |
18229df5 AW |
355 | p->first_page = page; |
356 | } | |
357 | } | |
358 | ||
59ff4216 | 359 | /* update __split_huge_page_refcount if you change this function */ |
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 | ||
c0a32fc5 SG |
400 | #ifdef CONFIG_DEBUG_PAGEALLOC |
401 | unsigned int _debug_guardpage_minorder; | |
402 | ||
403 | static int __init debug_guardpage_minorder_setup(char *buf) | |
404 | { | |
405 | unsigned long res; | |
406 | ||
407 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
408 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
409 | return 0; | |
410 | } | |
411 | _debug_guardpage_minorder = res; | |
412 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
413 | return 0; | |
414 | } | |
415 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
416 | ||
417 | static inline void set_page_guard_flag(struct page *page) | |
418 | { | |
419 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
420 | } | |
421 | ||
422 | static inline void clear_page_guard_flag(struct page *page) | |
423 | { | |
424 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
425 | } | |
426 | #else | |
427 | static inline void set_page_guard_flag(struct page *page) { } | |
428 | static inline void clear_page_guard_flag(struct page *page) { } | |
429 | #endif | |
430 | ||
6aa3001b AM |
431 | static inline void set_page_order(struct page *page, int order) |
432 | { | |
4c21e2f2 | 433 | set_page_private(page, order); |
676165a8 | 434 | __SetPageBuddy(page); |
1da177e4 LT |
435 | } |
436 | ||
437 | static inline void rmv_page_order(struct page *page) | |
438 | { | |
676165a8 | 439 | __ClearPageBuddy(page); |
4c21e2f2 | 440 | set_page_private(page, 0); |
1da177e4 LT |
441 | } |
442 | ||
443 | /* | |
444 | * Locate the struct page for both the matching buddy in our | |
445 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
446 | * | |
447 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
448 | * the following equation: | |
449 | * B2 = B1 ^ (1 << O) | |
450 | * For example, if the starting buddy (buddy2) is #8 its order | |
451 | * 1 buddy is #10: | |
452 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
453 | * | |
454 | * 2) Any buddy B will have an order O+1 parent P which | |
455 | * satisfies the following equation: | |
456 | * P = B & ~(1 << O) | |
457 | * | |
d6e05edc | 458 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 459 | */ |
1da177e4 | 460 | static inline unsigned long |
43506fad | 461 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 462 | { |
43506fad | 463 | return page_idx ^ (1 << order); |
1da177e4 LT |
464 | } |
465 | ||
466 | /* | |
467 | * This function checks whether a page is free && is the buddy | |
468 | * we can do coalesce a page and its buddy if | |
13e7444b | 469 | * (a) the buddy is not in a hole && |
676165a8 | 470 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
471 | * (c) a page and its buddy have the same order && |
472 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 473 | * |
5f24ce5f AA |
474 | * For recording whether a page is in the buddy system, we set ->_mapcount -2. |
475 | * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock. | |
1da177e4 | 476 | * |
676165a8 | 477 | * For recording page's order, we use page_private(page). |
1da177e4 | 478 | */ |
cb2b95e1 AW |
479 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
480 | int order) | |
1da177e4 | 481 | { |
14e07298 | 482 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 483 | return 0; |
13e7444b | 484 | |
cb2b95e1 AW |
485 | if (page_zone_id(page) != page_zone_id(buddy)) |
486 | return 0; | |
487 | ||
c0a32fc5 SG |
488 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
489 | VM_BUG_ON(page_count(buddy) != 0); | |
490 | return 1; | |
491 | } | |
492 | ||
cb2b95e1 | 493 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
a3af9c38 | 494 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 495 | return 1; |
676165a8 | 496 | } |
6aa3001b | 497 | return 0; |
1da177e4 LT |
498 | } |
499 | ||
500 | /* | |
501 | * Freeing function for a buddy system allocator. | |
502 | * | |
503 | * The concept of a buddy system is to maintain direct-mapped table | |
504 | * (containing bit values) for memory blocks of various "orders". | |
505 | * The bottom level table contains the map for the smallest allocatable | |
506 | * units of memory (here, pages), and each level above it describes | |
507 | * pairs of units from the levels below, hence, "buddies". | |
508 | * At a high level, all that happens here is marking the table entry | |
509 | * at the bottom level available, and propagating the changes upward | |
510 | * as necessary, plus some accounting needed to play nicely with other | |
511 | * parts of the VM system. | |
512 | * At each level, we keep a list of pages, which are heads of continuous | |
5f24ce5f | 513 | * free pages of length of (1 << order) and marked with _mapcount -2. Page's |
4c21e2f2 | 514 | * order is recorded in page_private(page) field. |
1da177e4 LT |
515 | * So when we are allocating or freeing one, we can derive the state of the |
516 | * other. That is, if we allocate a small block, and both were | |
517 | * free, the remainder of the region must be split into blocks. | |
518 | * If a block is freed, and its buddy is also free, then this | |
519 | * triggers coalescing into a block of larger size. | |
520 | * | |
521 | * -- wli | |
522 | */ | |
523 | ||
48db57f8 | 524 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
525 | struct zone *zone, unsigned int order, |
526 | int migratetype) | |
1da177e4 LT |
527 | { |
528 | unsigned long page_idx; | |
6dda9d55 | 529 | unsigned long combined_idx; |
43506fad | 530 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 531 | struct page *buddy; |
1da177e4 | 532 | |
224abf92 | 533 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
534 | if (unlikely(destroy_compound_page(page, order))) |
535 | return; | |
1da177e4 | 536 | |
ed0ae21d MG |
537 | VM_BUG_ON(migratetype == -1); |
538 | ||
1da177e4 LT |
539 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
540 | ||
f2260e6b | 541 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 542 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 543 | |
1da177e4 | 544 | while (order < MAX_ORDER-1) { |
43506fad KC |
545 | buddy_idx = __find_buddy_index(page_idx, order); |
546 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 547 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 548 | break; |
c0a32fc5 SG |
549 | /* |
550 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
551 | * merge with it and move up one order. | |
552 | */ | |
553 | if (page_is_guard(buddy)) { | |
554 | clear_page_guard_flag(buddy); | |
555 | set_page_private(page, 0); | |
556 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); | |
557 | } else { | |
558 | list_del(&buddy->lru); | |
559 | zone->free_area[order].nr_free--; | |
560 | rmv_page_order(buddy); | |
561 | } | |
43506fad | 562 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
563 | page = page + (combined_idx - page_idx); |
564 | page_idx = combined_idx; | |
565 | order++; | |
566 | } | |
567 | set_page_order(page, order); | |
6dda9d55 CZ |
568 | |
569 | /* | |
570 | * If this is not the largest possible page, check if the buddy | |
571 | * of the next-highest order is free. If it is, it's possible | |
572 | * that pages are being freed that will coalesce soon. In case, | |
573 | * that is happening, add the free page to the tail of the list | |
574 | * so it's less likely to be used soon and more likely to be merged | |
575 | * as a higher order page | |
576 | */ | |
b7f50cfa | 577 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 578 | struct page *higher_page, *higher_buddy; |
43506fad KC |
579 | combined_idx = buddy_idx & page_idx; |
580 | higher_page = page + (combined_idx - page_idx); | |
581 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
582 | higher_buddy = page + (buddy_idx - combined_idx); | |
6dda9d55 CZ |
583 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
584 | list_add_tail(&page->lru, | |
585 | &zone->free_area[order].free_list[migratetype]); | |
586 | goto out; | |
587 | } | |
588 | } | |
589 | ||
590 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
591 | out: | |
1da177e4 LT |
592 | zone->free_area[order].nr_free++; |
593 | } | |
594 | ||
092cead6 KM |
595 | /* |
596 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
597 | * Page should not be on lru, so no need to fix that up. | |
598 | * free_pages_check() will verify... | |
599 | */ | |
600 | static inline void free_page_mlock(struct page *page) | |
601 | { | |
092cead6 KM |
602 | __dec_zone_page_state(page, NR_MLOCK); |
603 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
604 | } | |
092cead6 | 605 | |
224abf92 | 606 | static inline int free_pages_check(struct page *page) |
1da177e4 | 607 | { |
92be2e33 NP |
608 | if (unlikely(page_mapcount(page) | |
609 | (page->mapping != NULL) | | |
a3af9c38 | 610 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
611 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
612 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 613 | bad_page(page); |
79f4b7bf | 614 | return 1; |
8cc3b392 | 615 | } |
79f4b7bf HD |
616 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
617 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
618 | return 0; | |
1da177e4 LT |
619 | } |
620 | ||
621 | /* | |
5f8dcc21 | 622 | * Frees a number of pages from the PCP lists |
1da177e4 | 623 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 624 | * count is the number of pages to free. |
1da177e4 LT |
625 | * |
626 | * If the zone was previously in an "all pages pinned" state then look to | |
627 | * see if this freeing clears that state. | |
628 | * | |
629 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
630 | * pinned" detection logic. | |
631 | */ | |
5f8dcc21 MG |
632 | static void free_pcppages_bulk(struct zone *zone, int count, |
633 | struct per_cpu_pages *pcp) | |
1da177e4 | 634 | { |
5f8dcc21 | 635 | int migratetype = 0; |
a6f9edd6 | 636 | int batch_free = 0; |
72853e29 | 637 | int to_free = count; |
5f8dcc21 | 638 | |
c54ad30c | 639 | spin_lock(&zone->lock); |
93e4a89a | 640 | zone->all_unreclaimable = 0; |
1da177e4 | 641 | zone->pages_scanned = 0; |
f2260e6b | 642 | |
72853e29 | 643 | while (to_free) { |
48db57f8 | 644 | struct page *page; |
5f8dcc21 MG |
645 | struct list_head *list; |
646 | ||
647 | /* | |
a6f9edd6 MG |
648 | * Remove pages from lists in a round-robin fashion. A |
649 | * batch_free count is maintained that is incremented when an | |
650 | * empty list is encountered. This is so more pages are freed | |
651 | * off fuller lists instead of spinning excessively around empty | |
652 | * lists | |
5f8dcc21 MG |
653 | */ |
654 | do { | |
a6f9edd6 | 655 | batch_free++; |
5f8dcc21 MG |
656 | if (++migratetype == MIGRATE_PCPTYPES) |
657 | migratetype = 0; | |
658 | list = &pcp->lists[migratetype]; | |
659 | } while (list_empty(list)); | |
48db57f8 | 660 | |
1d16871d NK |
661 | /* This is the only non-empty list. Free them all. */ |
662 | if (batch_free == MIGRATE_PCPTYPES) | |
663 | batch_free = to_free; | |
664 | ||
a6f9edd6 MG |
665 | do { |
666 | page = list_entry(list->prev, struct page, lru); | |
667 | /* must delete as __free_one_page list manipulates */ | |
668 | list_del(&page->lru); | |
a7016235 HD |
669 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
670 | __free_one_page(page, zone, 0, page_private(page)); | |
671 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); | |
72853e29 | 672 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 673 | } |
72853e29 | 674 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
c54ad30c | 675 | spin_unlock(&zone->lock); |
1da177e4 LT |
676 | } |
677 | ||
ed0ae21d MG |
678 | static void free_one_page(struct zone *zone, struct page *page, int order, |
679 | int migratetype) | |
1da177e4 | 680 | { |
006d22d9 | 681 | spin_lock(&zone->lock); |
93e4a89a | 682 | zone->all_unreclaimable = 0; |
006d22d9 | 683 | zone->pages_scanned = 0; |
f2260e6b | 684 | |
ed0ae21d | 685 | __free_one_page(page, zone, order, migratetype); |
72853e29 | 686 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); |
006d22d9 | 687 | spin_unlock(&zone->lock); |
48db57f8 NP |
688 | } |
689 | ||
ec95f53a | 690 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 691 | { |
1da177e4 | 692 | int i; |
8cc3b392 | 693 | int bad = 0; |
1da177e4 | 694 | |
b413d48a | 695 | trace_mm_page_free(page, order); |
b1eeab67 VN |
696 | kmemcheck_free_shadow(page, order); |
697 | ||
8dd60a3a AA |
698 | if (PageAnon(page)) |
699 | page->mapping = NULL; | |
700 | for (i = 0; i < (1 << order); i++) | |
701 | bad += free_pages_check(page + i); | |
8cc3b392 | 702 | if (bad) |
ec95f53a | 703 | return false; |
689bcebf | 704 | |
3ac7fe5a | 705 | if (!PageHighMem(page)) { |
9858db50 | 706 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
707 | debug_check_no_obj_freed(page_address(page), |
708 | PAGE_SIZE << order); | |
709 | } | |
dafb1367 | 710 | arch_free_page(page, order); |
48db57f8 | 711 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 712 | |
ec95f53a KM |
713 | return true; |
714 | } | |
715 | ||
716 | static void __free_pages_ok(struct page *page, unsigned int order) | |
717 | { | |
718 | unsigned long flags; | |
719 | int wasMlocked = __TestClearPageMlocked(page); | |
720 | ||
721 | if (!free_pages_prepare(page, order)) | |
722 | return; | |
723 | ||
c54ad30c | 724 | local_irq_save(flags); |
c277331d | 725 | if (unlikely(wasMlocked)) |
da456f14 | 726 | free_page_mlock(page); |
f8891e5e | 727 | __count_vm_events(PGFREE, 1 << order); |
ed0ae21d MG |
728 | free_one_page(page_zone(page), page, order, |
729 | get_pageblock_migratetype(page)); | |
c54ad30c | 730 | local_irq_restore(flags); |
1da177e4 LT |
731 | } |
732 | ||
a226f6c8 DH |
733 | /* |
734 | * permit the bootmem allocator to evade page validation on high-order frees | |
735 | */ | |
af370fb8 | 736 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
737 | { |
738 | if (order == 0) { | |
739 | __ClearPageReserved(page); | |
740 | set_page_count(page, 0); | |
7835e98b | 741 | set_page_refcounted(page); |
545b1ea9 | 742 | __free_page(page); |
a226f6c8 | 743 | } else { |
a226f6c8 DH |
744 | int loop; |
745 | ||
545b1ea9 | 746 | prefetchw(page); |
53348f27 | 747 | for (loop = 0; loop < (1 << order); loop++) { |
a226f6c8 DH |
748 | struct page *p = &page[loop]; |
749 | ||
53348f27 | 750 | if (loop + 1 < (1 << order)) |
545b1ea9 | 751 | prefetchw(p + 1); |
a226f6c8 DH |
752 | __ClearPageReserved(p); |
753 | set_page_count(p, 0); | |
754 | } | |
755 | ||
7835e98b | 756 | set_page_refcounted(page); |
545b1ea9 | 757 | __free_pages(page, order); |
a226f6c8 DH |
758 | } |
759 | } | |
760 | ||
1da177e4 LT |
761 | |
762 | /* | |
763 | * The order of subdivision here is critical for the IO subsystem. | |
764 | * Please do not alter this order without good reasons and regression | |
765 | * testing. Specifically, as large blocks of memory are subdivided, | |
766 | * the order in which smaller blocks are delivered depends on the order | |
767 | * they're subdivided in this function. This is the primary factor | |
768 | * influencing the order in which pages are delivered to the IO | |
769 | * subsystem according to empirical testing, and this is also justified | |
770 | * by considering the behavior of a buddy system containing a single | |
771 | * large block of memory acted on by a series of small allocations. | |
772 | * This behavior is a critical factor in sglist merging's success. | |
773 | * | |
774 | * -- wli | |
775 | */ | |
085cc7d5 | 776 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
777 | int low, int high, struct free_area *area, |
778 | int migratetype) | |
1da177e4 LT |
779 | { |
780 | unsigned long size = 1 << high; | |
781 | ||
782 | while (high > low) { | |
783 | area--; | |
784 | high--; | |
785 | size >>= 1; | |
725d704e | 786 | VM_BUG_ON(bad_range(zone, &page[size])); |
c0a32fc5 SG |
787 | |
788 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
789 | if (high < debug_guardpage_minorder()) { | |
790 | /* | |
791 | * Mark as guard pages (or page), that will allow to | |
792 | * merge back to allocator when buddy will be freed. | |
793 | * Corresponding page table entries will not be touched, | |
794 | * pages will stay not present in virtual address space | |
795 | */ | |
796 | INIT_LIST_HEAD(&page[size].lru); | |
797 | set_page_guard_flag(&page[size]); | |
798 | set_page_private(&page[size], high); | |
799 | /* Guard pages are not available for any usage */ | |
800 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << high)); | |
801 | continue; | |
802 | } | |
803 | #endif | |
b2a0ac88 | 804 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
805 | area->nr_free++; |
806 | set_page_order(&page[size], high); | |
807 | } | |
1da177e4 LT |
808 | } |
809 | ||
1da177e4 LT |
810 | /* |
811 | * This page is about to be returned from the page allocator | |
812 | */ | |
2a7684a2 | 813 | static inline int check_new_page(struct page *page) |
1da177e4 | 814 | { |
92be2e33 NP |
815 | if (unlikely(page_mapcount(page) | |
816 | (page->mapping != NULL) | | |
a3af9c38 | 817 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
818 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
819 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 820 | bad_page(page); |
689bcebf | 821 | return 1; |
8cc3b392 | 822 | } |
2a7684a2 WF |
823 | return 0; |
824 | } | |
825 | ||
826 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
827 | { | |
828 | int i; | |
829 | ||
830 | for (i = 0; i < (1 << order); i++) { | |
831 | struct page *p = page + i; | |
832 | if (unlikely(check_new_page(p))) | |
833 | return 1; | |
834 | } | |
689bcebf | 835 | |
4c21e2f2 | 836 | set_page_private(page, 0); |
7835e98b | 837 | set_page_refcounted(page); |
cc102509 NP |
838 | |
839 | arch_alloc_page(page, order); | |
1da177e4 | 840 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
841 | |
842 | if (gfp_flags & __GFP_ZERO) | |
843 | prep_zero_page(page, order, gfp_flags); | |
844 | ||
845 | if (order && (gfp_flags & __GFP_COMP)) | |
846 | prep_compound_page(page, order); | |
847 | ||
689bcebf | 848 | return 0; |
1da177e4 LT |
849 | } |
850 | ||
56fd56b8 MG |
851 | /* |
852 | * Go through the free lists for the given migratetype and remove | |
853 | * the smallest available page from the freelists | |
854 | */ | |
728ec980 MG |
855 | static inline |
856 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
857 | int migratetype) |
858 | { | |
859 | unsigned int current_order; | |
860 | struct free_area * area; | |
861 | struct page *page; | |
862 | ||
863 | /* Find a page of the appropriate size in the preferred list */ | |
864 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
865 | area = &(zone->free_area[current_order]); | |
866 | if (list_empty(&area->free_list[migratetype])) | |
867 | continue; | |
868 | ||
869 | page = list_entry(area->free_list[migratetype].next, | |
870 | struct page, lru); | |
871 | list_del(&page->lru); | |
872 | rmv_page_order(page); | |
873 | area->nr_free--; | |
56fd56b8 MG |
874 | expand(zone, page, order, current_order, area, migratetype); |
875 | return page; | |
876 | } | |
877 | ||
878 | return NULL; | |
879 | } | |
880 | ||
881 | ||
b2a0ac88 MG |
882 | /* |
883 | * This array describes the order lists are fallen back to when | |
884 | * the free lists for the desirable migrate type are depleted | |
885 | */ | |
886 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
887 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
888 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
889 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
890 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
891 | }; |
892 | ||
c361be55 MG |
893 | /* |
894 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 895 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
896 | * boundary. If alignment is required, use move_freepages_block() |
897 | */ | |
b69a7288 AB |
898 | static int move_freepages(struct zone *zone, |
899 | struct page *start_page, struct page *end_page, | |
900 | int migratetype) | |
c361be55 MG |
901 | { |
902 | struct page *page; | |
903 | unsigned long order; | |
d100313f | 904 | int pages_moved = 0; |
c361be55 MG |
905 | |
906 | #ifndef CONFIG_HOLES_IN_ZONE | |
907 | /* | |
908 | * page_zone is not safe to call in this context when | |
909 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
910 | * anyway as we check zone boundaries in move_freepages_block(). | |
911 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 912 | * grouping pages by mobility |
c361be55 MG |
913 | */ |
914 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
915 | #endif | |
916 | ||
917 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
918 | /* Make sure we are not inadvertently changing nodes */ |
919 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
920 | ||
c361be55 MG |
921 | if (!pfn_valid_within(page_to_pfn(page))) { |
922 | page++; | |
923 | continue; | |
924 | } | |
925 | ||
926 | if (!PageBuddy(page)) { | |
927 | page++; | |
928 | continue; | |
929 | } | |
930 | ||
931 | order = page_order(page); | |
84be48d8 KS |
932 | list_move(&page->lru, |
933 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 934 | page += 1 << order; |
d100313f | 935 | pages_moved += 1 << order; |
c361be55 MG |
936 | } |
937 | ||
d100313f | 938 | return pages_moved; |
c361be55 MG |
939 | } |
940 | ||
b69a7288 AB |
941 | static int move_freepages_block(struct zone *zone, struct page *page, |
942 | int migratetype) | |
c361be55 MG |
943 | { |
944 | unsigned long start_pfn, end_pfn; | |
945 | struct page *start_page, *end_page; | |
946 | ||
947 | start_pfn = page_to_pfn(page); | |
d9c23400 | 948 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 949 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
950 | end_page = start_page + pageblock_nr_pages - 1; |
951 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
952 | |
953 | /* Do not cross zone boundaries */ | |
954 | if (start_pfn < zone->zone_start_pfn) | |
955 | start_page = page; | |
956 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
957 | return 0; | |
958 | ||
959 | return move_freepages(zone, start_page, end_page, migratetype); | |
960 | } | |
961 | ||
2f66a68f MG |
962 | static void change_pageblock_range(struct page *pageblock_page, |
963 | int start_order, int migratetype) | |
964 | { | |
965 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
966 | ||
967 | while (nr_pageblocks--) { | |
968 | set_pageblock_migratetype(pageblock_page, migratetype); | |
969 | pageblock_page += pageblock_nr_pages; | |
970 | } | |
971 | } | |
972 | ||
b2a0ac88 | 973 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
974 | static inline struct page * |
975 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
976 | { |
977 | struct free_area * area; | |
978 | int current_order; | |
979 | struct page *page; | |
980 | int migratetype, i; | |
981 | ||
982 | /* Find the largest possible block of pages in the other list */ | |
983 | for (current_order = MAX_ORDER-1; current_order >= order; | |
984 | --current_order) { | |
985 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
986 | migratetype = fallbacks[start_migratetype][i]; | |
987 | ||
56fd56b8 MG |
988 | /* MIGRATE_RESERVE handled later if necessary */ |
989 | if (migratetype == MIGRATE_RESERVE) | |
990 | continue; | |
e010487d | 991 | |
b2a0ac88 MG |
992 | area = &(zone->free_area[current_order]); |
993 | if (list_empty(&area->free_list[migratetype])) | |
994 | continue; | |
995 | ||
996 | page = list_entry(area->free_list[migratetype].next, | |
997 | struct page, lru); | |
998 | area->nr_free--; | |
999 | ||
1000 | /* | |
c361be55 | 1001 | * If breaking a large block of pages, move all free |
46dafbca MG |
1002 | * pages to the preferred allocation list. If falling |
1003 | * back for a reclaimable kernel allocation, be more | |
25985edc | 1004 | * aggressive about taking ownership of free pages |
b2a0ac88 | 1005 | */ |
d9c23400 | 1006 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
dd5d241e MG |
1007 | start_migratetype == MIGRATE_RECLAIMABLE || |
1008 | page_group_by_mobility_disabled) { | |
46dafbca MG |
1009 | unsigned long pages; |
1010 | pages = move_freepages_block(zone, page, | |
1011 | start_migratetype); | |
1012 | ||
1013 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
1014 | if (pages >= (1 << (pageblock_order-1)) || |
1015 | page_group_by_mobility_disabled) | |
46dafbca MG |
1016 | set_pageblock_migratetype(page, |
1017 | start_migratetype); | |
1018 | ||
b2a0ac88 | 1019 | migratetype = start_migratetype; |
c361be55 | 1020 | } |
b2a0ac88 MG |
1021 | |
1022 | /* Remove the page from the freelists */ | |
1023 | list_del(&page->lru); | |
1024 | rmv_page_order(page); | |
b2a0ac88 | 1025 | |
2f66a68f MG |
1026 | /* Take ownership for orders >= pageblock_order */ |
1027 | if (current_order >= pageblock_order) | |
1028 | change_pageblock_range(page, current_order, | |
b2a0ac88 MG |
1029 | start_migratetype); |
1030 | ||
1031 | expand(zone, page, order, current_order, area, migratetype); | |
e0fff1bd MG |
1032 | |
1033 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
1034 | start_migratetype, migratetype); | |
1035 | ||
b2a0ac88 MG |
1036 | return page; |
1037 | } | |
1038 | } | |
1039 | ||
728ec980 | 1040 | return NULL; |
b2a0ac88 MG |
1041 | } |
1042 | ||
56fd56b8 | 1043 | /* |
1da177e4 LT |
1044 | * Do the hard work of removing an element from the buddy allocator. |
1045 | * Call me with the zone->lock already held. | |
1046 | */ | |
b2a0ac88 MG |
1047 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1048 | int migratetype) | |
1da177e4 | 1049 | { |
1da177e4 LT |
1050 | struct page *page; |
1051 | ||
728ec980 | 1052 | retry_reserve: |
56fd56b8 | 1053 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1054 | |
728ec980 | 1055 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1056 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1057 | |
728ec980 MG |
1058 | /* |
1059 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1060 | * is used because __rmqueue_smallest is an inline function | |
1061 | * and we want just one call site | |
1062 | */ | |
1063 | if (!page) { | |
1064 | migratetype = MIGRATE_RESERVE; | |
1065 | goto retry_reserve; | |
1066 | } | |
1067 | } | |
1068 | ||
0d3d062a | 1069 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1070 | return page; |
1da177e4 LT |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * Obtain a specified number of elements from the buddy allocator, all under | |
1075 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1076 | * Returns the number of new pages which were placed at *list. | |
1077 | */ | |
1078 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 | 1079 | unsigned long count, struct list_head *list, |
e084b2d9 | 1080 | int migratetype, int cold) |
1da177e4 | 1081 | { |
1da177e4 | 1082 | int i; |
1da177e4 | 1083 | |
c54ad30c | 1084 | spin_lock(&zone->lock); |
1da177e4 | 1085 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1086 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1087 | if (unlikely(page == NULL)) |
1da177e4 | 1088 | break; |
81eabcbe MG |
1089 | |
1090 | /* | |
1091 | * Split buddy pages returned by expand() are received here | |
1092 | * in physical page order. The page is added to the callers and | |
1093 | * list and the list head then moves forward. From the callers | |
1094 | * perspective, the linked list is ordered by page number in | |
1095 | * some conditions. This is useful for IO devices that can | |
1096 | * merge IO requests if the physical pages are ordered | |
1097 | * properly. | |
1098 | */ | |
e084b2d9 MG |
1099 | if (likely(cold == 0)) |
1100 | list_add(&page->lru, list); | |
1101 | else | |
1102 | list_add_tail(&page->lru, list); | |
535131e6 | 1103 | set_page_private(page, migratetype); |
81eabcbe | 1104 | list = &page->lru; |
1da177e4 | 1105 | } |
f2260e6b | 1106 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1107 | spin_unlock(&zone->lock); |
085cc7d5 | 1108 | return i; |
1da177e4 LT |
1109 | } |
1110 | ||
4ae7c039 | 1111 | #ifdef CONFIG_NUMA |
8fce4d8e | 1112 | /* |
4037d452 CL |
1113 | * Called from the vmstat counter updater to drain pagesets of this |
1114 | * currently executing processor on remote nodes after they have | |
1115 | * expired. | |
1116 | * | |
879336c3 CL |
1117 | * Note that this function must be called with the thread pinned to |
1118 | * a single processor. | |
8fce4d8e | 1119 | */ |
4037d452 | 1120 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1121 | { |
4ae7c039 | 1122 | unsigned long flags; |
4037d452 | 1123 | int to_drain; |
4ae7c039 | 1124 | |
4037d452 CL |
1125 | local_irq_save(flags); |
1126 | if (pcp->count >= pcp->batch) | |
1127 | to_drain = pcp->batch; | |
1128 | else | |
1129 | to_drain = pcp->count; | |
5f8dcc21 | 1130 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 CL |
1131 | pcp->count -= to_drain; |
1132 | local_irq_restore(flags); | |
4ae7c039 CL |
1133 | } |
1134 | #endif | |
1135 | ||
9f8f2172 CL |
1136 | /* |
1137 | * Drain pages of the indicated processor. | |
1138 | * | |
1139 | * The processor must either be the current processor and the | |
1140 | * thread pinned to the current processor or a processor that | |
1141 | * is not online. | |
1142 | */ | |
1143 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1144 | { |
c54ad30c | 1145 | unsigned long flags; |
1da177e4 | 1146 | struct zone *zone; |
1da177e4 | 1147 | |
ee99c71c | 1148 | for_each_populated_zone(zone) { |
1da177e4 | 1149 | struct per_cpu_pageset *pset; |
3dfa5721 | 1150 | struct per_cpu_pages *pcp; |
1da177e4 | 1151 | |
99dcc3e5 CL |
1152 | local_irq_save(flags); |
1153 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1154 | |
1155 | pcp = &pset->pcp; | |
2ff754fa DR |
1156 | if (pcp->count) { |
1157 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1158 | pcp->count = 0; | |
1159 | } | |
3dfa5721 | 1160 | local_irq_restore(flags); |
1da177e4 LT |
1161 | } |
1162 | } | |
1da177e4 | 1163 | |
9f8f2172 CL |
1164 | /* |
1165 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1166 | */ | |
1167 | void drain_local_pages(void *arg) | |
1168 | { | |
1169 | drain_pages(smp_processor_id()); | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1174 | */ | |
1175 | void drain_all_pages(void) | |
1176 | { | |
15c8b6c1 | 1177 | on_each_cpu(drain_local_pages, NULL, 1); |
9f8f2172 CL |
1178 | } |
1179 | ||
296699de | 1180 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1181 | |
1182 | void mark_free_pages(struct zone *zone) | |
1183 | { | |
f623f0db RW |
1184 | unsigned long pfn, max_zone_pfn; |
1185 | unsigned long flags; | |
b2a0ac88 | 1186 | int order, t; |
1da177e4 LT |
1187 | struct list_head *curr; |
1188 | ||
1189 | if (!zone->spanned_pages) | |
1190 | return; | |
1191 | ||
1192 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1193 | |
1194 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1195 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1196 | if (pfn_valid(pfn)) { | |
1197 | struct page *page = pfn_to_page(pfn); | |
1198 | ||
7be98234 RW |
1199 | if (!swsusp_page_is_forbidden(page)) |
1200 | swsusp_unset_page_free(page); | |
f623f0db | 1201 | } |
1da177e4 | 1202 | |
b2a0ac88 MG |
1203 | for_each_migratetype_order(order, t) { |
1204 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1205 | unsigned long i; |
1da177e4 | 1206 | |
f623f0db RW |
1207 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1208 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1209 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1210 | } |
b2a0ac88 | 1211 | } |
1da177e4 LT |
1212 | spin_unlock_irqrestore(&zone->lock, flags); |
1213 | } | |
e2c55dc8 | 1214 | #endif /* CONFIG_PM */ |
1da177e4 | 1215 | |
1da177e4 LT |
1216 | /* |
1217 | * Free a 0-order page | |
fc91668e | 1218 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1219 | */ |
fc91668e | 1220 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1221 | { |
1222 | struct zone *zone = page_zone(page); | |
1223 | struct per_cpu_pages *pcp; | |
1224 | unsigned long flags; | |
5f8dcc21 | 1225 | int migratetype; |
451ea25d | 1226 | int wasMlocked = __TestClearPageMlocked(page); |
1da177e4 | 1227 | |
ec95f53a | 1228 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1229 | return; |
1230 | ||
5f8dcc21 MG |
1231 | migratetype = get_pageblock_migratetype(page); |
1232 | set_page_private(page, migratetype); | |
1da177e4 | 1233 | local_irq_save(flags); |
c277331d | 1234 | if (unlikely(wasMlocked)) |
da456f14 | 1235 | free_page_mlock(page); |
f8891e5e | 1236 | __count_vm_event(PGFREE); |
da456f14 | 1237 | |
5f8dcc21 MG |
1238 | /* |
1239 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1240 | * Free ISOLATE pages back to the allocator because they are being | |
1241 | * offlined but treat RESERVE as movable pages so we can get those | |
1242 | * areas back if necessary. Otherwise, we may have to free | |
1243 | * excessively into the page allocator | |
1244 | */ | |
1245 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1246 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1247 | free_one_page(zone, page, 0, migratetype); | |
1248 | goto out; | |
1249 | } | |
1250 | migratetype = MIGRATE_MOVABLE; | |
1251 | } | |
1252 | ||
99dcc3e5 | 1253 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1254 | if (cold) |
5f8dcc21 | 1255 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1256 | else |
5f8dcc21 | 1257 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1258 | pcp->count++; |
48db57f8 | 1259 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1260 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1261 | pcp->count -= pcp->batch; |
1262 | } | |
5f8dcc21 MG |
1263 | |
1264 | out: | |
1da177e4 | 1265 | local_irq_restore(flags); |
1da177e4 LT |
1266 | } |
1267 | ||
cc59850e KK |
1268 | /* |
1269 | * Free a list of 0-order pages | |
1270 | */ | |
1271 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1272 | { | |
1273 | struct page *page, *next; | |
1274 | ||
1275 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1276 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1277 | free_hot_cold_page(page, cold); |
1278 | } | |
1279 | } | |
1280 | ||
8dfcc9ba NP |
1281 | /* |
1282 | * split_page takes a non-compound higher-order page, and splits it into | |
1283 | * n (1<<order) sub-pages: page[0..n] | |
1284 | * Each sub-page must be freed individually. | |
1285 | * | |
1286 | * Note: this is probably too low level an operation for use in drivers. | |
1287 | * Please consult with lkml before using this in your driver. | |
1288 | */ | |
1289 | void split_page(struct page *page, unsigned int order) | |
1290 | { | |
1291 | int i; | |
1292 | ||
725d704e NP |
1293 | VM_BUG_ON(PageCompound(page)); |
1294 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1295 | |
1296 | #ifdef CONFIG_KMEMCHECK | |
1297 | /* | |
1298 | * Split shadow pages too, because free(page[0]) would | |
1299 | * otherwise free the whole shadow. | |
1300 | */ | |
1301 | if (kmemcheck_page_is_tracked(page)) | |
1302 | split_page(virt_to_page(page[0].shadow), order); | |
1303 | #endif | |
1304 | ||
7835e98b NP |
1305 | for (i = 1; i < (1 << order); i++) |
1306 | set_page_refcounted(page + i); | |
8dfcc9ba | 1307 | } |
8dfcc9ba | 1308 | |
748446bb MG |
1309 | /* |
1310 | * Similar to split_page except the page is already free. As this is only | |
1311 | * being used for migration, the migratetype of the block also changes. | |
1312 | * As this is called with interrupts disabled, the caller is responsible | |
1313 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1314 | * are enabled. | |
1315 | * | |
1316 | * Note: this is probably too low level an operation for use in drivers. | |
1317 | * Please consult with lkml before using this in your driver. | |
1318 | */ | |
1319 | int split_free_page(struct page *page) | |
1320 | { | |
1321 | unsigned int order; | |
1322 | unsigned long watermark; | |
1323 | struct zone *zone; | |
1324 | ||
1325 | BUG_ON(!PageBuddy(page)); | |
1326 | ||
1327 | zone = page_zone(page); | |
1328 | order = page_order(page); | |
1329 | ||
1330 | /* Obey watermarks as if the page was being allocated */ | |
1331 | watermark = low_wmark_pages(zone) + (1 << order); | |
1332 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1333 | return 0; | |
1334 | ||
1335 | /* Remove page from free list */ | |
1336 | list_del(&page->lru); | |
1337 | zone->free_area[order].nr_free--; | |
1338 | rmv_page_order(page); | |
1339 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order)); | |
1340 | ||
1341 | /* Split into individual pages */ | |
1342 | set_page_refcounted(page); | |
1343 | split_page(page, order); | |
1344 | ||
1345 | if (order >= pageblock_order - 1) { | |
1346 | struct page *endpage = page + (1 << order) - 1; | |
1347 | for (; page < endpage; page += pageblock_nr_pages) | |
1348 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
1349 | } | |
1350 | ||
1351 | return 1 << order; | |
1352 | } | |
1353 | ||
1da177e4 LT |
1354 | /* |
1355 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1356 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1357 | * or two. | |
1358 | */ | |
0a15c3e9 MG |
1359 | static inline |
1360 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1361 | struct zone *zone, int order, gfp_t gfp_flags, |
1362 | int migratetype) | |
1da177e4 LT |
1363 | { |
1364 | unsigned long flags; | |
689bcebf | 1365 | struct page *page; |
1da177e4 LT |
1366 | int cold = !!(gfp_flags & __GFP_COLD); |
1367 | ||
689bcebf | 1368 | again: |
48db57f8 | 1369 | if (likely(order == 0)) { |
1da177e4 | 1370 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1371 | struct list_head *list; |
1da177e4 | 1372 | |
1da177e4 | 1373 | local_irq_save(flags); |
99dcc3e5 CL |
1374 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1375 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1376 | if (list_empty(list)) { |
535131e6 | 1377 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1378 | pcp->batch, list, |
e084b2d9 | 1379 | migratetype, cold); |
5f8dcc21 | 1380 | if (unlikely(list_empty(list))) |
6fb332fa | 1381 | goto failed; |
535131e6 | 1382 | } |
b92a6edd | 1383 | |
5f8dcc21 MG |
1384 | if (cold) |
1385 | page = list_entry(list->prev, struct page, lru); | |
1386 | else | |
1387 | page = list_entry(list->next, struct page, lru); | |
1388 | ||
b92a6edd MG |
1389 | list_del(&page->lru); |
1390 | pcp->count--; | |
7fb1d9fc | 1391 | } else { |
dab48dab AM |
1392 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1393 | /* | |
1394 | * __GFP_NOFAIL is not to be used in new code. | |
1395 | * | |
1396 | * All __GFP_NOFAIL callers should be fixed so that they | |
1397 | * properly detect and handle allocation failures. | |
1398 | * | |
1399 | * We most definitely don't want callers attempting to | |
4923abf9 | 1400 | * allocate greater than order-1 page units with |
dab48dab AM |
1401 | * __GFP_NOFAIL. |
1402 | */ | |
4923abf9 | 1403 | WARN_ON_ONCE(order > 1); |
dab48dab | 1404 | } |
1da177e4 | 1405 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1406 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1407 | spin_unlock(&zone->lock); |
1408 | if (!page) | |
1409 | goto failed; | |
6ccf80eb | 1410 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
1da177e4 LT |
1411 | } |
1412 | ||
f8891e5e | 1413 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1414 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1415 | local_irq_restore(flags); |
1da177e4 | 1416 | |
725d704e | 1417 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1418 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1419 | goto again; |
1da177e4 | 1420 | return page; |
a74609fa NP |
1421 | |
1422 | failed: | |
1423 | local_irq_restore(flags); | |
a74609fa | 1424 | return NULL; |
1da177e4 LT |
1425 | } |
1426 | ||
41858966 MG |
1427 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
1428 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1429 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1430 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1431 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1432 | ||
1433 | /* Mask to get the watermark bits */ | |
1434 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1435 | ||
3148890b NP |
1436 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
1437 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1438 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1439 | |
933e312e AM |
1440 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1441 | ||
b2588c4b | 1442 | static struct { |
933e312e AM |
1443 | struct fault_attr attr; |
1444 | ||
1445 | u32 ignore_gfp_highmem; | |
1446 | u32 ignore_gfp_wait; | |
54114994 | 1447 | u32 min_order; |
933e312e AM |
1448 | } fail_page_alloc = { |
1449 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1450 | .ignore_gfp_wait = 1, |
1451 | .ignore_gfp_highmem = 1, | |
54114994 | 1452 | .min_order = 1, |
933e312e AM |
1453 | }; |
1454 | ||
1455 | static int __init setup_fail_page_alloc(char *str) | |
1456 | { | |
1457 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1458 | } | |
1459 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1460 | ||
1461 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1462 | { | |
54114994 AM |
1463 | if (order < fail_page_alloc.min_order) |
1464 | return 0; | |
933e312e AM |
1465 | if (gfp_mask & __GFP_NOFAIL) |
1466 | return 0; | |
1467 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1468 | return 0; | |
1469 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1470 | return 0; | |
1471 | ||
1472 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1473 | } | |
1474 | ||
1475 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1476 | ||
1477 | static int __init fail_page_alloc_debugfs(void) | |
1478 | { | |
f4ae40a6 | 1479 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1480 | struct dentry *dir; |
933e312e | 1481 | |
dd48c085 AM |
1482 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1483 | &fail_page_alloc.attr); | |
1484 | if (IS_ERR(dir)) | |
1485 | return PTR_ERR(dir); | |
933e312e | 1486 | |
b2588c4b AM |
1487 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1488 | &fail_page_alloc.ignore_gfp_wait)) | |
1489 | goto fail; | |
1490 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1491 | &fail_page_alloc.ignore_gfp_highmem)) | |
1492 | goto fail; | |
1493 | if (!debugfs_create_u32("min-order", mode, dir, | |
1494 | &fail_page_alloc.min_order)) | |
1495 | goto fail; | |
1496 | ||
1497 | return 0; | |
1498 | fail: | |
dd48c085 | 1499 | debugfs_remove_recursive(dir); |
933e312e | 1500 | |
b2588c4b | 1501 | return -ENOMEM; |
933e312e AM |
1502 | } |
1503 | ||
1504 | late_initcall(fail_page_alloc_debugfs); | |
1505 | ||
1506 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1507 | ||
1508 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1509 | ||
1510 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1511 | { | |
1512 | return 0; | |
1513 | } | |
1514 | ||
1515 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1516 | ||
1da177e4 | 1517 | /* |
88f5acf8 | 1518 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1519 | * of the allocation. |
1520 | */ | |
88f5acf8 MG |
1521 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1522 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1523 | { |
1524 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1525 | long min = mark; |
1da177e4 LT |
1526 | int o; |
1527 | ||
88f5acf8 | 1528 | free_pages -= (1 << order) + 1; |
7fb1d9fc | 1529 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1530 | min -= min / 2; |
7fb1d9fc | 1531 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1532 | min -= min / 4; |
1533 | ||
1534 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 1535 | return false; |
1da177e4 LT |
1536 | for (o = 0; o < order; o++) { |
1537 | /* At the next order, this order's pages become unavailable */ | |
1538 | free_pages -= z->free_area[o].nr_free << o; | |
1539 | ||
1540 | /* Require fewer higher order pages to be free */ | |
1541 | min >>= 1; | |
1542 | ||
1543 | if (free_pages <= min) | |
88f5acf8 | 1544 | return false; |
1da177e4 | 1545 | } |
88f5acf8 MG |
1546 | return true; |
1547 | } | |
1548 | ||
1549 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1550 | int classzone_idx, int alloc_flags) | |
1551 | { | |
1552 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1553 | zone_page_state(z, NR_FREE_PAGES)); | |
1554 | } | |
1555 | ||
1556 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1557 | int classzone_idx, int alloc_flags) | |
1558 | { | |
1559 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1560 | ||
1561 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1562 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1563 | ||
1564 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1565 | free_pages); | |
1da177e4 LT |
1566 | } |
1567 | ||
9276b1bc PJ |
1568 | #ifdef CONFIG_NUMA |
1569 | /* | |
1570 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1571 | * skip over zones that are not allowed by the cpuset, or that have | |
1572 | * been recently (in last second) found to be nearly full. See further | |
1573 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1574 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1575 | * |
1576 | * If the zonelist cache is present in the passed in zonelist, then | |
1577 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1578 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1579 | * |
1580 | * If the zonelist cache is not available for this zonelist, does | |
1581 | * nothing and returns NULL. | |
1582 | * | |
1583 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1584 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1585 | * | |
1586 | * We hold off even calling zlc_setup, until after we've checked the | |
1587 | * first zone in the zonelist, on the theory that most allocations will | |
1588 | * be satisfied from that first zone, so best to examine that zone as | |
1589 | * quickly as we can. | |
1590 | */ | |
1591 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1592 | { | |
1593 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1594 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1595 | ||
1596 | zlc = zonelist->zlcache_ptr; | |
1597 | if (!zlc) | |
1598 | return NULL; | |
1599 | ||
f05111f5 | 1600 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1601 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1602 | zlc->last_full_zap = jiffies; | |
1603 | } | |
1604 | ||
1605 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1606 | &cpuset_current_mems_allowed : | |
37b07e41 | 1607 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1608 | return allowednodes; |
1609 | } | |
1610 | ||
1611 | /* | |
1612 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1613 | * if it is worth looking at further for free memory: | |
1614 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1615 | * bit set in the zonelist_cache fullzones BITMAP). | |
1616 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1617 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1618 | * Return true (non-zero) if zone is worth looking at further, or | |
1619 | * else return false (zero) if it is not. | |
1620 | * | |
1621 | * This check -ignores- the distinction between various watermarks, | |
1622 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1623 | * found to be full for any variation of these watermarks, it will | |
1624 | * be considered full for up to one second by all requests, unless | |
1625 | * we are so low on memory on all allowed nodes that we are forced | |
1626 | * into the second scan of the zonelist. | |
1627 | * | |
1628 | * In the second scan we ignore this zonelist cache and exactly | |
1629 | * apply the watermarks to all zones, even it is slower to do so. | |
1630 | * We are low on memory in the second scan, and should leave no stone | |
1631 | * unturned looking for a free page. | |
1632 | */ | |
dd1a239f | 1633 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1634 | nodemask_t *allowednodes) |
1635 | { | |
1636 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1637 | int i; /* index of *z in zonelist zones */ | |
1638 | int n; /* node that zone *z is on */ | |
1639 | ||
1640 | zlc = zonelist->zlcache_ptr; | |
1641 | if (!zlc) | |
1642 | return 1; | |
1643 | ||
dd1a239f | 1644 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1645 | n = zlc->z_to_n[i]; |
1646 | ||
1647 | /* This zone is worth trying if it is allowed but not full */ | |
1648 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1653 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1654 | * from that zone don't waste time re-examining it. | |
1655 | */ | |
dd1a239f | 1656 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1657 | { |
1658 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1659 | int i; /* index of *z in zonelist zones */ | |
1660 | ||
1661 | zlc = zonelist->zlcache_ptr; | |
1662 | if (!zlc) | |
1663 | return; | |
1664 | ||
dd1a239f | 1665 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1666 | |
1667 | set_bit(i, zlc->fullzones); | |
1668 | } | |
1669 | ||
76d3fbf8 MG |
1670 | /* |
1671 | * clear all zones full, called after direct reclaim makes progress so that | |
1672 | * a zone that was recently full is not skipped over for up to a second | |
1673 | */ | |
1674 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1675 | { | |
1676 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1677 | ||
1678 | zlc = zonelist->zlcache_ptr; | |
1679 | if (!zlc) | |
1680 | return; | |
1681 | ||
1682 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1683 | } | |
1684 | ||
9276b1bc PJ |
1685 | #else /* CONFIG_NUMA */ |
1686 | ||
1687 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1688 | { | |
1689 | return NULL; | |
1690 | } | |
1691 | ||
dd1a239f | 1692 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1693 | nodemask_t *allowednodes) |
1694 | { | |
1695 | return 1; | |
1696 | } | |
1697 | ||
dd1a239f | 1698 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1699 | { |
1700 | } | |
76d3fbf8 MG |
1701 | |
1702 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1703 | { | |
1704 | } | |
9276b1bc PJ |
1705 | #endif /* CONFIG_NUMA */ |
1706 | ||
7fb1d9fc | 1707 | /* |
0798e519 | 1708 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1709 | * a page. |
1710 | */ | |
1711 | static struct page * | |
19770b32 | 1712 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1713 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1714 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1715 | { |
dd1a239f | 1716 | struct zoneref *z; |
7fb1d9fc | 1717 | struct page *page = NULL; |
54a6eb5c | 1718 | int classzone_idx; |
5117f45d | 1719 | struct zone *zone; |
9276b1bc PJ |
1720 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1721 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1722 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1723 | |
19770b32 | 1724 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1725 | zonelist_scan: |
7fb1d9fc | 1726 | /* |
9276b1bc | 1727 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1728 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1729 | */ | |
19770b32 MG |
1730 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1731 | high_zoneidx, nodemask) { | |
9276b1bc PJ |
1732 | if (NUMA_BUILD && zlc_active && |
1733 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1734 | continue; | |
7fb1d9fc | 1735 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1736 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1737 | continue; |
a756cf59 JW |
1738 | /* |
1739 | * When allocating a page cache page for writing, we | |
1740 | * want to get it from a zone that is within its dirty | |
1741 | * limit, such that no single zone holds more than its | |
1742 | * proportional share of globally allowed dirty pages. | |
1743 | * The dirty limits take into account the zone's | |
1744 | * lowmem reserves and high watermark so that kswapd | |
1745 | * should be able to balance it without having to | |
1746 | * write pages from its LRU list. | |
1747 | * | |
1748 | * This may look like it could increase pressure on | |
1749 | * lower zones by failing allocations in higher zones | |
1750 | * before they are full. But the pages that do spill | |
1751 | * over are limited as the lower zones are protected | |
1752 | * by this very same mechanism. It should not become | |
1753 | * a practical burden to them. | |
1754 | * | |
1755 | * XXX: For now, allow allocations to potentially | |
1756 | * exceed the per-zone dirty limit in the slowpath | |
1757 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1758 | * which is important when on a NUMA setup the allowed | |
1759 | * zones are together not big enough to reach the | |
1760 | * global limit. The proper fix for these situations | |
1761 | * will require awareness of zones in the | |
1762 | * dirty-throttling and the flusher threads. | |
1763 | */ | |
1764 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1765 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
1766 | goto this_zone_full; | |
7fb1d9fc | 1767 | |
41858966 | 1768 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1769 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1770 | unsigned long mark; |
fa5e084e MG |
1771 | int ret; |
1772 | ||
41858966 | 1773 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1774 | if (zone_watermark_ok(zone, order, mark, |
1775 | classzone_idx, alloc_flags)) | |
1776 | goto try_this_zone; | |
1777 | ||
cd38b115 MG |
1778 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
1779 | /* | |
1780 | * we do zlc_setup if there are multiple nodes | |
1781 | * and before considering the first zone allowed | |
1782 | * by the cpuset. | |
1783 | */ | |
1784 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1785 | zlc_active = 1; | |
1786 | did_zlc_setup = 1; | |
1787 | } | |
1788 | ||
fa5e084e MG |
1789 | if (zone_reclaim_mode == 0) |
1790 | goto this_zone_full; | |
1791 | ||
cd38b115 MG |
1792 | /* |
1793 | * As we may have just activated ZLC, check if the first | |
1794 | * eligible zone has failed zone_reclaim recently. | |
1795 | */ | |
1796 | if (NUMA_BUILD && zlc_active && | |
1797 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1798 | continue; | |
1799 | ||
fa5e084e MG |
1800 | ret = zone_reclaim(zone, gfp_mask, order); |
1801 | switch (ret) { | |
1802 | case ZONE_RECLAIM_NOSCAN: | |
1803 | /* did not scan */ | |
cd38b115 | 1804 | continue; |
fa5e084e MG |
1805 | case ZONE_RECLAIM_FULL: |
1806 | /* scanned but unreclaimable */ | |
cd38b115 | 1807 | continue; |
fa5e084e MG |
1808 | default: |
1809 | /* did we reclaim enough */ | |
1810 | if (!zone_watermark_ok(zone, order, mark, | |
1811 | classzone_idx, alloc_flags)) | |
9276b1bc | 1812 | goto this_zone_full; |
0798e519 | 1813 | } |
7fb1d9fc RS |
1814 | } |
1815 | ||
fa5e084e | 1816 | try_this_zone: |
3dd28266 MG |
1817 | page = buffered_rmqueue(preferred_zone, zone, order, |
1818 | gfp_mask, migratetype); | |
0798e519 | 1819 | if (page) |
7fb1d9fc | 1820 | break; |
9276b1bc PJ |
1821 | this_zone_full: |
1822 | if (NUMA_BUILD) | |
1823 | zlc_mark_zone_full(zonelist, z); | |
54a6eb5c | 1824 | } |
9276b1bc PJ |
1825 | |
1826 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1827 | /* Disable zlc cache for second zonelist scan */ | |
1828 | zlc_active = 0; | |
1829 | goto zonelist_scan; | |
1830 | } | |
7fb1d9fc | 1831 | return page; |
753ee728 MH |
1832 | } |
1833 | ||
29423e77 DR |
1834 | /* |
1835 | * Large machines with many possible nodes should not always dump per-node | |
1836 | * meminfo in irq context. | |
1837 | */ | |
1838 | static inline bool should_suppress_show_mem(void) | |
1839 | { | |
1840 | bool ret = false; | |
1841 | ||
1842 | #if NODES_SHIFT > 8 | |
1843 | ret = in_interrupt(); | |
1844 | #endif | |
1845 | return ret; | |
1846 | } | |
1847 | ||
a238ab5b DH |
1848 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
1849 | DEFAULT_RATELIMIT_INTERVAL, | |
1850 | DEFAULT_RATELIMIT_BURST); | |
1851 | ||
1852 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
1853 | { | |
a238ab5b DH |
1854 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
1855 | ||
c0a32fc5 SG |
1856 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
1857 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
1858 | return; |
1859 | ||
1860 | /* | |
1861 | * This documents exceptions given to allocations in certain | |
1862 | * contexts that are allowed to allocate outside current's set | |
1863 | * of allowed nodes. | |
1864 | */ | |
1865 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
1866 | if (test_thread_flag(TIF_MEMDIE) || | |
1867 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
1868 | filter &= ~SHOW_MEM_FILTER_NODES; | |
1869 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
1870 | filter &= ~SHOW_MEM_FILTER_NODES; | |
1871 | ||
1872 | if (fmt) { | |
3ee9a4f0 JP |
1873 | struct va_format vaf; |
1874 | va_list args; | |
1875 | ||
a238ab5b | 1876 | va_start(args, fmt); |
3ee9a4f0 JP |
1877 | |
1878 | vaf.fmt = fmt; | |
1879 | vaf.va = &args; | |
1880 | ||
1881 | pr_warn("%pV", &vaf); | |
1882 | ||
a238ab5b DH |
1883 | va_end(args); |
1884 | } | |
1885 | ||
3ee9a4f0 JP |
1886 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
1887 | current->comm, order, gfp_mask); | |
a238ab5b DH |
1888 | |
1889 | dump_stack(); | |
1890 | if (!should_suppress_show_mem()) | |
1891 | show_mem(filter); | |
1892 | } | |
1893 | ||
11e33f6a MG |
1894 | static inline int |
1895 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 1896 | unsigned long did_some_progress, |
11e33f6a | 1897 | unsigned long pages_reclaimed) |
1da177e4 | 1898 | { |
11e33f6a MG |
1899 | /* Do not loop if specifically requested */ |
1900 | if (gfp_mask & __GFP_NORETRY) | |
1901 | return 0; | |
1da177e4 | 1902 | |
f90ac398 MG |
1903 | /* Always retry if specifically requested */ |
1904 | if (gfp_mask & __GFP_NOFAIL) | |
1905 | return 1; | |
1906 | ||
1907 | /* | |
1908 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
1909 | * making forward progress without invoking OOM. Suspend also disables | |
1910 | * storage devices so kswapd will not help. Bail if we are suspending. | |
1911 | */ | |
1912 | if (!did_some_progress && pm_suspended_storage()) | |
1913 | return 0; | |
1914 | ||
11e33f6a MG |
1915 | /* |
1916 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1917 | * means __GFP_NOFAIL, but that may not be true in other | |
1918 | * implementations. | |
1919 | */ | |
1920 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1921 | return 1; | |
1922 | ||
1923 | /* | |
1924 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1925 | * specified, then we retry until we no longer reclaim any pages | |
1926 | * (above), or we've reclaimed an order of pages at least as | |
1927 | * large as the allocation's order. In both cases, if the | |
1928 | * allocation still fails, we stop retrying. | |
1929 | */ | |
1930 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1931 | return 1; | |
cf40bd16 | 1932 | |
11e33f6a MG |
1933 | return 0; |
1934 | } | |
933e312e | 1935 | |
11e33f6a MG |
1936 | static inline struct page * |
1937 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1938 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
1939 | nodemask_t *nodemask, struct zone *preferred_zone, |
1940 | int migratetype) | |
11e33f6a MG |
1941 | { |
1942 | struct page *page; | |
1943 | ||
1944 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 1945 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 1946 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
1947 | return NULL; |
1948 | } | |
6b1de916 | 1949 | |
11e33f6a MG |
1950 | /* |
1951 | * Go through the zonelist yet one more time, keep very high watermark | |
1952 | * here, this is only to catch a parallel oom killing, we must fail if | |
1953 | * we're still under heavy pressure. | |
1954 | */ | |
1955 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1956 | order, zonelist, high_zoneidx, | |
5117f45d | 1957 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 1958 | preferred_zone, migratetype); |
7fb1d9fc | 1959 | if (page) |
11e33f6a MG |
1960 | goto out; |
1961 | ||
4365a567 KH |
1962 | if (!(gfp_mask & __GFP_NOFAIL)) { |
1963 | /* The OOM killer will not help higher order allocs */ | |
1964 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1965 | goto out; | |
03668b3c DR |
1966 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
1967 | if (high_zoneidx < ZONE_NORMAL) | |
1968 | goto out; | |
4365a567 KH |
1969 | /* |
1970 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
1971 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
1972 | * The caller should handle page allocation failure by itself if | |
1973 | * it specifies __GFP_THISNODE. | |
1974 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
1975 | */ | |
1976 | if (gfp_mask & __GFP_THISNODE) | |
1977 | goto out; | |
1978 | } | |
11e33f6a | 1979 | /* Exhausted what can be done so it's blamo time */ |
4365a567 | 1980 | out_of_memory(zonelist, gfp_mask, order, nodemask); |
11e33f6a MG |
1981 | |
1982 | out: | |
1983 | clear_zonelist_oom(zonelist, gfp_mask); | |
1984 | return page; | |
1985 | } | |
1986 | ||
56de7263 MG |
1987 | #ifdef CONFIG_COMPACTION |
1988 | /* Try memory compaction for high-order allocations before reclaim */ | |
1989 | static struct page * | |
1990 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
1991 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1992 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
77f1fe6b MG |
1993 | int migratetype, unsigned long *did_some_progress, |
1994 | bool sync_migration) | |
56de7263 MG |
1995 | { |
1996 | struct page *page; | |
1997 | ||
4f92e258 | 1998 | if (!order || compaction_deferred(preferred_zone)) |
56de7263 MG |
1999 | return NULL; |
2000 | ||
c06b1fca | 2001 | current->flags |= PF_MEMALLOC; |
56de7263 | 2002 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
77f1fe6b | 2003 | nodemask, sync_migration); |
c06b1fca | 2004 | current->flags &= ~PF_MEMALLOC; |
56de7263 MG |
2005 | if (*did_some_progress != COMPACT_SKIPPED) { |
2006 | ||
2007 | /* Page migration frees to the PCP lists but we want merging */ | |
2008 | drain_pages(get_cpu()); | |
2009 | put_cpu(); | |
2010 | ||
2011 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2012 | order, zonelist, high_zoneidx, | |
2013 | alloc_flags, preferred_zone, | |
2014 | migratetype); | |
2015 | if (page) { | |
4f92e258 MG |
2016 | preferred_zone->compact_considered = 0; |
2017 | preferred_zone->compact_defer_shift = 0; | |
56de7263 MG |
2018 | count_vm_event(COMPACTSUCCESS); |
2019 | return page; | |
2020 | } | |
2021 | ||
2022 | /* | |
2023 | * It's bad if compaction run occurs and fails. | |
2024 | * The most likely reason is that pages exist, | |
2025 | * but not enough to satisfy watermarks. | |
2026 | */ | |
2027 | count_vm_event(COMPACTFAIL); | |
4f92e258 | 2028 | defer_compaction(preferred_zone); |
56de7263 MG |
2029 | |
2030 | cond_resched(); | |
2031 | } | |
2032 | ||
2033 | return NULL; | |
2034 | } | |
2035 | #else | |
2036 | static inline struct page * | |
2037 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2038 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2039 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
77f1fe6b MG |
2040 | int migratetype, unsigned long *did_some_progress, |
2041 | bool sync_migration) | |
56de7263 MG |
2042 | { |
2043 | return NULL; | |
2044 | } | |
2045 | #endif /* CONFIG_COMPACTION */ | |
2046 | ||
11e33f6a MG |
2047 | /* The really slow allocator path where we enter direct reclaim */ |
2048 | static inline struct page * | |
2049 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2050 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
5117f45d | 2051 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
3dd28266 | 2052 | int migratetype, unsigned long *did_some_progress) |
11e33f6a MG |
2053 | { |
2054 | struct page *page = NULL; | |
2055 | struct reclaim_state reclaim_state; | |
9ee493ce | 2056 | bool drained = false; |
11e33f6a MG |
2057 | |
2058 | cond_resched(); | |
2059 | ||
2060 | /* We now go into synchronous reclaim */ | |
2061 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2062 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2063 | lockdep_set_current_reclaim_state(gfp_mask); |
2064 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2065 | current->reclaim_state = &reclaim_state; |
11e33f6a MG |
2066 | |
2067 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
2068 | ||
c06b1fca | 2069 | current->reclaim_state = NULL; |
11e33f6a | 2070 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2071 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2072 | |
2073 | cond_resched(); | |
2074 | ||
9ee493ce MG |
2075 | if (unlikely(!(*did_some_progress))) |
2076 | return NULL; | |
11e33f6a | 2077 | |
76d3fbf8 MG |
2078 | /* After successful reclaim, reconsider all zones for allocation */ |
2079 | if (NUMA_BUILD) | |
2080 | zlc_clear_zones_full(zonelist); | |
2081 | ||
9ee493ce MG |
2082 | retry: |
2083 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2084 | zonelist, high_zoneidx, |
3dd28266 MG |
2085 | alloc_flags, preferred_zone, |
2086 | migratetype); | |
9ee493ce MG |
2087 | |
2088 | /* | |
2089 | * If an allocation failed after direct reclaim, it could be because | |
2090 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2091 | */ | |
2092 | if (!page && !drained) { | |
2093 | drain_all_pages(); | |
2094 | drained = true; | |
2095 | goto retry; | |
2096 | } | |
2097 | ||
11e33f6a MG |
2098 | return page; |
2099 | } | |
2100 | ||
1da177e4 | 2101 | /* |
11e33f6a MG |
2102 | * This is called in the allocator slow-path if the allocation request is of |
2103 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2104 | */ |
11e33f6a MG |
2105 | static inline struct page * |
2106 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2107 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2108 | nodemask_t *nodemask, struct zone *preferred_zone, |
2109 | int migratetype) | |
11e33f6a MG |
2110 | { |
2111 | struct page *page; | |
2112 | ||
2113 | do { | |
2114 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2115 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2116 | preferred_zone, migratetype); |
11e33f6a MG |
2117 | |
2118 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2119 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2120 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2121 | ||
2122 | return page; | |
2123 | } | |
2124 | ||
2125 | static inline | |
2126 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
99504748 MG |
2127 | enum zone_type high_zoneidx, |
2128 | enum zone_type classzone_idx) | |
1da177e4 | 2129 | { |
dd1a239f MG |
2130 | struct zoneref *z; |
2131 | struct zone *zone; | |
1da177e4 | 2132 | |
11e33f6a | 2133 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
99504748 | 2134 | wakeup_kswapd(zone, order, classzone_idx); |
11e33f6a | 2135 | } |
cf40bd16 | 2136 | |
341ce06f PZ |
2137 | static inline int |
2138 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2139 | { | |
341ce06f PZ |
2140 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2141 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2142 | |
a56f57ff | 2143 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2144 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2145 | |
341ce06f PZ |
2146 | /* |
2147 | * The caller may dip into page reserves a bit more if the caller | |
2148 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2149 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2150 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2151 | */ | |
e6223a3b | 2152 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2153 | |
341ce06f | 2154 | if (!wait) { |
5c3240d9 AA |
2155 | /* |
2156 | * Not worth trying to allocate harder for | |
2157 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2158 | */ | |
2159 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2160 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2161 | /* |
341ce06f PZ |
2162 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2163 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2164 | */ |
341ce06f | 2165 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2166 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2167 | alloc_flags |= ALLOC_HARDER; |
2168 | ||
2169 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
2170 | if (!in_interrupt() && | |
c06b1fca | 2171 | ((current->flags & PF_MEMALLOC) || |
341ce06f PZ |
2172 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
2173 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1da177e4 | 2174 | } |
6b1de916 | 2175 | |
341ce06f PZ |
2176 | return alloc_flags; |
2177 | } | |
2178 | ||
11e33f6a MG |
2179 | static inline struct page * |
2180 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2181 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2182 | nodemask_t *nodemask, struct zone *preferred_zone, |
2183 | int migratetype) | |
11e33f6a MG |
2184 | { |
2185 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2186 | struct page *page = NULL; | |
2187 | int alloc_flags; | |
2188 | unsigned long pages_reclaimed = 0; | |
2189 | unsigned long did_some_progress; | |
77f1fe6b | 2190 | bool sync_migration = false; |
1da177e4 | 2191 | |
72807a74 MG |
2192 | /* |
2193 | * In the slowpath, we sanity check order to avoid ever trying to | |
2194 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2195 | * be using allocators in order of preference for an area that is | |
2196 | * too large. | |
2197 | */ | |
1fc28b70 MG |
2198 | if (order >= MAX_ORDER) { |
2199 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2200 | return NULL; |
1fc28b70 | 2201 | } |
1da177e4 | 2202 | |
952f3b51 CL |
2203 | /* |
2204 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2205 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2206 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2207 | * using a larger set of nodes after it has established that the | |
2208 | * allowed per node queues are empty and that nodes are | |
2209 | * over allocated. | |
2210 | */ | |
2211 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
2212 | goto nopage; | |
2213 | ||
cc4a6851 | 2214 | restart: |
32dba98e AA |
2215 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2216 | wake_all_kswapd(order, zonelist, high_zoneidx, | |
99504748 | 2217 | zone_idx(preferred_zone)); |
1da177e4 | 2218 | |
9bf2229f | 2219 | /* |
7fb1d9fc RS |
2220 | * OK, we're below the kswapd watermark and have kicked background |
2221 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2222 | * to how we want to proceed. | |
9bf2229f | 2223 | */ |
341ce06f | 2224 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2225 | |
f33261d7 DR |
2226 | /* |
2227 | * Find the true preferred zone if the allocation is unconstrained by | |
2228 | * cpusets. | |
2229 | */ | |
2230 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2231 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2232 | &preferred_zone); | |
2233 | ||
cfa54a0f | 2234 | rebalance: |
341ce06f | 2235 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2236 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2237 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2238 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2239 | if (page) |
2240 | goto got_pg; | |
1da177e4 | 2241 | |
11e33f6a | 2242 | /* Allocate without watermarks if the context allows */ |
341ce06f PZ |
2243 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
2244 | page = __alloc_pages_high_priority(gfp_mask, order, | |
2245 | zonelist, high_zoneidx, nodemask, | |
2246 | preferred_zone, migratetype); | |
2247 | if (page) | |
2248 | goto got_pg; | |
1da177e4 LT |
2249 | } |
2250 | ||
2251 | /* Atomic allocations - we can't balance anything */ | |
2252 | if (!wait) | |
2253 | goto nopage; | |
2254 | ||
341ce06f | 2255 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2256 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2257 | goto nopage; |
2258 | ||
6583bb64 DR |
2259 | /* Avoid allocations with no watermarks from looping endlessly */ |
2260 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2261 | goto nopage; | |
2262 | ||
77f1fe6b MG |
2263 | /* |
2264 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2265 | * attempts after direct reclaim are synchronous | |
2266 | */ | |
56de7263 MG |
2267 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2268 | zonelist, high_zoneidx, | |
2269 | nodemask, | |
2270 | alloc_flags, preferred_zone, | |
77f1fe6b MG |
2271 | migratetype, &did_some_progress, |
2272 | sync_migration); | |
56de7263 MG |
2273 | if (page) |
2274 | goto got_pg; | |
c6a140bf | 2275 | sync_migration = true; |
56de7263 | 2276 | |
11e33f6a MG |
2277 | /* Try direct reclaim and then allocating */ |
2278 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2279 | zonelist, high_zoneidx, | |
2280 | nodemask, | |
5117f45d | 2281 | alloc_flags, preferred_zone, |
3dd28266 | 2282 | migratetype, &did_some_progress); |
11e33f6a MG |
2283 | if (page) |
2284 | goto got_pg; | |
1da177e4 | 2285 | |
e33c3b5e | 2286 | /* |
11e33f6a MG |
2287 | * If we failed to make any progress reclaiming, then we are |
2288 | * running out of options and have to consider going OOM | |
e33c3b5e | 2289 | */ |
11e33f6a MG |
2290 | if (!did_some_progress) { |
2291 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
2292 | if (oom_killer_disabled) |
2293 | goto nopage; | |
11e33f6a MG |
2294 | page = __alloc_pages_may_oom(gfp_mask, order, |
2295 | zonelist, high_zoneidx, | |
3dd28266 MG |
2296 | nodemask, preferred_zone, |
2297 | migratetype); | |
11e33f6a MG |
2298 | if (page) |
2299 | goto got_pg; | |
1da177e4 | 2300 | |
03668b3c DR |
2301 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2302 | /* | |
2303 | * The oom killer is not called for high-order | |
2304 | * allocations that may fail, so if no progress | |
2305 | * is being made, there are no other options and | |
2306 | * retrying is unlikely to help. | |
2307 | */ | |
2308 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2309 | goto nopage; | |
2310 | /* | |
2311 | * The oom killer is not called for lowmem | |
2312 | * allocations to prevent needlessly killing | |
2313 | * innocent tasks. | |
2314 | */ | |
2315 | if (high_zoneidx < ZONE_NORMAL) | |
2316 | goto nopage; | |
2317 | } | |
e2c55dc8 | 2318 | |
ff0ceb9d DR |
2319 | goto restart; |
2320 | } | |
1da177e4 LT |
2321 | } |
2322 | ||
11e33f6a | 2323 | /* Check if we should retry the allocation */ |
a41f24ea | 2324 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2325 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2326 | pages_reclaimed)) { | |
11e33f6a | 2327 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2328 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2329 | goto rebalance; |
3e7d3449 MG |
2330 | } else { |
2331 | /* | |
2332 | * High-order allocations do not necessarily loop after | |
2333 | * direct reclaim and reclaim/compaction depends on compaction | |
2334 | * being called after reclaim so call directly if necessary | |
2335 | */ | |
2336 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2337 | zonelist, high_zoneidx, | |
2338 | nodemask, | |
2339 | alloc_flags, preferred_zone, | |
77f1fe6b MG |
2340 | migratetype, &did_some_progress, |
2341 | sync_migration); | |
3e7d3449 MG |
2342 | if (page) |
2343 | goto got_pg; | |
1da177e4 LT |
2344 | } |
2345 | ||
2346 | nopage: | |
a238ab5b | 2347 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2348 | return page; |
1da177e4 | 2349 | got_pg: |
b1eeab67 VN |
2350 | if (kmemcheck_enabled) |
2351 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1da177e4 | 2352 | return page; |
11e33f6a | 2353 | |
1da177e4 | 2354 | } |
11e33f6a MG |
2355 | |
2356 | /* | |
2357 | * This is the 'heart' of the zoned buddy allocator. | |
2358 | */ | |
2359 | struct page * | |
2360 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2361 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2362 | { | |
2363 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2364 | struct zone *preferred_zone; |
11e33f6a | 2365 | struct page *page; |
3dd28266 | 2366 | int migratetype = allocflags_to_migratetype(gfp_mask); |
11e33f6a | 2367 | |
dcce284a BH |
2368 | gfp_mask &= gfp_allowed_mask; |
2369 | ||
11e33f6a MG |
2370 | lockdep_trace_alloc(gfp_mask); |
2371 | ||
2372 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2373 | ||
2374 | if (should_fail_alloc_page(gfp_mask, order)) | |
2375 | return NULL; | |
2376 | ||
2377 | /* | |
2378 | * Check the zones suitable for the gfp_mask contain at least one | |
2379 | * valid zone. It's possible to have an empty zonelist as a result | |
2380 | * of GFP_THISNODE and a memoryless node | |
2381 | */ | |
2382 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2383 | return NULL; | |
2384 | ||
c0ff7453 | 2385 | get_mems_allowed(); |
5117f45d | 2386 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2387 | first_zones_zonelist(zonelist, high_zoneidx, |
2388 | nodemask ? : &cpuset_current_mems_allowed, | |
2389 | &preferred_zone); | |
c0ff7453 MX |
2390 | if (!preferred_zone) { |
2391 | put_mems_allowed(); | |
5117f45d | 2392 | return NULL; |
c0ff7453 | 2393 | } |
5117f45d MG |
2394 | |
2395 | /* First allocation attempt */ | |
11e33f6a | 2396 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
5117f45d | 2397 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
3dd28266 | 2398 | preferred_zone, migratetype); |
11e33f6a MG |
2399 | if (unlikely(!page)) |
2400 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 2401 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2402 | preferred_zone, migratetype); |
c0ff7453 | 2403 | put_mems_allowed(); |
11e33f6a | 2404 | |
4b4f278c | 2405 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
11e33f6a | 2406 | return page; |
1da177e4 | 2407 | } |
d239171e | 2408 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2409 | |
2410 | /* | |
2411 | * Common helper functions. | |
2412 | */ | |
920c7a5d | 2413 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2414 | { |
945a1113 AM |
2415 | struct page *page; |
2416 | ||
2417 | /* | |
2418 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2419 | * a highmem page | |
2420 | */ | |
2421 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2422 | ||
1da177e4 LT |
2423 | page = alloc_pages(gfp_mask, order); |
2424 | if (!page) | |
2425 | return 0; | |
2426 | return (unsigned long) page_address(page); | |
2427 | } | |
1da177e4 LT |
2428 | EXPORT_SYMBOL(__get_free_pages); |
2429 | ||
920c7a5d | 2430 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2431 | { |
945a1113 | 2432 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2433 | } |
1da177e4 LT |
2434 | EXPORT_SYMBOL(get_zeroed_page); |
2435 | ||
920c7a5d | 2436 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2437 | { |
b5810039 | 2438 | if (put_page_testzero(page)) { |
1da177e4 | 2439 | if (order == 0) |
fc91668e | 2440 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2441 | else |
2442 | __free_pages_ok(page, order); | |
2443 | } | |
2444 | } | |
2445 | ||
2446 | EXPORT_SYMBOL(__free_pages); | |
2447 | ||
920c7a5d | 2448 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2449 | { |
2450 | if (addr != 0) { | |
725d704e | 2451 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2452 | __free_pages(virt_to_page((void *)addr), order); |
2453 | } | |
2454 | } | |
2455 | ||
2456 | EXPORT_SYMBOL(free_pages); | |
2457 | ||
ee85c2e1 AK |
2458 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2459 | { | |
2460 | if (addr) { | |
2461 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2462 | unsigned long used = addr + PAGE_ALIGN(size); | |
2463 | ||
2464 | split_page(virt_to_page((void *)addr), order); | |
2465 | while (used < alloc_end) { | |
2466 | free_page(used); | |
2467 | used += PAGE_SIZE; | |
2468 | } | |
2469 | } | |
2470 | return (void *)addr; | |
2471 | } | |
2472 | ||
2be0ffe2 TT |
2473 | /** |
2474 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2475 | * @size: the number of bytes to allocate | |
2476 | * @gfp_mask: GFP flags for the allocation | |
2477 | * | |
2478 | * This function is similar to alloc_pages(), except that it allocates the | |
2479 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2480 | * allocate memory in power-of-two pages. | |
2481 | * | |
2482 | * This function is also limited by MAX_ORDER. | |
2483 | * | |
2484 | * Memory allocated by this function must be released by free_pages_exact(). | |
2485 | */ | |
2486 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2487 | { | |
2488 | unsigned int order = get_order(size); | |
2489 | unsigned long addr; | |
2490 | ||
2491 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2492 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2493 | } |
2494 | EXPORT_SYMBOL(alloc_pages_exact); | |
2495 | ||
ee85c2e1 AK |
2496 | /** |
2497 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2498 | * pages on a node. | |
b5e6ab58 | 2499 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2500 | * @size: the number of bytes to allocate |
2501 | * @gfp_mask: GFP flags for the allocation | |
2502 | * | |
2503 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2504 | * back. | |
2505 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2506 | * but is not exact. | |
2507 | */ | |
2508 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2509 | { | |
2510 | unsigned order = get_order(size); | |
2511 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2512 | if (!p) | |
2513 | return NULL; | |
2514 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2515 | } | |
2516 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2517 | ||
2be0ffe2 TT |
2518 | /** |
2519 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2520 | * @virt: the value returned by alloc_pages_exact. | |
2521 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2522 | * | |
2523 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2524 | */ | |
2525 | void free_pages_exact(void *virt, size_t size) | |
2526 | { | |
2527 | unsigned long addr = (unsigned long)virt; | |
2528 | unsigned long end = addr + PAGE_ALIGN(size); | |
2529 | ||
2530 | while (addr < end) { | |
2531 | free_page(addr); | |
2532 | addr += PAGE_SIZE; | |
2533 | } | |
2534 | } | |
2535 | EXPORT_SYMBOL(free_pages_exact); | |
2536 | ||
1da177e4 LT |
2537 | static unsigned int nr_free_zone_pages(int offset) |
2538 | { | |
dd1a239f | 2539 | struct zoneref *z; |
54a6eb5c MG |
2540 | struct zone *zone; |
2541 | ||
e310fd43 | 2542 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2543 | unsigned int sum = 0; |
2544 | ||
0e88460d | 2545 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2546 | |
54a6eb5c | 2547 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2548 | unsigned long size = zone->present_pages; |
41858966 | 2549 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2550 | if (size > high) |
2551 | sum += size - high; | |
1da177e4 LT |
2552 | } |
2553 | ||
2554 | return sum; | |
2555 | } | |
2556 | ||
2557 | /* | |
2558 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2559 | */ | |
2560 | unsigned int nr_free_buffer_pages(void) | |
2561 | { | |
af4ca457 | 2562 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2563 | } |
c2f1a551 | 2564 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2565 | |
2566 | /* | |
2567 | * Amount of free RAM allocatable within all zones | |
2568 | */ | |
2569 | unsigned int nr_free_pagecache_pages(void) | |
2570 | { | |
2a1e274a | 2571 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2572 | } |
08e0f6a9 CL |
2573 | |
2574 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2575 | { |
08e0f6a9 | 2576 | if (NUMA_BUILD) |
25ba77c1 | 2577 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2578 | } |
1da177e4 | 2579 | |
1da177e4 LT |
2580 | void si_meminfo(struct sysinfo *val) |
2581 | { | |
2582 | val->totalram = totalram_pages; | |
2583 | val->sharedram = 0; | |
d23ad423 | 2584 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2585 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2586 | val->totalhigh = totalhigh_pages; |
2587 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2588 | val->mem_unit = PAGE_SIZE; |
2589 | } | |
2590 | ||
2591 | EXPORT_SYMBOL(si_meminfo); | |
2592 | ||
2593 | #ifdef CONFIG_NUMA | |
2594 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2595 | { | |
2596 | pg_data_t *pgdat = NODE_DATA(nid); | |
2597 | ||
2598 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2599 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2600 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2601 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2602 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2603 | NR_FREE_PAGES); | |
98d2b0eb CL |
2604 | #else |
2605 | val->totalhigh = 0; | |
2606 | val->freehigh = 0; | |
2607 | #endif | |
1da177e4 LT |
2608 | val->mem_unit = PAGE_SIZE; |
2609 | } | |
2610 | #endif | |
2611 | ||
ddd588b5 | 2612 | /* |
7bf02ea2 DR |
2613 | * Determine whether the node should be displayed or not, depending on whether |
2614 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 2615 | */ |
7bf02ea2 | 2616 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
2617 | { |
2618 | bool ret = false; | |
2619 | ||
2620 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
2621 | goto out; | |
2622 | ||
2623 | get_mems_allowed(); | |
7bf02ea2 | 2624 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
ddd588b5 DR |
2625 | put_mems_allowed(); |
2626 | out: | |
2627 | return ret; | |
2628 | } | |
2629 | ||
1da177e4 LT |
2630 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
2631 | ||
2632 | /* | |
2633 | * Show free area list (used inside shift_scroll-lock stuff) | |
2634 | * We also calculate the percentage fragmentation. We do this by counting the | |
2635 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
2636 | * Suppresses nodes that are not allowed by current's cpuset if |
2637 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 2638 | */ |
7bf02ea2 | 2639 | void show_free_areas(unsigned int filter) |
1da177e4 | 2640 | { |
c7241913 | 2641 | int cpu; |
1da177e4 LT |
2642 | struct zone *zone; |
2643 | ||
ee99c71c | 2644 | for_each_populated_zone(zone) { |
7bf02ea2 | 2645 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2646 | continue; |
c7241913 JS |
2647 | show_node(zone); |
2648 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2649 | |
6b482c67 | 2650 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2651 | struct per_cpu_pageset *pageset; |
2652 | ||
99dcc3e5 | 2653 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 2654 | |
3dfa5721 CL |
2655 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2656 | cpu, pageset->pcp.high, | |
2657 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2658 | } |
2659 | } | |
2660 | ||
a731286d KM |
2661 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2662 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2663 | " unevictable:%lu" |
b76146ed | 2664 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2665 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
4b02108a | 2666 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
4f98a2fe | 2667 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2668 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2669 | global_page_state(NR_ISOLATED_ANON), |
2670 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2671 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2672 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2673 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2674 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2675 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2676 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2677 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2678 | global_page_state(NR_SLAB_RECLAIMABLE), |
2679 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2680 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2681 | global_page_state(NR_SHMEM), |
a25700a5 AM |
2682 | global_page_state(NR_PAGETABLE), |
2683 | global_page_state(NR_BOUNCE)); | |
1da177e4 | 2684 | |
ee99c71c | 2685 | for_each_populated_zone(zone) { |
1da177e4 LT |
2686 | int i; |
2687 | ||
7bf02ea2 | 2688 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2689 | continue; |
1da177e4 LT |
2690 | show_node(zone); |
2691 | printk("%s" | |
2692 | " free:%lukB" | |
2693 | " min:%lukB" | |
2694 | " low:%lukB" | |
2695 | " high:%lukB" | |
4f98a2fe RR |
2696 | " active_anon:%lukB" |
2697 | " inactive_anon:%lukB" | |
2698 | " active_file:%lukB" | |
2699 | " inactive_file:%lukB" | |
7b854121 | 2700 | " unevictable:%lukB" |
a731286d KM |
2701 | " isolated(anon):%lukB" |
2702 | " isolated(file):%lukB" | |
1da177e4 | 2703 | " present:%lukB" |
4a0aa73f KM |
2704 | " mlocked:%lukB" |
2705 | " dirty:%lukB" | |
2706 | " writeback:%lukB" | |
2707 | " mapped:%lukB" | |
4b02108a | 2708 | " shmem:%lukB" |
4a0aa73f KM |
2709 | " slab_reclaimable:%lukB" |
2710 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 2711 | " kernel_stack:%lukB" |
4a0aa73f KM |
2712 | " pagetables:%lukB" |
2713 | " unstable:%lukB" | |
2714 | " bounce:%lukB" | |
2715 | " writeback_tmp:%lukB" | |
1da177e4 LT |
2716 | " pages_scanned:%lu" |
2717 | " all_unreclaimable? %s" | |
2718 | "\n", | |
2719 | zone->name, | |
88f5acf8 | 2720 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
2721 | K(min_wmark_pages(zone)), |
2722 | K(low_wmark_pages(zone)), | |
2723 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
2724 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
2725 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2726 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2727 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 2728 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
2729 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
2730 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 2731 | K(zone->present_pages), |
4a0aa73f KM |
2732 | K(zone_page_state(zone, NR_MLOCK)), |
2733 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
2734 | K(zone_page_state(zone, NR_WRITEBACK)), | |
2735 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 2736 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
2737 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
2738 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
2739 | zone_page_state(zone, NR_KERNEL_STACK) * |
2740 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
2741 | K(zone_page_state(zone, NR_PAGETABLE)), |
2742 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
2743 | K(zone_page_state(zone, NR_BOUNCE)), | |
2744 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), | |
1da177e4 | 2745 | zone->pages_scanned, |
93e4a89a | 2746 | (zone->all_unreclaimable ? "yes" : "no") |
1da177e4 LT |
2747 | ); |
2748 | printk("lowmem_reserve[]:"); | |
2749 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2750 | printk(" %lu", zone->lowmem_reserve[i]); | |
2751 | printk("\n"); | |
2752 | } | |
2753 | ||
ee99c71c | 2754 | for_each_populated_zone(zone) { |
8f9de51a | 2755 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 2756 | |
7bf02ea2 | 2757 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2758 | continue; |
1da177e4 LT |
2759 | show_node(zone); |
2760 | printk("%s: ", zone->name); | |
1da177e4 LT |
2761 | |
2762 | spin_lock_irqsave(&zone->lock, flags); | |
2763 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
2764 | nr[order] = zone->free_area[order].nr_free; |
2765 | total += nr[order] << order; | |
1da177e4 LT |
2766 | } |
2767 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
2768 | for (order = 0; order < MAX_ORDER; order++) |
2769 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
2770 | printk("= %lukB\n", K(total)); |
2771 | } | |
2772 | ||
e6f3602d LW |
2773 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
2774 | ||
1da177e4 LT |
2775 | show_swap_cache_info(); |
2776 | } | |
2777 | ||
19770b32 MG |
2778 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
2779 | { | |
2780 | zoneref->zone = zone; | |
2781 | zoneref->zone_idx = zone_idx(zone); | |
2782 | } | |
2783 | ||
1da177e4 LT |
2784 | /* |
2785 | * Builds allocation fallback zone lists. | |
1a93205b CL |
2786 | * |
2787 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 2788 | */ |
f0c0b2b8 KH |
2789 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
2790 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 2791 | { |
1a93205b CL |
2792 | struct zone *zone; |
2793 | ||
98d2b0eb | 2794 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 2795 | zone_type++; |
02a68a5e CL |
2796 | |
2797 | do { | |
2f6726e5 | 2798 | zone_type--; |
070f8032 | 2799 | zone = pgdat->node_zones + zone_type; |
1a93205b | 2800 | if (populated_zone(zone)) { |
dd1a239f MG |
2801 | zoneref_set_zone(zone, |
2802 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 2803 | check_highest_zone(zone_type); |
1da177e4 | 2804 | } |
02a68a5e | 2805 | |
2f6726e5 | 2806 | } while (zone_type); |
070f8032 | 2807 | return nr_zones; |
1da177e4 LT |
2808 | } |
2809 | ||
f0c0b2b8 KH |
2810 | |
2811 | /* | |
2812 | * zonelist_order: | |
2813 | * 0 = automatic detection of better ordering. | |
2814 | * 1 = order by ([node] distance, -zonetype) | |
2815 | * 2 = order by (-zonetype, [node] distance) | |
2816 | * | |
2817 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2818 | * the same zonelist. So only NUMA can configure this param. | |
2819 | */ | |
2820 | #define ZONELIST_ORDER_DEFAULT 0 | |
2821 | #define ZONELIST_ORDER_NODE 1 | |
2822 | #define ZONELIST_ORDER_ZONE 2 | |
2823 | ||
2824 | /* zonelist order in the kernel. | |
2825 | * set_zonelist_order() will set this to NODE or ZONE. | |
2826 | */ | |
2827 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2828 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2829 | ||
2830 | ||
1da177e4 | 2831 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
2832 | /* The value user specified ....changed by config */ |
2833 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2834 | /* string for sysctl */ | |
2835 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2836 | char numa_zonelist_order[16] = "default"; | |
2837 | ||
2838 | /* | |
2839 | * interface for configure zonelist ordering. | |
2840 | * command line option "numa_zonelist_order" | |
2841 | * = "[dD]efault - default, automatic configuration. | |
2842 | * = "[nN]ode - order by node locality, then by zone within node | |
2843 | * = "[zZ]one - order by zone, then by locality within zone | |
2844 | */ | |
2845 | ||
2846 | static int __parse_numa_zonelist_order(char *s) | |
2847 | { | |
2848 | if (*s == 'd' || *s == 'D') { | |
2849 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2850 | } else if (*s == 'n' || *s == 'N') { | |
2851 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2852 | } else if (*s == 'z' || *s == 'Z') { | |
2853 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2854 | } else { | |
2855 | printk(KERN_WARNING | |
2856 | "Ignoring invalid numa_zonelist_order value: " | |
2857 | "%s\n", s); | |
2858 | return -EINVAL; | |
2859 | } | |
2860 | return 0; | |
2861 | } | |
2862 | ||
2863 | static __init int setup_numa_zonelist_order(char *s) | |
2864 | { | |
ecb256f8 VL |
2865 | int ret; |
2866 | ||
2867 | if (!s) | |
2868 | return 0; | |
2869 | ||
2870 | ret = __parse_numa_zonelist_order(s); | |
2871 | if (ret == 0) | |
2872 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
2873 | ||
2874 | return ret; | |
f0c0b2b8 KH |
2875 | } |
2876 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2877 | ||
2878 | /* | |
2879 | * sysctl handler for numa_zonelist_order | |
2880 | */ | |
2881 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 2882 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
2883 | loff_t *ppos) |
2884 | { | |
2885 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2886 | int ret; | |
443c6f14 | 2887 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 2888 | |
443c6f14 | 2889 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 2890 | if (write) |
443c6f14 | 2891 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 2892 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 2893 | if (ret) |
443c6f14 | 2894 | goto out; |
f0c0b2b8 KH |
2895 | if (write) { |
2896 | int oldval = user_zonelist_order; | |
2897 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2898 | /* | |
2899 | * bogus value. restore saved string | |
2900 | */ | |
2901 | strncpy((char*)table->data, saved_string, | |
2902 | NUMA_ZONELIST_ORDER_LEN); | |
2903 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
2904 | } else if (oldval != user_zonelist_order) { |
2905 | mutex_lock(&zonelists_mutex); | |
1f522509 | 2906 | build_all_zonelists(NULL); |
4eaf3f64 HL |
2907 | mutex_unlock(&zonelists_mutex); |
2908 | } | |
f0c0b2b8 | 2909 | } |
443c6f14 AK |
2910 | out: |
2911 | mutex_unlock(&zl_order_mutex); | |
2912 | return ret; | |
f0c0b2b8 KH |
2913 | } |
2914 | ||
2915 | ||
62bc62a8 | 2916 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
2917 | static int node_load[MAX_NUMNODES]; |
2918 | ||
1da177e4 | 2919 | /** |
4dc3b16b | 2920 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2921 | * @node: node whose fallback list we're appending |
2922 | * @used_node_mask: nodemask_t of already used nodes | |
2923 | * | |
2924 | * We use a number of factors to determine which is the next node that should | |
2925 | * appear on a given node's fallback list. The node should not have appeared | |
2926 | * already in @node's fallback list, and it should be the next closest node | |
2927 | * according to the distance array (which contains arbitrary distance values | |
2928 | * from each node to each node in the system), and should also prefer nodes | |
2929 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2930 | * on them otherwise. | |
2931 | * It returns -1 if no node is found. | |
2932 | */ | |
f0c0b2b8 | 2933 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2934 | { |
4cf808eb | 2935 | int n, val; |
1da177e4 LT |
2936 | int min_val = INT_MAX; |
2937 | int best_node = -1; | |
a70f7302 | 2938 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 2939 | |
4cf808eb LT |
2940 | /* Use the local node if we haven't already */ |
2941 | if (!node_isset(node, *used_node_mask)) { | |
2942 | node_set(node, *used_node_mask); | |
2943 | return node; | |
2944 | } | |
1da177e4 | 2945 | |
37b07e41 | 2946 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
2947 | |
2948 | /* Don't want a node to appear more than once */ | |
2949 | if (node_isset(n, *used_node_mask)) | |
2950 | continue; | |
2951 | ||
1da177e4 LT |
2952 | /* Use the distance array to find the distance */ |
2953 | val = node_distance(node, n); | |
2954 | ||
4cf808eb LT |
2955 | /* Penalize nodes under us ("prefer the next node") */ |
2956 | val += (n < node); | |
2957 | ||
1da177e4 | 2958 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
2959 | tmp = cpumask_of_node(n); |
2960 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
2961 | val += PENALTY_FOR_NODE_WITH_CPUS; |
2962 | ||
2963 | /* Slight preference for less loaded node */ | |
2964 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2965 | val += node_load[n]; | |
2966 | ||
2967 | if (val < min_val) { | |
2968 | min_val = val; | |
2969 | best_node = n; | |
2970 | } | |
2971 | } | |
2972 | ||
2973 | if (best_node >= 0) | |
2974 | node_set(best_node, *used_node_mask); | |
2975 | ||
2976 | return best_node; | |
2977 | } | |
2978 | ||
f0c0b2b8 KH |
2979 | |
2980 | /* | |
2981 | * Build zonelists ordered by node and zones within node. | |
2982 | * This results in maximum locality--normal zone overflows into local | |
2983 | * DMA zone, if any--but risks exhausting DMA zone. | |
2984 | */ | |
2985 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2986 | { |
f0c0b2b8 | 2987 | int j; |
1da177e4 | 2988 | struct zonelist *zonelist; |
f0c0b2b8 | 2989 | |
54a6eb5c | 2990 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 2991 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
2992 | ; |
2993 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2994 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
2995 | zonelist->_zonerefs[j].zone = NULL; |
2996 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
2997 | } |
2998 | ||
523b9458 CL |
2999 | /* |
3000 | * Build gfp_thisnode zonelists | |
3001 | */ | |
3002 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3003 | { | |
523b9458 CL |
3004 | int j; |
3005 | struct zonelist *zonelist; | |
3006 | ||
54a6eb5c MG |
3007 | zonelist = &pgdat->node_zonelists[1]; |
3008 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
3009 | zonelist->_zonerefs[j].zone = NULL; |
3010 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3011 | } |
3012 | ||
f0c0b2b8 KH |
3013 | /* |
3014 | * Build zonelists ordered by zone and nodes within zones. | |
3015 | * This results in conserving DMA zone[s] until all Normal memory is | |
3016 | * exhausted, but results in overflowing to remote node while memory | |
3017 | * may still exist in local DMA zone. | |
3018 | */ | |
3019 | static int node_order[MAX_NUMNODES]; | |
3020 | ||
3021 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3022 | { | |
f0c0b2b8 KH |
3023 | int pos, j, node; |
3024 | int zone_type; /* needs to be signed */ | |
3025 | struct zone *z; | |
3026 | struct zonelist *zonelist; | |
3027 | ||
54a6eb5c MG |
3028 | zonelist = &pgdat->node_zonelists[0]; |
3029 | pos = 0; | |
3030 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3031 | for (j = 0; j < nr_nodes; j++) { | |
3032 | node = node_order[j]; | |
3033 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3034 | if (populated_zone(z)) { | |
dd1a239f MG |
3035 | zoneref_set_zone(z, |
3036 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3037 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3038 | } |
3039 | } | |
f0c0b2b8 | 3040 | } |
dd1a239f MG |
3041 | zonelist->_zonerefs[pos].zone = NULL; |
3042 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3043 | } |
3044 | ||
3045 | static int default_zonelist_order(void) | |
3046 | { | |
3047 | int nid, zone_type; | |
3048 | unsigned long low_kmem_size,total_size; | |
3049 | struct zone *z; | |
3050 | int average_size; | |
3051 | /* | |
88393161 | 3052 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3053 | * If they are really small and used heavily, the system can fall |
3054 | * into OOM very easily. | |
e325c90f | 3055 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3056 | */ |
3057 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3058 | low_kmem_size = 0; | |
3059 | total_size = 0; | |
3060 | for_each_online_node(nid) { | |
3061 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3062 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3063 | if (populated_zone(z)) { | |
3064 | if (zone_type < ZONE_NORMAL) | |
3065 | low_kmem_size += z->present_pages; | |
3066 | total_size += z->present_pages; | |
e325c90f DR |
3067 | } else if (zone_type == ZONE_NORMAL) { |
3068 | /* | |
3069 | * If any node has only lowmem, then node order | |
3070 | * is preferred to allow kernel allocations | |
3071 | * locally; otherwise, they can easily infringe | |
3072 | * on other nodes when there is an abundance of | |
3073 | * lowmem available to allocate from. | |
3074 | */ | |
3075 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3076 | } |
3077 | } | |
3078 | } | |
3079 | if (!low_kmem_size || /* there are no DMA area. */ | |
3080 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3081 | return ZONELIST_ORDER_NODE; | |
3082 | /* | |
3083 | * look into each node's config. | |
3084 | * If there is a node whose DMA/DMA32 memory is very big area on | |
3085 | * local memory, NODE_ORDER may be suitable. | |
3086 | */ | |
37b07e41 LS |
3087 | average_size = total_size / |
3088 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
3089 | for_each_online_node(nid) { |
3090 | low_kmem_size = 0; | |
3091 | total_size = 0; | |
3092 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3093 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3094 | if (populated_zone(z)) { | |
3095 | if (zone_type < ZONE_NORMAL) | |
3096 | low_kmem_size += z->present_pages; | |
3097 | total_size += z->present_pages; | |
3098 | } | |
3099 | } | |
3100 | if (low_kmem_size && | |
3101 | total_size > average_size && /* ignore small node */ | |
3102 | low_kmem_size > total_size * 70/100) | |
3103 | return ZONELIST_ORDER_NODE; | |
3104 | } | |
3105 | return ZONELIST_ORDER_ZONE; | |
3106 | } | |
3107 | ||
3108 | static void set_zonelist_order(void) | |
3109 | { | |
3110 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3111 | current_zonelist_order = default_zonelist_order(); | |
3112 | else | |
3113 | current_zonelist_order = user_zonelist_order; | |
3114 | } | |
3115 | ||
3116 | static void build_zonelists(pg_data_t *pgdat) | |
3117 | { | |
3118 | int j, node, load; | |
3119 | enum zone_type i; | |
1da177e4 | 3120 | nodemask_t used_mask; |
f0c0b2b8 KH |
3121 | int local_node, prev_node; |
3122 | struct zonelist *zonelist; | |
3123 | int order = current_zonelist_order; | |
1da177e4 LT |
3124 | |
3125 | /* initialize zonelists */ | |
523b9458 | 3126 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3127 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3128 | zonelist->_zonerefs[0].zone = NULL; |
3129 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3130 | } |
3131 | ||
3132 | /* NUMA-aware ordering of nodes */ | |
3133 | local_node = pgdat->node_id; | |
62bc62a8 | 3134 | load = nr_online_nodes; |
1da177e4 LT |
3135 | prev_node = local_node; |
3136 | nodes_clear(used_mask); | |
f0c0b2b8 | 3137 | |
f0c0b2b8 KH |
3138 | memset(node_order, 0, sizeof(node_order)); |
3139 | j = 0; | |
3140 | ||
1da177e4 | 3141 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
3142 | int distance = node_distance(local_node, node); |
3143 | ||
3144 | /* | |
3145 | * If another node is sufficiently far away then it is better | |
3146 | * to reclaim pages in a zone before going off node. | |
3147 | */ | |
3148 | if (distance > RECLAIM_DISTANCE) | |
3149 | zone_reclaim_mode = 1; | |
3150 | ||
1da177e4 LT |
3151 | /* |
3152 | * We don't want to pressure a particular node. | |
3153 | * So adding penalty to the first node in same | |
3154 | * distance group to make it round-robin. | |
3155 | */ | |
9eeff239 | 3156 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
3157 | node_load[node] = load; |
3158 | ||
1da177e4 LT |
3159 | prev_node = node; |
3160 | load--; | |
f0c0b2b8 KH |
3161 | if (order == ZONELIST_ORDER_NODE) |
3162 | build_zonelists_in_node_order(pgdat, node); | |
3163 | else | |
3164 | node_order[j++] = node; /* remember order */ | |
3165 | } | |
1da177e4 | 3166 | |
f0c0b2b8 KH |
3167 | if (order == ZONELIST_ORDER_ZONE) { |
3168 | /* calculate node order -- i.e., DMA last! */ | |
3169 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3170 | } |
523b9458 CL |
3171 | |
3172 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3173 | } |
3174 | ||
9276b1bc | 3175 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3176 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3177 | { |
54a6eb5c MG |
3178 | struct zonelist *zonelist; |
3179 | struct zonelist_cache *zlc; | |
dd1a239f | 3180 | struct zoneref *z; |
9276b1bc | 3181 | |
54a6eb5c MG |
3182 | zonelist = &pgdat->node_zonelists[0]; |
3183 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3184 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3185 | for (z = zonelist->_zonerefs; z->zone; z++) |
3186 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3187 | } |
3188 | ||
7aac7898 LS |
3189 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3190 | /* | |
3191 | * Return node id of node used for "local" allocations. | |
3192 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3193 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3194 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3195 | */ | |
3196 | int local_memory_node(int node) | |
3197 | { | |
3198 | struct zone *zone; | |
3199 | ||
3200 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3201 | gfp_zone(GFP_KERNEL), | |
3202 | NULL, | |
3203 | &zone); | |
3204 | return zone->node; | |
3205 | } | |
3206 | #endif | |
f0c0b2b8 | 3207 | |
1da177e4 LT |
3208 | #else /* CONFIG_NUMA */ |
3209 | ||
f0c0b2b8 KH |
3210 | static void set_zonelist_order(void) |
3211 | { | |
3212 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3213 | } | |
3214 | ||
3215 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3216 | { |
19655d34 | 3217 | int node, local_node; |
54a6eb5c MG |
3218 | enum zone_type j; |
3219 | struct zonelist *zonelist; | |
1da177e4 LT |
3220 | |
3221 | local_node = pgdat->node_id; | |
1da177e4 | 3222 | |
54a6eb5c MG |
3223 | zonelist = &pgdat->node_zonelists[0]; |
3224 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 3225 | |
54a6eb5c MG |
3226 | /* |
3227 | * Now we build the zonelist so that it contains the zones | |
3228 | * of all the other nodes. | |
3229 | * We don't want to pressure a particular node, so when | |
3230 | * building the zones for node N, we make sure that the | |
3231 | * zones coming right after the local ones are those from | |
3232 | * node N+1 (modulo N) | |
3233 | */ | |
3234 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3235 | if (!node_online(node)) | |
3236 | continue; | |
3237 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3238 | MAX_NR_ZONES - 1); | |
1da177e4 | 3239 | } |
54a6eb5c MG |
3240 | for (node = 0; node < local_node; node++) { |
3241 | if (!node_online(node)) | |
3242 | continue; | |
3243 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3244 | MAX_NR_ZONES - 1); | |
3245 | } | |
3246 | ||
dd1a239f MG |
3247 | zonelist->_zonerefs[j].zone = NULL; |
3248 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3249 | } |
3250 | ||
9276b1bc | 3251 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3252 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3253 | { |
54a6eb5c | 3254 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3255 | } |
3256 | ||
1da177e4 LT |
3257 | #endif /* CONFIG_NUMA */ |
3258 | ||
99dcc3e5 CL |
3259 | /* |
3260 | * Boot pageset table. One per cpu which is going to be used for all | |
3261 | * zones and all nodes. The parameters will be set in such a way | |
3262 | * that an item put on a list will immediately be handed over to | |
3263 | * the buddy list. This is safe since pageset manipulation is done | |
3264 | * with interrupts disabled. | |
3265 | * | |
3266 | * The boot_pagesets must be kept even after bootup is complete for | |
3267 | * unused processors and/or zones. They do play a role for bootstrapping | |
3268 | * hotplugged processors. | |
3269 | * | |
3270 | * zoneinfo_show() and maybe other functions do | |
3271 | * not check if the processor is online before following the pageset pointer. | |
3272 | * Other parts of the kernel may not check if the zone is available. | |
3273 | */ | |
3274 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3275 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3276 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3277 | |
4eaf3f64 HL |
3278 | /* |
3279 | * Global mutex to protect against size modification of zonelists | |
3280 | * as well as to serialize pageset setup for the new populated zone. | |
3281 | */ | |
3282 | DEFINE_MUTEX(zonelists_mutex); | |
3283 | ||
9b1a4d38 | 3284 | /* return values int ....just for stop_machine() */ |
1f522509 | 3285 | static __init_refok int __build_all_zonelists(void *data) |
1da177e4 | 3286 | { |
6811378e | 3287 | int nid; |
99dcc3e5 | 3288 | int cpu; |
9276b1bc | 3289 | |
7f9cfb31 BL |
3290 | #ifdef CONFIG_NUMA |
3291 | memset(node_load, 0, sizeof(node_load)); | |
3292 | #endif | |
9276b1bc | 3293 | for_each_online_node(nid) { |
7ea1530a CL |
3294 | pg_data_t *pgdat = NODE_DATA(nid); |
3295 | ||
3296 | build_zonelists(pgdat); | |
3297 | build_zonelist_cache(pgdat); | |
9276b1bc | 3298 | } |
99dcc3e5 CL |
3299 | |
3300 | /* | |
3301 | * Initialize the boot_pagesets that are going to be used | |
3302 | * for bootstrapping processors. The real pagesets for | |
3303 | * each zone will be allocated later when the per cpu | |
3304 | * allocator is available. | |
3305 | * | |
3306 | * boot_pagesets are used also for bootstrapping offline | |
3307 | * cpus if the system is already booted because the pagesets | |
3308 | * are needed to initialize allocators on a specific cpu too. | |
3309 | * F.e. the percpu allocator needs the page allocator which | |
3310 | * needs the percpu allocator in order to allocate its pagesets | |
3311 | * (a chicken-egg dilemma). | |
3312 | */ | |
7aac7898 | 3313 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3314 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3315 | ||
7aac7898 LS |
3316 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3317 | /* | |
3318 | * We now know the "local memory node" for each node-- | |
3319 | * i.e., the node of the first zone in the generic zonelist. | |
3320 | * Set up numa_mem percpu variable for on-line cpus. During | |
3321 | * boot, only the boot cpu should be on-line; we'll init the | |
3322 | * secondary cpus' numa_mem as they come on-line. During | |
3323 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3324 | */ | |
3325 | if (cpu_online(cpu)) | |
3326 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3327 | #endif | |
3328 | } | |
3329 | ||
6811378e YG |
3330 | return 0; |
3331 | } | |
3332 | ||
4eaf3f64 HL |
3333 | /* |
3334 | * Called with zonelists_mutex held always | |
3335 | * unless system_state == SYSTEM_BOOTING. | |
3336 | */ | |
9f6ae448 | 3337 | void __ref build_all_zonelists(void *data) |
6811378e | 3338 | { |
f0c0b2b8 KH |
3339 | set_zonelist_order(); |
3340 | ||
6811378e | 3341 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3342 | __build_all_zonelists(NULL); |
68ad8df4 | 3343 | mminit_verify_zonelist(); |
6811378e YG |
3344 | cpuset_init_current_mems_allowed(); |
3345 | } else { | |
183ff22b | 3346 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 3347 | of zonelist */ |
e9959f0f KH |
3348 | #ifdef CONFIG_MEMORY_HOTPLUG |
3349 | if (data) | |
3350 | setup_zone_pageset((struct zone *)data); | |
3351 | #endif | |
3352 | stop_machine(__build_all_zonelists, NULL, NULL); | |
6811378e YG |
3353 | /* cpuset refresh routine should be here */ |
3354 | } | |
bd1e22b8 | 3355 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3356 | /* |
3357 | * Disable grouping by mobility if the number of pages in the | |
3358 | * system is too low to allow the mechanism to work. It would be | |
3359 | * more accurate, but expensive to check per-zone. This check is | |
3360 | * made on memory-hotadd so a system can start with mobility | |
3361 | * disabled and enable it later | |
3362 | */ | |
d9c23400 | 3363 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3364 | page_group_by_mobility_disabled = 1; |
3365 | else | |
3366 | page_group_by_mobility_disabled = 0; | |
3367 | ||
3368 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3369 | "Total pages: %ld\n", | |
62bc62a8 | 3370 | nr_online_nodes, |
f0c0b2b8 | 3371 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3372 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3373 | vm_total_pages); |
3374 | #ifdef CONFIG_NUMA | |
3375 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3376 | #endif | |
1da177e4 LT |
3377 | } |
3378 | ||
3379 | /* | |
3380 | * Helper functions to size the waitqueue hash table. | |
3381 | * Essentially these want to choose hash table sizes sufficiently | |
3382 | * large so that collisions trying to wait on pages are rare. | |
3383 | * But in fact, the number of active page waitqueues on typical | |
3384 | * systems is ridiculously low, less than 200. So this is even | |
3385 | * conservative, even though it seems large. | |
3386 | * | |
3387 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3388 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3389 | */ | |
3390 | #define PAGES_PER_WAITQUEUE 256 | |
3391 | ||
cca448fe | 3392 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3393 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3394 | { |
3395 | unsigned long size = 1; | |
3396 | ||
3397 | pages /= PAGES_PER_WAITQUEUE; | |
3398 | ||
3399 | while (size < pages) | |
3400 | size <<= 1; | |
3401 | ||
3402 | /* | |
3403 | * Once we have dozens or even hundreds of threads sleeping | |
3404 | * on IO we've got bigger problems than wait queue collision. | |
3405 | * Limit the size of the wait table to a reasonable size. | |
3406 | */ | |
3407 | size = min(size, 4096UL); | |
3408 | ||
3409 | return max(size, 4UL); | |
3410 | } | |
cca448fe YG |
3411 | #else |
3412 | /* | |
3413 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3414 | * a suitable size for its wait_table. So we use the maximum size now. | |
3415 | * | |
3416 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3417 | * | |
3418 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3419 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3420 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3421 | * | |
3422 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3423 | * or more by the traditional way. (See above). It equals: | |
3424 | * | |
3425 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3426 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3427 | * powerpc (64K page size) : = (32G +16M)byte. | |
3428 | */ | |
3429 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3430 | { | |
3431 | return 4096UL; | |
3432 | } | |
3433 | #endif | |
1da177e4 LT |
3434 | |
3435 | /* | |
3436 | * This is an integer logarithm so that shifts can be used later | |
3437 | * to extract the more random high bits from the multiplicative | |
3438 | * hash function before the remainder is taken. | |
3439 | */ | |
3440 | static inline unsigned long wait_table_bits(unsigned long size) | |
3441 | { | |
3442 | return ffz(~size); | |
3443 | } | |
3444 | ||
3445 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
3446 | ||
6d3163ce AH |
3447 | /* |
3448 | * Check if a pageblock contains reserved pages | |
3449 | */ | |
3450 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3451 | { | |
3452 | unsigned long pfn; | |
3453 | ||
3454 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3455 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3456 | return 1; | |
3457 | } | |
3458 | return 0; | |
3459 | } | |
3460 | ||
56fd56b8 | 3461 | /* |
d9c23400 | 3462 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3463 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3464 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3465 | * higher will lead to a bigger reserve which will get freed as contiguous |
3466 | * blocks as reclaim kicks in | |
3467 | */ | |
3468 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3469 | { | |
6d3163ce | 3470 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3471 | struct page *page; |
78986a67 MG |
3472 | unsigned long block_migratetype; |
3473 | int reserve; | |
56fd56b8 | 3474 | |
d0215638 MH |
3475 | /* |
3476 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3477 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3478 | * make sure that we always check pfn_valid for the first page in | |
3479 | * the block. | |
3480 | */ | |
56fd56b8 MG |
3481 | start_pfn = zone->zone_start_pfn; |
3482 | end_pfn = start_pfn + zone->spanned_pages; | |
d0215638 | 3483 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3484 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3485 | pageblock_order; |
56fd56b8 | 3486 | |
78986a67 MG |
3487 | /* |
3488 | * Reserve blocks are generally in place to help high-order atomic | |
3489 | * allocations that are short-lived. A min_free_kbytes value that | |
3490 | * would result in more than 2 reserve blocks for atomic allocations | |
3491 | * is assumed to be in place to help anti-fragmentation for the | |
3492 | * future allocation of hugepages at runtime. | |
3493 | */ | |
3494 | reserve = min(2, reserve); | |
3495 | ||
d9c23400 | 3496 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3497 | if (!pfn_valid(pfn)) |
3498 | continue; | |
3499 | page = pfn_to_page(pfn); | |
3500 | ||
344c790e AL |
3501 | /* Watch out for overlapping nodes */ |
3502 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3503 | continue; | |
3504 | ||
56fd56b8 MG |
3505 | block_migratetype = get_pageblock_migratetype(page); |
3506 | ||
938929f1 MG |
3507 | /* Only test what is necessary when the reserves are not met */ |
3508 | if (reserve > 0) { | |
3509 | /* | |
3510 | * Blocks with reserved pages will never free, skip | |
3511 | * them. | |
3512 | */ | |
3513 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
3514 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
3515 | continue; | |
56fd56b8 | 3516 | |
938929f1 MG |
3517 | /* If this block is reserved, account for it */ |
3518 | if (block_migratetype == MIGRATE_RESERVE) { | |
3519 | reserve--; | |
3520 | continue; | |
3521 | } | |
3522 | ||
3523 | /* Suitable for reserving if this block is movable */ | |
3524 | if (block_migratetype == MIGRATE_MOVABLE) { | |
3525 | set_pageblock_migratetype(page, | |
3526 | MIGRATE_RESERVE); | |
3527 | move_freepages_block(zone, page, | |
3528 | MIGRATE_RESERVE); | |
3529 | reserve--; | |
3530 | continue; | |
3531 | } | |
56fd56b8 MG |
3532 | } |
3533 | ||
3534 | /* | |
3535 | * If the reserve is met and this is a previous reserved block, | |
3536 | * take it back | |
3537 | */ | |
3538 | if (block_migratetype == MIGRATE_RESERVE) { | |
3539 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3540 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3541 | } | |
3542 | } | |
3543 | } | |
ac0e5b7a | 3544 | |
1da177e4 LT |
3545 | /* |
3546 | * Initially all pages are reserved - free ones are freed | |
3547 | * up by free_all_bootmem() once the early boot process is | |
3548 | * done. Non-atomic initialization, single-pass. | |
3549 | */ | |
c09b4240 | 3550 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3551 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3552 | { |
1da177e4 | 3553 | struct page *page; |
29751f69 AW |
3554 | unsigned long end_pfn = start_pfn + size; |
3555 | unsigned long pfn; | |
86051ca5 | 3556 | struct zone *z; |
1da177e4 | 3557 | |
22b31eec HD |
3558 | if (highest_memmap_pfn < end_pfn - 1) |
3559 | highest_memmap_pfn = end_pfn - 1; | |
3560 | ||
86051ca5 | 3561 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3562 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3563 | /* |
3564 | * There can be holes in boot-time mem_map[]s | |
3565 | * handed to this function. They do not | |
3566 | * exist on hotplugged memory. | |
3567 | */ | |
3568 | if (context == MEMMAP_EARLY) { | |
3569 | if (!early_pfn_valid(pfn)) | |
3570 | continue; | |
3571 | if (!early_pfn_in_nid(pfn, nid)) | |
3572 | continue; | |
3573 | } | |
d41dee36 AW |
3574 | page = pfn_to_page(pfn); |
3575 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 3576 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 3577 | init_page_count(page); |
1da177e4 LT |
3578 | reset_page_mapcount(page); |
3579 | SetPageReserved(page); | |
b2a0ac88 MG |
3580 | /* |
3581 | * Mark the block movable so that blocks are reserved for | |
3582 | * movable at startup. This will force kernel allocations | |
3583 | * to reserve their blocks rather than leaking throughout | |
3584 | * the address space during boot when many long-lived | |
56fd56b8 MG |
3585 | * kernel allocations are made. Later some blocks near |
3586 | * the start are marked MIGRATE_RESERVE by | |
3587 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
3588 | * |
3589 | * bitmap is created for zone's valid pfn range. but memmap | |
3590 | * can be created for invalid pages (for alignment) | |
3591 | * check here not to call set_pageblock_migratetype() against | |
3592 | * pfn out of zone. | |
b2a0ac88 | 3593 | */ |
86051ca5 KH |
3594 | if ((z->zone_start_pfn <= pfn) |
3595 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
3596 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 3597 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 3598 | |
1da177e4 LT |
3599 | INIT_LIST_HEAD(&page->lru); |
3600 | #ifdef WANT_PAGE_VIRTUAL | |
3601 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3602 | if (!is_highmem_idx(zone)) | |
3212c6be | 3603 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 3604 | #endif |
1da177e4 LT |
3605 | } |
3606 | } | |
3607 | ||
1e548deb | 3608 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3609 | { |
b2a0ac88 MG |
3610 | int order, t; |
3611 | for_each_migratetype_order(order, t) { | |
3612 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3613 | zone->free_area[order].nr_free = 0; |
3614 | } | |
3615 | } | |
3616 | ||
3617 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3618 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3619 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3620 | #endif |
3621 | ||
1d6f4e60 | 3622 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3623 | { |
3a6be87f | 3624 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3625 | int batch; |
3626 | ||
3627 | /* | |
3628 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3629 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3630 | * |
3631 | * OK, so we don't know how big the cache is. So guess. | |
3632 | */ | |
3633 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3634 | if (batch * PAGE_SIZE > 512 * 1024) |
3635 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3636 | batch /= 4; /* We effectively *= 4 below */ |
3637 | if (batch < 1) | |
3638 | batch = 1; | |
3639 | ||
3640 | /* | |
0ceaacc9 NP |
3641 | * Clamp the batch to a 2^n - 1 value. Having a power |
3642 | * of 2 value was found to be more likely to have | |
3643 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3644 | * |
0ceaacc9 NP |
3645 | * For example if 2 tasks are alternately allocating |
3646 | * batches of pages, one task can end up with a lot | |
3647 | * of pages of one half of the possible page colors | |
3648 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3649 | */ |
9155203a | 3650 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3651 | |
e7c8d5c9 | 3652 | return batch; |
3a6be87f DH |
3653 | |
3654 | #else | |
3655 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3656 | * conditions. | |
3657 | * | |
3658 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3659 | * of contiguous memory as there's no hardware page translation to | |
3660 | * assemble apparent contiguous memory from discontiguous pages. | |
3661 | * | |
3662 | * Queueing large contiguous runs of pages for batching, however, | |
3663 | * causes the pages to actually be freed in smaller chunks. As there | |
3664 | * can be a significant delay between the individual batches being | |
3665 | * recycled, this leads to the once large chunks of space being | |
3666 | * fragmented and becoming unavailable for high-order allocations. | |
3667 | */ | |
3668 | return 0; | |
3669 | #endif | |
e7c8d5c9 CL |
3670 | } |
3671 | ||
b69a7288 | 3672 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3673 | { |
3674 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3675 | int migratetype; |
2caaad41 | 3676 | |
1c6fe946 MD |
3677 | memset(p, 0, sizeof(*p)); |
3678 | ||
3dfa5721 | 3679 | pcp = &p->pcp; |
2caaad41 | 3680 | pcp->count = 0; |
2caaad41 CL |
3681 | pcp->high = 6 * batch; |
3682 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3683 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3684 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3685 | } |
3686 | ||
8ad4b1fb RS |
3687 | /* |
3688 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3689 | * to the value high for the pageset p. | |
3690 | */ | |
3691 | ||
3692 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3693 | unsigned long high) | |
3694 | { | |
3695 | struct per_cpu_pages *pcp; | |
3696 | ||
3dfa5721 | 3697 | pcp = &p->pcp; |
8ad4b1fb RS |
3698 | pcp->high = high; |
3699 | pcp->batch = max(1UL, high/4); | |
3700 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3701 | pcp->batch = PAGE_SHIFT * 8; | |
3702 | } | |
3703 | ||
58c2ee40 | 3704 | static void setup_zone_pageset(struct zone *zone) |
319774e2 WF |
3705 | { |
3706 | int cpu; | |
3707 | ||
3708 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
3709 | ||
3710 | for_each_possible_cpu(cpu) { | |
3711 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
3712 | ||
3713 | setup_pageset(pcp, zone_batchsize(zone)); | |
3714 | ||
3715 | if (percpu_pagelist_fraction) | |
3716 | setup_pagelist_highmark(pcp, | |
3717 | (zone->present_pages / | |
3718 | percpu_pagelist_fraction)); | |
3719 | } | |
3720 | } | |
3721 | ||
2caaad41 | 3722 | /* |
99dcc3e5 CL |
3723 | * Allocate per cpu pagesets and initialize them. |
3724 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 3725 | */ |
99dcc3e5 | 3726 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 3727 | { |
99dcc3e5 | 3728 | struct zone *zone; |
e7c8d5c9 | 3729 | |
319774e2 WF |
3730 | for_each_populated_zone(zone) |
3731 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
3732 | } |
3733 | ||
577a32f6 | 3734 | static noinline __init_refok |
cca448fe | 3735 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
3736 | { |
3737 | int i; | |
3738 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 3739 | size_t alloc_size; |
ed8ece2e DH |
3740 | |
3741 | /* | |
3742 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3743 | * per zone. | |
3744 | */ | |
02b694de YG |
3745 | zone->wait_table_hash_nr_entries = |
3746 | wait_table_hash_nr_entries(zone_size_pages); | |
3747 | zone->wait_table_bits = | |
3748 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
3749 | alloc_size = zone->wait_table_hash_nr_entries |
3750 | * sizeof(wait_queue_head_t); | |
3751 | ||
cd94b9db | 3752 | if (!slab_is_available()) { |
cca448fe | 3753 | zone->wait_table = (wait_queue_head_t *) |
8f389a99 | 3754 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
cca448fe YG |
3755 | } else { |
3756 | /* | |
3757 | * This case means that a zone whose size was 0 gets new memory | |
3758 | * via memory hot-add. | |
3759 | * But it may be the case that a new node was hot-added. In | |
3760 | * this case vmalloc() will not be able to use this new node's | |
3761 | * memory - this wait_table must be initialized to use this new | |
3762 | * node itself as well. | |
3763 | * To use this new node's memory, further consideration will be | |
3764 | * necessary. | |
3765 | */ | |
8691f3a7 | 3766 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
3767 | } |
3768 | if (!zone->wait_table) | |
3769 | return -ENOMEM; | |
ed8ece2e | 3770 | |
02b694de | 3771 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 3772 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
3773 | |
3774 | return 0; | |
ed8ece2e DH |
3775 | } |
3776 | ||
112067f0 SL |
3777 | static int __zone_pcp_update(void *data) |
3778 | { | |
3779 | struct zone *zone = data; | |
3780 | int cpu; | |
3781 | unsigned long batch = zone_batchsize(zone), flags; | |
3782 | ||
2d30a1f6 | 3783 | for_each_possible_cpu(cpu) { |
112067f0 SL |
3784 | struct per_cpu_pageset *pset; |
3785 | struct per_cpu_pages *pcp; | |
3786 | ||
99dcc3e5 | 3787 | pset = per_cpu_ptr(zone->pageset, cpu); |
112067f0 SL |
3788 | pcp = &pset->pcp; |
3789 | ||
3790 | local_irq_save(flags); | |
5f8dcc21 | 3791 | free_pcppages_bulk(zone, pcp->count, pcp); |
112067f0 SL |
3792 | setup_pageset(pset, batch); |
3793 | local_irq_restore(flags); | |
3794 | } | |
3795 | return 0; | |
3796 | } | |
3797 | ||
3798 | void zone_pcp_update(struct zone *zone) | |
3799 | { | |
3800 | stop_machine(__zone_pcp_update, zone, NULL); | |
3801 | } | |
3802 | ||
c09b4240 | 3803 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 3804 | { |
99dcc3e5 CL |
3805 | /* |
3806 | * per cpu subsystem is not up at this point. The following code | |
3807 | * relies on the ability of the linker to provide the | |
3808 | * offset of a (static) per cpu variable into the per cpu area. | |
3809 | */ | |
3810 | zone->pageset = &boot_pageset; | |
ed8ece2e | 3811 | |
f5335c0f | 3812 | if (zone->present_pages) |
99dcc3e5 CL |
3813 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
3814 | zone->name, zone->present_pages, | |
3815 | zone_batchsize(zone)); | |
ed8ece2e DH |
3816 | } |
3817 | ||
718127cc YG |
3818 | __meminit int init_currently_empty_zone(struct zone *zone, |
3819 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
3820 | unsigned long size, |
3821 | enum memmap_context context) | |
ed8ece2e DH |
3822 | { |
3823 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
3824 | int ret; |
3825 | ret = zone_wait_table_init(zone, size); | |
3826 | if (ret) | |
3827 | return ret; | |
ed8ece2e DH |
3828 | pgdat->nr_zones = zone_idx(zone) + 1; |
3829 | ||
ed8ece2e DH |
3830 | zone->zone_start_pfn = zone_start_pfn; |
3831 | ||
708614e6 MG |
3832 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
3833 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3834 | pgdat->node_id, | |
3835 | (unsigned long)zone_idx(zone), | |
3836 | zone_start_pfn, (zone_start_pfn + size)); | |
3837 | ||
1e548deb | 3838 | zone_init_free_lists(zone); |
718127cc YG |
3839 | |
3840 | return 0; | |
ed8ece2e DH |
3841 | } |
3842 | ||
0ee332c1 | 3843 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
3844 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
3845 | /* | |
3846 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3847 | * Architectures may implement their own version but if add_active_range() | |
3848 | * was used and there are no special requirements, this is a convenient | |
3849 | * alternative | |
3850 | */ | |
f2dbcfa7 | 3851 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 3852 | { |
c13291a5 TH |
3853 | unsigned long start_pfn, end_pfn; |
3854 | int i, nid; | |
c713216d | 3855 | |
c13291a5 | 3856 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
c713216d | 3857 | if (start_pfn <= pfn && pfn < end_pfn) |
c13291a5 | 3858 | return nid; |
cc2559bc KH |
3859 | /* This is a memory hole */ |
3860 | return -1; | |
c713216d MG |
3861 | } |
3862 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3863 | ||
f2dbcfa7 KH |
3864 | int __meminit early_pfn_to_nid(unsigned long pfn) |
3865 | { | |
cc2559bc KH |
3866 | int nid; |
3867 | ||
3868 | nid = __early_pfn_to_nid(pfn); | |
3869 | if (nid >= 0) | |
3870 | return nid; | |
3871 | /* just returns 0 */ | |
3872 | return 0; | |
f2dbcfa7 KH |
3873 | } |
3874 | ||
cc2559bc KH |
3875 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
3876 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3877 | { | |
3878 | int nid; | |
3879 | ||
3880 | nid = __early_pfn_to_nid(pfn); | |
3881 | if (nid >= 0 && nid != node) | |
3882 | return false; | |
3883 | return true; | |
3884 | } | |
3885 | #endif | |
f2dbcfa7 | 3886 | |
c713216d MG |
3887 | /** |
3888 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
3889 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
3890 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
3891 | * |
3892 | * If an architecture guarantees that all ranges registered with | |
3893 | * add_active_ranges() contain no holes and may be freed, this | |
3894 | * this function may be used instead of calling free_bootmem() manually. | |
3895 | */ | |
c13291a5 | 3896 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 3897 | { |
c13291a5 TH |
3898 | unsigned long start_pfn, end_pfn; |
3899 | int i, this_nid; | |
edbe7d23 | 3900 | |
c13291a5 TH |
3901 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
3902 | start_pfn = min(start_pfn, max_low_pfn); | |
3903 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 3904 | |
c13291a5 TH |
3905 | if (start_pfn < end_pfn) |
3906 | free_bootmem_node(NODE_DATA(this_nid), | |
3907 | PFN_PHYS(start_pfn), | |
3908 | (end_pfn - start_pfn) << PAGE_SHIFT); | |
edbe7d23 | 3909 | } |
edbe7d23 | 3910 | } |
edbe7d23 | 3911 | |
08677214 YL |
3912 | int __init add_from_early_node_map(struct range *range, int az, |
3913 | int nr_range, int nid) | |
3914 | { | |
c13291a5 | 3915 | unsigned long start_pfn, end_pfn; |
08677214 | 3916 | int i; |
08677214 YL |
3917 | |
3918 | /* need to go over early_node_map to find out good range for node */ | |
c13291a5 TH |
3919 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) |
3920 | nr_range = add_range(range, az, nr_range, start_pfn, end_pfn); | |
08677214 YL |
3921 | return nr_range; |
3922 | } | |
3923 | ||
c713216d MG |
3924 | /** |
3925 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 3926 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
3927 | * |
3928 | * If an architecture guarantees that all ranges registered with | |
3929 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 3930 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
3931 | */ |
3932 | void __init sparse_memory_present_with_active_regions(int nid) | |
3933 | { | |
c13291a5 TH |
3934 | unsigned long start_pfn, end_pfn; |
3935 | int i, this_nid; | |
c713216d | 3936 | |
c13291a5 TH |
3937 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
3938 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
3939 | } |
3940 | ||
3941 | /** | |
3942 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
3943 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
3944 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3945 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
3946 | * |
3947 | * It returns the start and end page frame of a node based on information | |
3948 | * provided by an arch calling add_active_range(). If called for a node | |
3949 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 3950 | * PFNs will be 0. |
c713216d | 3951 | */ |
a3142c8e | 3952 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3953 | unsigned long *start_pfn, unsigned long *end_pfn) |
3954 | { | |
c13291a5 | 3955 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 3956 | int i; |
c13291a5 | 3957 | |
c713216d MG |
3958 | *start_pfn = -1UL; |
3959 | *end_pfn = 0; | |
3960 | ||
c13291a5 TH |
3961 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
3962 | *start_pfn = min(*start_pfn, this_start_pfn); | |
3963 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
3964 | } |
3965 | ||
633c0666 | 3966 | if (*start_pfn == -1UL) |
c713216d | 3967 | *start_pfn = 0; |
c713216d MG |
3968 | } |
3969 | ||
2a1e274a MG |
3970 | /* |
3971 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3972 | * assumption is made that zones within a node are ordered in monotonic | |
3973 | * increasing memory addresses so that the "highest" populated zone is used | |
3974 | */ | |
b69a7288 | 3975 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
3976 | { |
3977 | int zone_index; | |
3978 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3979 | if (zone_index == ZONE_MOVABLE) | |
3980 | continue; | |
3981 | ||
3982 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3983 | arch_zone_lowest_possible_pfn[zone_index]) | |
3984 | break; | |
3985 | } | |
3986 | ||
3987 | VM_BUG_ON(zone_index == -1); | |
3988 | movable_zone = zone_index; | |
3989 | } | |
3990 | ||
3991 | /* | |
3992 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 3993 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
3994 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
3995 | * in each node depending on the size of each node and how evenly kernelcore | |
3996 | * is distributed. This helper function adjusts the zone ranges | |
3997 | * provided by the architecture for a given node by using the end of the | |
3998 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3999 | * zones within a node are in order of monotonic increases memory addresses | |
4000 | */ | |
b69a7288 | 4001 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4002 | unsigned long zone_type, |
4003 | unsigned long node_start_pfn, | |
4004 | unsigned long node_end_pfn, | |
4005 | unsigned long *zone_start_pfn, | |
4006 | unsigned long *zone_end_pfn) | |
4007 | { | |
4008 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4009 | if (zone_movable_pfn[nid]) { | |
4010 | /* Size ZONE_MOVABLE */ | |
4011 | if (zone_type == ZONE_MOVABLE) { | |
4012 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4013 | *zone_end_pfn = min(node_end_pfn, | |
4014 | arch_zone_highest_possible_pfn[movable_zone]); | |
4015 | ||
4016 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4017 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4018 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4019 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4020 | ||
4021 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4022 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4023 | *zone_start_pfn = *zone_end_pfn; | |
4024 | } | |
4025 | } | |
4026 | ||
c713216d MG |
4027 | /* |
4028 | * Return the number of pages a zone spans in a node, including holes | |
4029 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4030 | */ | |
6ea6e688 | 4031 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4032 | unsigned long zone_type, |
4033 | unsigned long *ignored) | |
4034 | { | |
4035 | unsigned long node_start_pfn, node_end_pfn; | |
4036 | unsigned long zone_start_pfn, zone_end_pfn; | |
4037 | ||
4038 | /* Get the start and end of the node and zone */ | |
4039 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
4040 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
4041 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4042 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4043 | node_start_pfn, node_end_pfn, | |
4044 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4045 | |
4046 | /* Check that this node has pages within the zone's required range */ | |
4047 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4048 | return 0; | |
4049 | ||
4050 | /* Move the zone boundaries inside the node if necessary */ | |
4051 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4052 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4053 | ||
4054 | /* Return the spanned pages */ | |
4055 | return zone_end_pfn - zone_start_pfn; | |
4056 | } | |
4057 | ||
4058 | /* | |
4059 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4060 | * then all holes in the requested range will be accounted for. |
c713216d | 4061 | */ |
32996250 | 4062 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4063 | unsigned long range_start_pfn, |
4064 | unsigned long range_end_pfn) | |
4065 | { | |
96e907d1 TH |
4066 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4067 | unsigned long start_pfn, end_pfn; | |
4068 | int i; | |
c713216d | 4069 | |
96e907d1 TH |
4070 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4071 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4072 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4073 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4074 | } |
96e907d1 | 4075 | return nr_absent; |
c713216d MG |
4076 | } |
4077 | ||
4078 | /** | |
4079 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4080 | * @start_pfn: The start PFN to start searching for holes | |
4081 | * @end_pfn: The end PFN to stop searching for holes | |
4082 | * | |
88ca3b94 | 4083 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4084 | */ |
4085 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4086 | unsigned long end_pfn) | |
4087 | { | |
4088 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4089 | } | |
4090 | ||
4091 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4092 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4093 | unsigned long zone_type, |
4094 | unsigned long *ignored) | |
4095 | { | |
96e907d1 TH |
4096 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4097 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4098 | unsigned long node_start_pfn, node_end_pfn; |
4099 | unsigned long zone_start_pfn, zone_end_pfn; | |
4100 | ||
4101 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
96e907d1 TH |
4102 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4103 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4104 | |
2a1e274a MG |
4105 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4106 | node_start_pfn, node_end_pfn, | |
4107 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4108 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4109 | } |
0e0b864e | 4110 | |
0ee332c1 | 4111 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4112 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4113 | unsigned long zone_type, |
4114 | unsigned long *zones_size) | |
4115 | { | |
4116 | return zones_size[zone_type]; | |
4117 | } | |
4118 | ||
6ea6e688 | 4119 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4120 | unsigned long zone_type, |
4121 | unsigned long *zholes_size) | |
4122 | { | |
4123 | if (!zholes_size) | |
4124 | return 0; | |
4125 | ||
4126 | return zholes_size[zone_type]; | |
4127 | } | |
0e0b864e | 4128 | |
0ee332c1 | 4129 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4130 | |
a3142c8e | 4131 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
4132 | unsigned long *zones_size, unsigned long *zholes_size) |
4133 | { | |
4134 | unsigned long realtotalpages, totalpages = 0; | |
4135 | enum zone_type i; | |
4136 | ||
4137 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4138 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
4139 | zones_size); | |
4140 | pgdat->node_spanned_pages = totalpages; | |
4141 | ||
4142 | realtotalpages = totalpages; | |
4143 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4144 | realtotalpages -= | |
4145 | zone_absent_pages_in_node(pgdat->node_id, i, | |
4146 | zholes_size); | |
4147 | pgdat->node_present_pages = realtotalpages; | |
4148 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4149 | realtotalpages); | |
4150 | } | |
4151 | ||
835c134e MG |
4152 | #ifndef CONFIG_SPARSEMEM |
4153 | /* | |
4154 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4155 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4156 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4157 | * round what is now in bits to nearest long in bits, then return it in |
4158 | * bytes. | |
4159 | */ | |
4160 | static unsigned long __init usemap_size(unsigned long zonesize) | |
4161 | { | |
4162 | unsigned long usemapsize; | |
4163 | ||
d9c23400 MG |
4164 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4165 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4166 | usemapsize *= NR_PAGEBLOCK_BITS; |
4167 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4168 | ||
4169 | return usemapsize / 8; | |
4170 | } | |
4171 | ||
4172 | static void __init setup_usemap(struct pglist_data *pgdat, | |
4173 | struct zone *zone, unsigned long zonesize) | |
4174 | { | |
4175 | unsigned long usemapsize = usemap_size(zonesize); | |
4176 | zone->pageblock_flags = NULL; | |
58a01a45 | 4177 | if (usemapsize) |
8f389a99 YL |
4178 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
4179 | usemapsize); | |
835c134e MG |
4180 | } |
4181 | #else | |
fa9f90be | 4182 | static inline void setup_usemap(struct pglist_data *pgdat, |
835c134e MG |
4183 | struct zone *zone, unsigned long zonesize) {} |
4184 | #endif /* CONFIG_SPARSEMEM */ | |
4185 | ||
d9c23400 | 4186 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
4187 | |
4188 | /* Return a sensible default order for the pageblock size. */ | |
4189 | static inline int pageblock_default_order(void) | |
4190 | { | |
4191 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
4192 | return HUGETLB_PAGE_ORDER; | |
4193 | ||
4194 | return MAX_ORDER-1; | |
4195 | } | |
4196 | ||
d9c23400 MG |
4197 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
4198 | static inline void __init set_pageblock_order(unsigned int order) | |
4199 | { | |
4200 | /* Check that pageblock_nr_pages has not already been setup */ | |
4201 | if (pageblock_order) | |
4202 | return; | |
4203 | ||
4204 | /* | |
4205 | * Assume the largest contiguous order of interest is a huge page. | |
4206 | * This value may be variable depending on boot parameters on IA64 | |
4207 | */ | |
4208 | pageblock_order = order; | |
4209 | } | |
4210 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4211 | ||
ba72cb8c MG |
4212 | /* |
4213 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
4214 | * and pageblock_default_order() are unused as pageblock_order is set | |
4215 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
4216 | * pageblock_order based on the kernel config | |
4217 | */ | |
4218 | static inline int pageblock_default_order(unsigned int order) | |
4219 | { | |
4220 | return MAX_ORDER-1; | |
4221 | } | |
d9c23400 MG |
4222 | #define set_pageblock_order(x) do {} while (0) |
4223 | ||
4224 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4225 | ||
1da177e4 LT |
4226 | /* |
4227 | * Set up the zone data structures: | |
4228 | * - mark all pages reserved | |
4229 | * - mark all memory queues empty | |
4230 | * - clear the memory bitmaps | |
4231 | */ | |
b5a0e011 | 4232 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
4233 | unsigned long *zones_size, unsigned long *zholes_size) |
4234 | { | |
2f1b6248 | 4235 | enum zone_type j; |
ed8ece2e | 4236 | int nid = pgdat->node_id; |
1da177e4 | 4237 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4238 | int ret; |
1da177e4 | 4239 | |
208d54e5 | 4240 | pgdat_resize_init(pgdat); |
1da177e4 LT |
4241 | pgdat->nr_zones = 0; |
4242 | init_waitqueue_head(&pgdat->kswapd_wait); | |
4243 | pgdat->kswapd_max_order = 0; | |
52d4b9ac | 4244 | pgdat_page_cgroup_init(pgdat); |
1da177e4 LT |
4245 | |
4246 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
4247 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 4248 | unsigned long size, realsize, memmap_pages; |
b69408e8 | 4249 | enum lru_list l; |
1da177e4 | 4250 | |
c713216d MG |
4251 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
4252 | realsize = size - zone_absent_pages_in_node(nid, j, | |
4253 | zholes_size); | |
1da177e4 | 4254 | |
0e0b864e MG |
4255 | /* |
4256 | * Adjust realsize so that it accounts for how much memory | |
4257 | * is used by this zone for memmap. This affects the watermark | |
4258 | * and per-cpu initialisations | |
4259 | */ | |
f7232154 JW |
4260 | memmap_pages = |
4261 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
4262 | if (realsize >= memmap_pages) { |
4263 | realsize -= memmap_pages; | |
5594c8c8 YL |
4264 | if (memmap_pages) |
4265 | printk(KERN_DEBUG | |
4266 | " %s zone: %lu pages used for memmap\n", | |
4267 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4268 | } else |
4269 | printk(KERN_WARNING | |
4270 | " %s zone: %lu pages exceeds realsize %lu\n", | |
4271 | zone_names[j], memmap_pages, realsize); | |
4272 | ||
6267276f CL |
4273 | /* Account for reserved pages */ |
4274 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 4275 | realsize -= dma_reserve; |
d903ef9f | 4276 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4277 | zone_names[0], dma_reserve); |
0e0b864e MG |
4278 | } |
4279 | ||
98d2b0eb | 4280 | if (!is_highmem_idx(j)) |
1da177e4 LT |
4281 | nr_kernel_pages += realsize; |
4282 | nr_all_pages += realsize; | |
4283 | ||
4284 | zone->spanned_pages = size; | |
4285 | zone->present_pages = realsize; | |
9614634f | 4286 | #ifdef CONFIG_NUMA |
d5f541ed | 4287 | zone->node = nid; |
8417bba4 | 4288 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 4289 | / 100; |
0ff38490 | 4290 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 4291 | #endif |
1da177e4 LT |
4292 | zone->name = zone_names[j]; |
4293 | spin_lock_init(&zone->lock); | |
4294 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4295 | zone_seqlock_init(zone); |
1da177e4 | 4296 | zone->zone_pgdat = pgdat; |
1da177e4 | 4297 | |
ed8ece2e | 4298 | zone_pcp_init(zone); |
246e87a9 | 4299 | for_each_lru(l) |
b69408e8 | 4300 | INIT_LIST_HEAD(&zone->lru[l].list); |
6e901571 KM |
4301 | zone->reclaim_stat.recent_rotated[0] = 0; |
4302 | zone->reclaim_stat.recent_rotated[1] = 0; | |
4303 | zone->reclaim_stat.recent_scanned[0] = 0; | |
4304 | zone->reclaim_stat.recent_scanned[1] = 0; | |
2244b95a | 4305 | zap_zone_vm_stats(zone); |
e815af95 | 4306 | zone->flags = 0; |
1da177e4 LT |
4307 | if (!size) |
4308 | continue; | |
4309 | ||
ba72cb8c | 4310 | set_pageblock_order(pageblock_default_order()); |
835c134e | 4311 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
4312 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4313 | size, MEMMAP_EARLY); | |
718127cc | 4314 | BUG_ON(ret); |
76cdd58e | 4315 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4316 | zone_start_pfn += size; |
1da177e4 LT |
4317 | } |
4318 | } | |
4319 | ||
577a32f6 | 4320 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4321 | { |
1da177e4 LT |
4322 | /* Skip empty nodes */ |
4323 | if (!pgdat->node_spanned_pages) | |
4324 | return; | |
4325 | ||
d41dee36 | 4326 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4327 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4328 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4329 | unsigned long size, start, end; |
d41dee36 AW |
4330 | struct page *map; |
4331 | ||
e984bb43 BP |
4332 | /* |
4333 | * The zone's endpoints aren't required to be MAX_ORDER | |
4334 | * aligned but the node_mem_map endpoints must be in order | |
4335 | * for the buddy allocator to function correctly. | |
4336 | */ | |
4337 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
4338 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
4339 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
4340 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4341 | map = alloc_remap(pgdat->node_id, size); |
4342 | if (!map) | |
8f389a99 | 4343 | map = alloc_bootmem_node_nopanic(pgdat, size); |
e984bb43 | 4344 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4345 | } |
12d810c1 | 4346 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4347 | /* |
4348 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4349 | */ | |
c713216d | 4350 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4351 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4352 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4353 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4354 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4355 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4356 | } |
1da177e4 | 4357 | #endif |
d41dee36 | 4358 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4359 | } |
4360 | ||
9109fb7b JW |
4361 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4362 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4363 | { |
9109fb7b JW |
4364 | pg_data_t *pgdat = NODE_DATA(nid); |
4365 | ||
1da177e4 LT |
4366 | pgdat->node_id = nid; |
4367 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 4368 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
4369 | |
4370 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4371 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4372 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4373 | nid, (unsigned long)pgdat, | |
4374 | (unsigned long)pgdat->node_mem_map); | |
4375 | #endif | |
1da177e4 LT |
4376 | |
4377 | free_area_init_core(pgdat, zones_size, zholes_size); | |
4378 | } | |
4379 | ||
0ee332c1 | 4380 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4381 | |
4382 | #if MAX_NUMNODES > 1 | |
4383 | /* | |
4384 | * Figure out the number of possible node ids. | |
4385 | */ | |
4386 | static void __init setup_nr_node_ids(void) | |
4387 | { | |
4388 | unsigned int node; | |
4389 | unsigned int highest = 0; | |
4390 | ||
4391 | for_each_node_mask(node, node_possible_map) | |
4392 | highest = node; | |
4393 | nr_node_ids = highest + 1; | |
4394 | } | |
4395 | #else | |
4396 | static inline void setup_nr_node_ids(void) | |
4397 | { | |
4398 | } | |
4399 | #endif | |
4400 | ||
1e01979c TH |
4401 | /** |
4402 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4403 | * | |
4404 | * This function should be called after node map is populated and sorted. | |
4405 | * It calculates the maximum power of two alignment which can distinguish | |
4406 | * all the nodes. | |
4407 | * | |
4408 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4409 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4410 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4411 | * shifted, 1GiB is enough and this function will indicate so. | |
4412 | * | |
4413 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4414 | * model has fine enough granularity to avoid incorrect mapping for the | |
4415 | * populated node map. | |
4416 | * | |
4417 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4418 | * requirement (single node). | |
4419 | */ | |
4420 | unsigned long __init node_map_pfn_alignment(void) | |
4421 | { | |
4422 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4423 | unsigned long start, end, mask; |
1e01979c | 4424 | int last_nid = -1; |
c13291a5 | 4425 | int i, nid; |
1e01979c | 4426 | |
c13291a5 | 4427 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
4428 | if (!start || last_nid < 0 || last_nid == nid) { |
4429 | last_nid = nid; | |
4430 | last_end = end; | |
4431 | continue; | |
4432 | } | |
4433 | ||
4434 | /* | |
4435 | * Start with a mask granular enough to pin-point to the | |
4436 | * start pfn and tick off bits one-by-one until it becomes | |
4437 | * too coarse to separate the current node from the last. | |
4438 | */ | |
4439 | mask = ~((1 << __ffs(start)) - 1); | |
4440 | while (mask && last_end <= (start & (mask << 1))) | |
4441 | mask <<= 1; | |
4442 | ||
4443 | /* accumulate all internode masks */ | |
4444 | accl_mask |= mask; | |
4445 | } | |
4446 | ||
4447 | /* convert mask to number of pages */ | |
4448 | return ~accl_mask + 1; | |
4449 | } | |
4450 | ||
a6af2bc3 | 4451 | /* Find the lowest pfn for a node */ |
b69a7288 | 4452 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 4453 | { |
a6af2bc3 | 4454 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
4455 | unsigned long start_pfn; |
4456 | int i; | |
1abbfb41 | 4457 | |
c13291a5 TH |
4458 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
4459 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 4460 | |
a6af2bc3 MG |
4461 | if (min_pfn == ULONG_MAX) { |
4462 | printk(KERN_WARNING | |
2bc0d261 | 4463 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4464 | return 0; |
4465 | } | |
4466 | ||
4467 | return min_pfn; | |
c713216d MG |
4468 | } |
4469 | ||
4470 | /** | |
4471 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4472 | * | |
4473 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4474 | * add_active_range(). |
c713216d MG |
4475 | */ |
4476 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4477 | { | |
4478 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4479 | } | |
4480 | ||
37b07e41 LS |
4481 | /* |
4482 | * early_calculate_totalpages() | |
4483 | * Sum pages in active regions for movable zone. | |
4484 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4485 | */ | |
484f51f8 | 4486 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 4487 | { |
7e63efef | 4488 | unsigned long totalpages = 0; |
c13291a5 TH |
4489 | unsigned long start_pfn, end_pfn; |
4490 | int i, nid; | |
4491 | ||
4492 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4493 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 4494 | |
37b07e41 LS |
4495 | totalpages += pages; |
4496 | if (pages) | |
c13291a5 | 4497 | node_set_state(nid, N_HIGH_MEMORY); |
37b07e41 LS |
4498 | } |
4499 | return totalpages; | |
7e63efef MG |
4500 | } |
4501 | ||
2a1e274a MG |
4502 | /* |
4503 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4504 | * is spread evenly between nodes as long as the nodes have enough | |
4505 | * memory. When they don't, some nodes will have more kernelcore than | |
4506 | * others | |
4507 | */ | |
b69a7288 | 4508 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
2a1e274a MG |
4509 | { |
4510 | int i, nid; | |
4511 | unsigned long usable_startpfn; | |
4512 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4513 | /* save the state before borrow the nodemask */ |
4514 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4515 | unsigned long totalpages = early_calculate_totalpages(); |
4516 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4517 | |
7e63efef MG |
4518 | /* |
4519 | * If movablecore was specified, calculate what size of | |
4520 | * kernelcore that corresponds so that memory usable for | |
4521 | * any allocation type is evenly spread. If both kernelcore | |
4522 | * and movablecore are specified, then the value of kernelcore | |
4523 | * will be used for required_kernelcore if it's greater than | |
4524 | * what movablecore would have allowed. | |
4525 | */ | |
4526 | if (required_movablecore) { | |
7e63efef MG |
4527 | unsigned long corepages; |
4528 | ||
4529 | /* | |
4530 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4531 | * was requested by the user | |
4532 | */ | |
4533 | required_movablecore = | |
4534 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4535 | corepages = totalpages - required_movablecore; | |
4536 | ||
4537 | required_kernelcore = max(required_kernelcore, corepages); | |
4538 | } | |
4539 | ||
2a1e274a MG |
4540 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4541 | if (!required_kernelcore) | |
66918dcd | 4542 | goto out; |
2a1e274a MG |
4543 | |
4544 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4545 | find_usable_zone_for_movable(); | |
4546 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4547 | ||
4548 | restart: | |
4549 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4550 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4551 | for_each_node_state(nid, N_HIGH_MEMORY) { |
c13291a5 TH |
4552 | unsigned long start_pfn, end_pfn; |
4553 | ||
2a1e274a MG |
4554 | /* |
4555 | * Recalculate kernelcore_node if the division per node | |
4556 | * now exceeds what is necessary to satisfy the requested | |
4557 | * amount of memory for the kernel | |
4558 | */ | |
4559 | if (required_kernelcore < kernelcore_node) | |
4560 | kernelcore_node = required_kernelcore / usable_nodes; | |
4561 | ||
4562 | /* | |
4563 | * As the map is walked, we track how much memory is usable | |
4564 | * by the kernel using kernelcore_remaining. When it is | |
4565 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4566 | */ | |
4567 | kernelcore_remaining = kernelcore_node; | |
4568 | ||
4569 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 4570 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
4571 | unsigned long size_pages; |
4572 | ||
c13291a5 | 4573 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
4574 | if (start_pfn >= end_pfn) |
4575 | continue; | |
4576 | ||
4577 | /* Account for what is only usable for kernelcore */ | |
4578 | if (start_pfn < usable_startpfn) { | |
4579 | unsigned long kernel_pages; | |
4580 | kernel_pages = min(end_pfn, usable_startpfn) | |
4581 | - start_pfn; | |
4582 | ||
4583 | kernelcore_remaining -= min(kernel_pages, | |
4584 | kernelcore_remaining); | |
4585 | required_kernelcore -= min(kernel_pages, | |
4586 | required_kernelcore); | |
4587 | ||
4588 | /* Continue if range is now fully accounted */ | |
4589 | if (end_pfn <= usable_startpfn) { | |
4590 | ||
4591 | /* | |
4592 | * Push zone_movable_pfn to the end so | |
4593 | * that if we have to rebalance | |
4594 | * kernelcore across nodes, we will | |
4595 | * not double account here | |
4596 | */ | |
4597 | zone_movable_pfn[nid] = end_pfn; | |
4598 | continue; | |
4599 | } | |
4600 | start_pfn = usable_startpfn; | |
4601 | } | |
4602 | ||
4603 | /* | |
4604 | * The usable PFN range for ZONE_MOVABLE is from | |
4605 | * start_pfn->end_pfn. Calculate size_pages as the | |
4606 | * number of pages used as kernelcore | |
4607 | */ | |
4608 | size_pages = end_pfn - start_pfn; | |
4609 | if (size_pages > kernelcore_remaining) | |
4610 | size_pages = kernelcore_remaining; | |
4611 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4612 | ||
4613 | /* | |
4614 | * Some kernelcore has been met, update counts and | |
4615 | * break if the kernelcore for this node has been | |
4616 | * satisified | |
4617 | */ | |
4618 | required_kernelcore -= min(required_kernelcore, | |
4619 | size_pages); | |
4620 | kernelcore_remaining -= size_pages; | |
4621 | if (!kernelcore_remaining) | |
4622 | break; | |
4623 | } | |
4624 | } | |
4625 | ||
4626 | /* | |
4627 | * If there is still required_kernelcore, we do another pass with one | |
4628 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4629 | * along on the nodes that still have memory until kernelcore is | |
4630 | * satisified | |
4631 | */ | |
4632 | usable_nodes--; | |
4633 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4634 | goto restart; | |
4635 | ||
4636 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4637 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4638 | zone_movable_pfn[nid] = | |
4639 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4640 | |
4641 | out: | |
4642 | /* restore the node_state */ | |
4643 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4644 | } |
4645 | ||
37b07e41 LS |
4646 | /* Any regular memory on that node ? */ |
4647 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4648 | { | |
4649 | #ifdef CONFIG_HIGHMEM | |
4650 | enum zone_type zone_type; | |
4651 | ||
4652 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4653 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4654 | if (zone->present_pages) | |
4655 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4656 | } | |
4657 | #endif | |
4658 | } | |
4659 | ||
c713216d MG |
4660 | /** |
4661 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4662 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4663 | * |
4664 | * This will call free_area_init_node() for each active node in the system. | |
4665 | * Using the page ranges provided by add_active_range(), the size of each | |
4666 | * zone in each node and their holes is calculated. If the maximum PFN | |
4667 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4668 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4669 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4670 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4671 | * at arch_max_dma_pfn. | |
4672 | */ | |
4673 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4674 | { | |
c13291a5 TH |
4675 | unsigned long start_pfn, end_pfn; |
4676 | int i, nid; | |
a6af2bc3 | 4677 | |
c713216d MG |
4678 | /* Record where the zone boundaries are */ |
4679 | memset(arch_zone_lowest_possible_pfn, 0, | |
4680 | sizeof(arch_zone_lowest_possible_pfn)); | |
4681 | memset(arch_zone_highest_possible_pfn, 0, | |
4682 | sizeof(arch_zone_highest_possible_pfn)); | |
4683 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4684 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4685 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4686 | if (i == ZONE_MOVABLE) |
4687 | continue; | |
c713216d MG |
4688 | arch_zone_lowest_possible_pfn[i] = |
4689 | arch_zone_highest_possible_pfn[i-1]; | |
4690 | arch_zone_highest_possible_pfn[i] = | |
4691 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4692 | } | |
2a1e274a MG |
4693 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4694 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4695 | ||
4696 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4697 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4698 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 4699 | |
c713216d MG |
4700 | /* Print out the zone ranges */ |
4701 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
4702 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4703 | if (i == ZONE_MOVABLE) | |
4704 | continue; | |
72f0ba02 DR |
4705 | printk(" %-8s ", zone_names[i]); |
4706 | if (arch_zone_lowest_possible_pfn[i] == | |
4707 | arch_zone_highest_possible_pfn[i]) | |
4708 | printk("empty\n"); | |
4709 | else | |
4710 | printk("%0#10lx -> %0#10lx\n", | |
c713216d MG |
4711 | arch_zone_lowest_possible_pfn[i], |
4712 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
4713 | } |
4714 | ||
4715 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4716 | printk("Movable zone start PFN for each node\n"); | |
4717 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4718 | if (zone_movable_pfn[i]) | |
4719 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4720 | } | |
c713216d MG |
4721 | |
4722 | /* Print out the early_node_map[] */ | |
c13291a5 TH |
4723 | printk("Early memory PFN ranges\n"); |
4724 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) | |
4725 | printk(" %3d: %0#10lx -> %0#10lx\n", nid, start_pfn, end_pfn); | |
c713216d MG |
4726 | |
4727 | /* Initialise every node */ | |
708614e6 | 4728 | mminit_verify_pageflags_layout(); |
8ef82866 | 4729 | setup_nr_node_ids(); |
c713216d MG |
4730 | for_each_online_node(nid) { |
4731 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4732 | free_area_init_node(nid, NULL, |
c713216d | 4733 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4734 | |
4735 | /* Any memory on that node */ | |
4736 | if (pgdat->node_present_pages) | |
4737 | node_set_state(nid, N_HIGH_MEMORY); | |
4738 | check_for_regular_memory(pgdat); | |
c713216d MG |
4739 | } |
4740 | } | |
2a1e274a | 4741 | |
7e63efef | 4742 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4743 | { |
4744 | unsigned long long coremem; | |
4745 | if (!p) | |
4746 | return -EINVAL; | |
4747 | ||
4748 | coremem = memparse(p, &p); | |
7e63efef | 4749 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 4750 | |
7e63efef | 4751 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
4752 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
4753 | ||
4754 | return 0; | |
4755 | } | |
ed7ed365 | 4756 | |
7e63efef MG |
4757 | /* |
4758 | * kernelcore=size sets the amount of memory for use for allocations that | |
4759 | * cannot be reclaimed or migrated. | |
4760 | */ | |
4761 | static int __init cmdline_parse_kernelcore(char *p) | |
4762 | { | |
4763 | return cmdline_parse_core(p, &required_kernelcore); | |
4764 | } | |
4765 | ||
4766 | /* | |
4767 | * movablecore=size sets the amount of memory for use for allocations that | |
4768 | * can be reclaimed or migrated. | |
4769 | */ | |
4770 | static int __init cmdline_parse_movablecore(char *p) | |
4771 | { | |
4772 | return cmdline_parse_core(p, &required_movablecore); | |
4773 | } | |
4774 | ||
ed7ed365 | 4775 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 4776 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 4777 | |
0ee332c1 | 4778 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4779 | |
0e0b864e | 4780 | /** |
88ca3b94 RD |
4781 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
4782 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
4783 | * |
4784 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4785 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4786 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
4787 | * function may optionally be used to account for unfreeable pages in the |
4788 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4789 | * smaller per-cpu batchsize. | |
0e0b864e MG |
4790 | */ |
4791 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4792 | { | |
4793 | dma_reserve = new_dma_reserve; | |
4794 | } | |
4795 | ||
1da177e4 LT |
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; | |
ab8fabd4 JW |
4862 | /* |
4863 | * Lowmem reserves are not available to | |
4864 | * GFP_HIGHUSER page cache allocations and | |
4865 | * kswapd tries to balance zones to their high | |
4866 | * watermark. As a result, neither should be | |
4867 | * regarded as dirtyable memory, to prevent a | |
4868 | * situation where reclaim has to clean pages | |
4869 | * in order to balance the zones. | |
4870 | */ | |
4871 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
4872 | } |
4873 | } | |
ab8fabd4 | 4874 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
4875 | totalreserve_pages = reserve_pages; |
4876 | } | |
4877 | ||
1da177e4 LT |
4878 | /* |
4879 | * setup_per_zone_lowmem_reserve - called whenever | |
4880 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4881 | * has a correct pages reserved value, so an adequate number of | |
4882 | * pages are left in the zone after a successful __alloc_pages(). | |
4883 | */ | |
4884 | static void setup_per_zone_lowmem_reserve(void) | |
4885 | { | |
4886 | struct pglist_data *pgdat; | |
2f6726e5 | 4887 | enum zone_type j, idx; |
1da177e4 | 4888 | |
ec936fc5 | 4889 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4890 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4891 | struct zone *zone = pgdat->node_zones + j; | |
4892 | unsigned long present_pages = zone->present_pages; | |
4893 | ||
4894 | zone->lowmem_reserve[j] = 0; | |
4895 | ||
2f6726e5 CL |
4896 | idx = j; |
4897 | while (idx) { | |
1da177e4 LT |
4898 | struct zone *lower_zone; |
4899 | ||
2f6726e5 CL |
4900 | idx--; |
4901 | ||
1da177e4 LT |
4902 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4903 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4904 | ||
4905 | lower_zone = pgdat->node_zones + idx; | |
4906 | lower_zone->lowmem_reserve[j] = present_pages / | |
4907 | sysctl_lowmem_reserve_ratio[idx]; | |
4908 | present_pages += lower_zone->present_pages; | |
4909 | } | |
4910 | } | |
4911 | } | |
cb45b0e9 HA |
4912 | |
4913 | /* update totalreserve_pages */ | |
4914 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4915 | } |
4916 | ||
88ca3b94 | 4917 | /** |
bc75d33f | 4918 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
bce7394a | 4919 | * or when memory is hot-{added|removed} |
88ca3b94 | 4920 | * |
bc75d33f MK |
4921 | * Ensures that the watermark[min,low,high] values for each zone are set |
4922 | * correctly with respect to min_free_kbytes. | |
1da177e4 | 4923 | */ |
bc75d33f | 4924 | void setup_per_zone_wmarks(void) |
1da177e4 LT |
4925 | { |
4926 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4927 | unsigned long lowmem_pages = 0; | |
4928 | struct zone *zone; | |
4929 | unsigned long flags; | |
4930 | ||
4931 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4932 | for_each_zone(zone) { | |
4933 | if (!is_highmem(zone)) | |
4934 | lowmem_pages += zone->present_pages; | |
4935 | } | |
4936 | ||
4937 | for_each_zone(zone) { | |
ac924c60 AM |
4938 | u64 tmp; |
4939 | ||
1125b4e3 | 4940 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
4941 | tmp = (u64)pages_min * zone->present_pages; |
4942 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4943 | if (is_highmem(zone)) { |
4944 | /* | |
669ed175 NP |
4945 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4946 | * need highmem pages, so cap pages_min to a small | |
4947 | * value here. | |
4948 | * | |
41858966 | 4949 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
4950 | * deltas controls asynch page reclaim, and so should |
4951 | * not be capped for highmem. | |
1da177e4 LT |
4952 | */ |
4953 | int min_pages; | |
4954 | ||
4955 | min_pages = zone->present_pages / 1024; | |
4956 | if (min_pages < SWAP_CLUSTER_MAX) | |
4957 | min_pages = SWAP_CLUSTER_MAX; | |
4958 | if (min_pages > 128) | |
4959 | min_pages = 128; | |
41858966 | 4960 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 4961 | } else { |
669ed175 NP |
4962 | /* |
4963 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4964 | * proportionate to the zone's size. |
4965 | */ | |
41858966 | 4966 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
4967 | } |
4968 | ||
41858966 MG |
4969 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
4970 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
56fd56b8 | 4971 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 4972 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 4973 | } |
cb45b0e9 HA |
4974 | |
4975 | /* update totalreserve_pages */ | |
4976 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4977 | } |
4978 | ||
55a4462a | 4979 | /* |
556adecb RR |
4980 | * The inactive anon list should be small enough that the VM never has to |
4981 | * do too much work, but large enough that each inactive page has a chance | |
4982 | * to be referenced again before it is swapped out. | |
4983 | * | |
4984 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4985 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4986 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4987 | * the anonymous pages are kept on the inactive list. | |
4988 | * | |
4989 | * total target max | |
4990 | * memory ratio inactive anon | |
4991 | * ------------------------------------- | |
4992 | * 10MB 1 5MB | |
4993 | * 100MB 1 50MB | |
4994 | * 1GB 3 250MB | |
4995 | * 10GB 10 0.9GB | |
4996 | * 100GB 31 3GB | |
4997 | * 1TB 101 10GB | |
4998 | * 10TB 320 32GB | |
4999 | */ | |
1b79acc9 | 5000 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5001 | { |
96cb4df5 | 5002 | unsigned int gb, ratio; |
556adecb | 5003 | |
96cb4df5 MK |
5004 | /* Zone size in gigabytes */ |
5005 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
5006 | if (gb) | |
556adecb | 5007 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5008 | else |
5009 | ratio = 1; | |
556adecb | 5010 | |
96cb4df5 MK |
5011 | zone->inactive_ratio = ratio; |
5012 | } | |
556adecb | 5013 | |
839a4fcc | 5014 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5015 | { |
5016 | struct zone *zone; | |
5017 | ||
5018 | for_each_zone(zone) | |
5019 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5020 | } |
5021 | ||
1da177e4 LT |
5022 | /* |
5023 | * Initialise min_free_kbytes. | |
5024 | * | |
5025 | * For small machines we want it small (128k min). For large machines | |
5026 | * we want it large (64MB max). But it is not linear, because network | |
5027 | * bandwidth does not increase linearly with machine size. We use | |
5028 | * | |
5029 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
5030 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
5031 | * | |
5032 | * which yields | |
5033 | * | |
5034 | * 16MB: 512k | |
5035 | * 32MB: 724k | |
5036 | * 64MB: 1024k | |
5037 | * 128MB: 1448k | |
5038 | * 256MB: 2048k | |
5039 | * 512MB: 2896k | |
5040 | * 1024MB: 4096k | |
5041 | * 2048MB: 5792k | |
5042 | * 4096MB: 8192k | |
5043 | * 8192MB: 11584k | |
5044 | * 16384MB: 16384k | |
5045 | */ | |
1b79acc9 | 5046 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5047 | { |
5048 | unsigned long lowmem_kbytes; | |
5049 | ||
5050 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5051 | ||
5052 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
5053 | if (min_free_kbytes < 128) | |
5054 | min_free_kbytes = 128; | |
5055 | if (min_free_kbytes > 65536) | |
5056 | min_free_kbytes = 65536; | |
bc75d33f | 5057 | setup_per_zone_wmarks(); |
a6cccdc3 | 5058 | refresh_zone_stat_thresholds(); |
1da177e4 | 5059 | setup_per_zone_lowmem_reserve(); |
556adecb | 5060 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5061 | return 0; |
5062 | } | |
bc75d33f | 5063 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5064 | |
5065 | /* | |
5066 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
5067 | * that we can call two helper functions whenever min_free_kbytes | |
5068 | * changes. | |
5069 | */ | |
5070 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5071 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5072 | { |
8d65af78 | 5073 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 5074 | if (write) |
bc75d33f | 5075 | setup_per_zone_wmarks(); |
1da177e4 LT |
5076 | return 0; |
5077 | } | |
5078 | ||
9614634f CL |
5079 | #ifdef CONFIG_NUMA |
5080 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5081 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5082 | { |
5083 | struct zone *zone; | |
5084 | int rc; | |
5085 | ||
8d65af78 | 5086 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5087 | if (rc) |
5088 | return rc; | |
5089 | ||
5090 | for_each_zone(zone) | |
8417bba4 | 5091 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
5092 | sysctl_min_unmapped_ratio) / 100; |
5093 | return 0; | |
5094 | } | |
0ff38490 CL |
5095 | |
5096 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5097 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5098 | { |
5099 | struct zone *zone; | |
5100 | int rc; | |
5101 | ||
8d65af78 | 5102 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5103 | if (rc) |
5104 | return rc; | |
5105 | ||
5106 | for_each_zone(zone) | |
5107 | zone->min_slab_pages = (zone->present_pages * | |
5108 | sysctl_min_slab_ratio) / 100; | |
5109 | return 0; | |
5110 | } | |
9614634f CL |
5111 | #endif |
5112 | ||
1da177e4 LT |
5113 | /* |
5114 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5115 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5116 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5117 | * | |
5118 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5119 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5120 | * if in function of the boot time zone sizes. |
5121 | */ | |
5122 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5123 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5124 | { |
8d65af78 | 5125 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5126 | setup_per_zone_lowmem_reserve(); |
5127 | return 0; | |
5128 | } | |
5129 | ||
8ad4b1fb RS |
5130 | /* |
5131 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
5132 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
5133 | * can have before it gets flushed back to buddy allocator. | |
5134 | */ | |
5135 | ||
5136 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5137 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5138 | { |
5139 | struct zone *zone; | |
5140 | unsigned int cpu; | |
5141 | int ret; | |
5142 | ||
8d65af78 | 5143 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
8ad4b1fb RS |
5144 | if (!write || (ret == -EINVAL)) |
5145 | return ret; | |
364df0eb | 5146 | for_each_populated_zone(zone) { |
99dcc3e5 | 5147 | for_each_possible_cpu(cpu) { |
8ad4b1fb RS |
5148 | unsigned long high; |
5149 | high = zone->present_pages / percpu_pagelist_fraction; | |
99dcc3e5 CL |
5150 | setup_pagelist_highmark( |
5151 | per_cpu_ptr(zone->pageset, cpu), high); | |
8ad4b1fb RS |
5152 | } |
5153 | } | |
5154 | return 0; | |
5155 | } | |
5156 | ||
f034b5d4 | 5157 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5158 | |
5159 | #ifdef CONFIG_NUMA | |
5160 | static int __init set_hashdist(char *str) | |
5161 | { | |
5162 | if (!str) | |
5163 | return 0; | |
5164 | hashdist = simple_strtoul(str, &str, 0); | |
5165 | return 1; | |
5166 | } | |
5167 | __setup("hashdist=", set_hashdist); | |
5168 | #endif | |
5169 | ||
5170 | /* | |
5171 | * allocate a large system hash table from bootmem | |
5172 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5173 | * quantity of entries | |
5174 | * - limit is the number of hash buckets, not the total allocation size | |
5175 | */ | |
5176 | void *__init alloc_large_system_hash(const char *tablename, | |
5177 | unsigned long bucketsize, | |
5178 | unsigned long numentries, | |
5179 | int scale, | |
5180 | int flags, | |
5181 | unsigned int *_hash_shift, | |
5182 | unsigned int *_hash_mask, | |
5183 | unsigned long limit) | |
5184 | { | |
5185 | unsigned long long max = limit; | |
5186 | unsigned long log2qty, size; | |
5187 | void *table = NULL; | |
5188 | ||
5189 | /* allow the kernel cmdline to have a say */ | |
5190 | if (!numentries) { | |
5191 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5192 | numentries = nr_kernel_pages; |
1da177e4 LT |
5193 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
5194 | numentries >>= 20 - PAGE_SHIFT; | |
5195 | numentries <<= 20 - PAGE_SHIFT; | |
5196 | ||
5197 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5198 | if (scale > PAGE_SHIFT) | |
5199 | numentries >>= (scale - PAGE_SHIFT); | |
5200 | else | |
5201 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5202 | |
5203 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5204 | if (unlikely(flags & HASH_SMALL)) { |
5205 | /* Makes no sense without HASH_EARLY */ | |
5206 | WARN_ON(!(flags & HASH_EARLY)); | |
5207 | if (!(numentries >> *_hash_shift)) { | |
5208 | numentries = 1UL << *_hash_shift; | |
5209 | BUG_ON(!numentries); | |
5210 | } | |
5211 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5212 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5213 | } |
6e692ed3 | 5214 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5215 | |
5216 | /* limit allocation size to 1/16 total memory by default */ | |
5217 | if (max == 0) { | |
5218 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5219 | do_div(max, bucketsize); | |
5220 | } | |
5221 | ||
5222 | if (numentries > max) | |
5223 | numentries = max; | |
5224 | ||
f0d1b0b3 | 5225 | log2qty = ilog2(numentries); |
1da177e4 LT |
5226 | |
5227 | do { | |
5228 | size = bucketsize << log2qty; | |
5229 | if (flags & HASH_EARLY) | |
74768ed8 | 5230 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5231 | else if (hashdist) |
5232 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5233 | else { | |
1037b83b ED |
5234 | /* |
5235 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5236 | * some pages at the end of hash table which |
5237 | * alloc_pages_exact() automatically does | |
1037b83b | 5238 | */ |
264ef8a9 | 5239 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5240 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5241 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5242 | } | |
1da177e4 LT |
5243 | } |
5244 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5245 | ||
5246 | if (!table) | |
5247 | panic("Failed to allocate %s hash table\n", tablename); | |
5248 | ||
f241e660 | 5249 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5250 | tablename, |
f241e660 | 5251 | (1UL << log2qty), |
f0d1b0b3 | 5252 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5253 | size); |
5254 | ||
5255 | if (_hash_shift) | |
5256 | *_hash_shift = log2qty; | |
5257 | if (_hash_mask) | |
5258 | *_hash_mask = (1 << log2qty) - 1; | |
5259 | ||
5260 | return table; | |
5261 | } | |
a117e66e | 5262 | |
835c134e MG |
5263 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5264 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5265 | unsigned long pfn) | |
5266 | { | |
5267 | #ifdef CONFIG_SPARSEMEM | |
5268 | return __pfn_to_section(pfn)->pageblock_flags; | |
5269 | #else | |
5270 | return zone->pageblock_flags; | |
5271 | #endif /* CONFIG_SPARSEMEM */ | |
5272 | } | |
5273 | ||
5274 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5275 | { | |
5276 | #ifdef CONFIG_SPARSEMEM | |
5277 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5278 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5279 | #else |
5280 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 5281 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5282 | #endif /* CONFIG_SPARSEMEM */ |
5283 | } | |
5284 | ||
5285 | /** | |
d9c23400 | 5286 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5287 | * @page: The page within the block of interest |
5288 | * @start_bitidx: The first bit of interest to retrieve | |
5289 | * @end_bitidx: The last bit of interest | |
5290 | * returns pageblock_bits flags | |
5291 | */ | |
5292 | unsigned long get_pageblock_flags_group(struct page *page, | |
5293 | int start_bitidx, int end_bitidx) | |
5294 | { | |
5295 | struct zone *zone; | |
5296 | unsigned long *bitmap; | |
5297 | unsigned long pfn, bitidx; | |
5298 | unsigned long flags = 0; | |
5299 | unsigned long value = 1; | |
5300 | ||
5301 | zone = page_zone(page); | |
5302 | pfn = page_to_pfn(page); | |
5303 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5304 | bitidx = pfn_to_bitidx(zone, pfn); | |
5305 | ||
5306 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5307 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5308 | flags |= value; | |
6220ec78 | 5309 | |
835c134e MG |
5310 | return flags; |
5311 | } | |
5312 | ||
5313 | /** | |
d9c23400 | 5314 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5315 | * @page: The page within the block of interest |
5316 | * @start_bitidx: The first bit of interest | |
5317 | * @end_bitidx: The last bit of interest | |
5318 | * @flags: The flags to set | |
5319 | */ | |
5320 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5321 | int start_bitidx, int end_bitidx) | |
5322 | { | |
5323 | struct zone *zone; | |
5324 | unsigned long *bitmap; | |
5325 | unsigned long pfn, bitidx; | |
5326 | unsigned long value = 1; | |
5327 | ||
5328 | zone = page_zone(page); | |
5329 | pfn = page_to_pfn(page); | |
5330 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5331 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
5332 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
5333 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
5334 | |
5335 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5336 | if (flags & value) | |
5337 | __set_bit(bitidx + start_bitidx, bitmap); | |
5338 | else | |
5339 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5340 | } | |
a5d76b54 KH |
5341 | |
5342 | /* | |
5343 | * This is designed as sub function...plz see page_isolation.c also. | |
5344 | * set/clear page block's type to be ISOLATE. | |
5345 | * page allocater never alloc memory from ISOLATE block. | |
5346 | */ | |
5347 | ||
49ac8255 KH |
5348 | static int |
5349 | __count_immobile_pages(struct zone *zone, struct page *page, int count) | |
5350 | { | |
5351 | unsigned long pfn, iter, found; | |
5352 | /* | |
5353 | * For avoiding noise data, lru_add_drain_all() should be called | |
5354 | * If ZONE_MOVABLE, the zone never contains immobile pages | |
5355 | */ | |
5356 | if (zone_idx(zone) == ZONE_MOVABLE) | |
5357 | return true; | |
5358 | ||
5359 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE) | |
5360 | return true; | |
5361 | ||
5362 | pfn = page_to_pfn(page); | |
5363 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5364 | unsigned long check = pfn + iter; | |
5365 | ||
29723fcc | 5366 | if (!pfn_valid_within(check)) |
49ac8255 | 5367 | continue; |
29723fcc | 5368 | |
49ac8255 KH |
5369 | page = pfn_to_page(check); |
5370 | if (!page_count(page)) { | |
5371 | if (PageBuddy(page)) | |
5372 | iter += (1 << page_order(page)) - 1; | |
5373 | continue; | |
5374 | } | |
5375 | if (!PageLRU(page)) | |
5376 | found++; | |
5377 | /* | |
5378 | * If there are RECLAIMABLE pages, we need to check it. | |
5379 | * But now, memory offline itself doesn't call shrink_slab() | |
5380 | * and it still to be fixed. | |
5381 | */ | |
5382 | /* | |
5383 | * If the page is not RAM, page_count()should be 0. | |
5384 | * we don't need more check. This is an _used_ not-movable page. | |
5385 | * | |
5386 | * The problematic thing here is PG_reserved pages. PG_reserved | |
5387 | * is set to both of a memory hole page and a _used_ kernel | |
5388 | * page at boot. | |
5389 | */ | |
5390 | if (found > count) | |
5391 | return false; | |
5392 | } | |
5393 | return true; | |
5394 | } | |
5395 | ||
5396 | bool is_pageblock_removable_nolock(struct page *page) | |
5397 | { | |
5398 | struct zone *zone = page_zone(page); | |
5399 | return __count_immobile_pages(zone, page, 0); | |
5400 | } | |
5401 | ||
a5d76b54 KH |
5402 | int set_migratetype_isolate(struct page *page) |
5403 | { | |
5404 | struct zone *zone; | |
49ac8255 | 5405 | unsigned long flags, pfn; |
925cc71e RJ |
5406 | struct memory_isolate_notify arg; |
5407 | int notifier_ret; | |
a5d76b54 KH |
5408 | int ret = -EBUSY; |
5409 | ||
5410 | zone = page_zone(page); | |
925cc71e | 5411 | |
a5d76b54 | 5412 | spin_lock_irqsave(&zone->lock, flags); |
925cc71e RJ |
5413 | |
5414 | pfn = page_to_pfn(page); | |
5415 | arg.start_pfn = pfn; | |
5416 | arg.nr_pages = pageblock_nr_pages; | |
5417 | arg.pages_found = 0; | |
5418 | ||
a5d76b54 | 5419 | /* |
925cc71e RJ |
5420 | * It may be possible to isolate a pageblock even if the |
5421 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
5422 | * notifier chain is used by balloon drivers to return the | |
5423 | * number of pages in a range that are held by the balloon | |
5424 | * driver to shrink memory. If all the pages are accounted for | |
5425 | * by balloons, are free, or on the LRU, isolation can continue. | |
5426 | * Later, for example, when memory hotplug notifier runs, these | |
5427 | * pages reported as "can be isolated" should be isolated(freed) | |
5428 | * by the balloon driver through the memory notifier chain. | |
a5d76b54 | 5429 | */ |
925cc71e RJ |
5430 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); |
5431 | notifier_ret = notifier_to_errno(notifier_ret); | |
4b20477f | 5432 | if (notifier_ret) |
a5d76b54 | 5433 | goto out; |
49ac8255 KH |
5434 | /* |
5435 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
5436 | * We just check MOVABLE pages. | |
5437 | */ | |
5438 | if (__count_immobile_pages(zone, page, arg.pages_found)) | |
925cc71e RJ |
5439 | ret = 0; |
5440 | ||
49ac8255 KH |
5441 | /* |
5442 | * immobile means "not-on-lru" paes. If immobile is larger than | |
5443 | * removable-by-driver pages reported by notifier, we'll fail. | |
5444 | */ | |
5445 | ||
a5d76b54 | 5446 | out: |
925cc71e RJ |
5447 | if (!ret) { |
5448 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
5449 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
5450 | } | |
5451 | ||
a5d76b54 KH |
5452 | spin_unlock_irqrestore(&zone->lock, flags); |
5453 | if (!ret) | |
9f8f2172 | 5454 | drain_all_pages(); |
a5d76b54 KH |
5455 | return ret; |
5456 | } | |
5457 | ||
5458 | void unset_migratetype_isolate(struct page *page) | |
5459 | { | |
5460 | struct zone *zone; | |
5461 | unsigned long flags; | |
5462 | zone = page_zone(page); | |
5463 | spin_lock_irqsave(&zone->lock, flags); | |
5464 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
5465 | goto out; | |
5466 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5467 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
5468 | out: | |
5469 | spin_unlock_irqrestore(&zone->lock, flags); | |
5470 | } | |
0c0e6195 KH |
5471 | |
5472 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5473 | /* | |
5474 | * All pages in the range must be isolated before calling this. | |
5475 | */ | |
5476 | void | |
5477 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5478 | { | |
5479 | struct page *page; | |
5480 | struct zone *zone; | |
5481 | int order, i; | |
5482 | unsigned long pfn; | |
5483 | unsigned long flags; | |
5484 | /* find the first valid pfn */ | |
5485 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5486 | if (pfn_valid(pfn)) | |
5487 | break; | |
5488 | if (pfn == end_pfn) | |
5489 | return; | |
5490 | zone = page_zone(pfn_to_page(pfn)); | |
5491 | spin_lock_irqsave(&zone->lock, flags); | |
5492 | pfn = start_pfn; | |
5493 | while (pfn < end_pfn) { | |
5494 | if (!pfn_valid(pfn)) { | |
5495 | pfn++; | |
5496 | continue; | |
5497 | } | |
5498 | page = pfn_to_page(pfn); | |
5499 | BUG_ON(page_count(page)); | |
5500 | BUG_ON(!PageBuddy(page)); | |
5501 | order = page_order(page); | |
5502 | #ifdef CONFIG_DEBUG_VM | |
5503 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5504 | pfn, 1 << order, end_pfn); | |
5505 | #endif | |
5506 | list_del(&page->lru); | |
5507 | rmv_page_order(page); | |
5508 | zone->free_area[order].nr_free--; | |
5509 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
5510 | - (1UL << order)); | |
5511 | for (i = 0; i < (1 << order); i++) | |
5512 | SetPageReserved((page+i)); | |
5513 | pfn += (1 << order); | |
5514 | } | |
5515 | spin_unlock_irqrestore(&zone->lock, flags); | |
5516 | } | |
5517 | #endif | |
8d22ba1b WF |
5518 | |
5519 | #ifdef CONFIG_MEMORY_FAILURE | |
5520 | bool is_free_buddy_page(struct page *page) | |
5521 | { | |
5522 | struct zone *zone = page_zone(page); | |
5523 | unsigned long pfn = page_to_pfn(page); | |
5524 | unsigned long flags; | |
5525 | int order; | |
5526 | ||
5527 | spin_lock_irqsave(&zone->lock, flags); | |
5528 | for (order = 0; order < MAX_ORDER; order++) { | |
5529 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
5530 | ||
5531 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
5532 | break; | |
5533 | } | |
5534 | spin_unlock_irqrestore(&zone->lock, flags); | |
5535 | ||
5536 | return order < MAX_ORDER; | |
5537 | } | |
5538 | #endif | |
718a3821 WF |
5539 | |
5540 | static struct trace_print_flags pageflag_names[] = { | |
5541 | {1UL << PG_locked, "locked" }, | |
5542 | {1UL << PG_error, "error" }, | |
5543 | {1UL << PG_referenced, "referenced" }, | |
5544 | {1UL << PG_uptodate, "uptodate" }, | |
5545 | {1UL << PG_dirty, "dirty" }, | |
5546 | {1UL << PG_lru, "lru" }, | |
5547 | {1UL << PG_active, "active" }, | |
5548 | {1UL << PG_slab, "slab" }, | |
5549 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
5550 | {1UL << PG_arch_1, "arch_1" }, | |
5551 | {1UL << PG_reserved, "reserved" }, | |
5552 | {1UL << PG_private, "private" }, | |
5553 | {1UL << PG_private_2, "private_2" }, | |
5554 | {1UL << PG_writeback, "writeback" }, | |
5555 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
5556 | {1UL << PG_head, "head" }, | |
5557 | {1UL << PG_tail, "tail" }, | |
5558 | #else | |
5559 | {1UL << PG_compound, "compound" }, | |
5560 | #endif | |
5561 | {1UL << PG_swapcache, "swapcache" }, | |
5562 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
5563 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
5564 | {1UL << PG_swapbacked, "swapbacked" }, |
5565 | {1UL << PG_unevictable, "unevictable" }, | |
5566 | #ifdef CONFIG_MMU | |
5567 | {1UL << PG_mlocked, "mlocked" }, | |
5568 | #endif | |
5569 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
5570 | {1UL << PG_uncached, "uncached" }, | |
5571 | #endif | |
5572 | #ifdef CONFIG_MEMORY_FAILURE | |
5573 | {1UL << PG_hwpoison, "hwpoison" }, | |
5574 | #endif | |
5575 | {-1UL, NULL }, | |
5576 | }; | |
5577 | ||
5578 | static void dump_page_flags(unsigned long flags) | |
5579 | { | |
5580 | const char *delim = ""; | |
5581 | unsigned long mask; | |
5582 | int i; | |
5583 | ||
5584 | printk(KERN_ALERT "page flags: %#lx(", flags); | |
5585 | ||
5586 | /* remove zone id */ | |
5587 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
5588 | ||
5589 | for (i = 0; pageflag_names[i].name && flags; i++) { | |
5590 | ||
5591 | mask = pageflag_names[i].mask; | |
5592 | if ((flags & mask) != mask) | |
5593 | continue; | |
5594 | ||
5595 | flags &= ~mask; | |
5596 | printk("%s%s", delim, pageflag_names[i].name); | |
5597 | delim = "|"; | |
5598 | } | |
5599 | ||
5600 | /* check for left over flags */ | |
5601 | if (flags) | |
5602 | printk("%s%#lx", delim, flags); | |
5603 | ||
5604 | printk(")\n"); | |
5605 | } | |
5606 | ||
5607 | void dump_page(struct page *page) | |
5608 | { | |
5609 | printk(KERN_ALERT | |
5610 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 5611 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
5612 | page->mapping, page->index); |
5613 | dump_page_flags(page->flags); | |
f212ad7c | 5614 | mem_cgroup_print_bad_page(page); |
718a3821 | 5615 | } |