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> |
56de7263 | 54 | #include <linux/compaction.h> |
0d3d062a | 55 | #include <trace/events/kmem.h> |
718a3821 | 56 | #include <linux/ftrace_event.h> |
f212ad7c | 57 | #include <linux/memcontrol.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
041d3a8c | 59 | #include <linux/migrate.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 | ||
1b76b02f | 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 | ||
702d1a6e MK |
221 | /* |
222 | * NOTE: | |
223 | * Don't use set_pageblock_migratetype(page, MIGRATE_ISOLATE) directly. | |
224 | * Instead, use {un}set_pageblock_isolate. | |
225 | */ | |
ee6f509c | 226 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 227 | { |
49255c61 MG |
228 | |
229 | if (unlikely(page_group_by_mobility_disabled)) | |
230 | migratetype = MIGRATE_UNMOVABLE; | |
231 | ||
b2a0ac88 MG |
232 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
233 | PB_migrate, PB_migrate_end); | |
234 | } | |
235 | ||
7f33d49a RW |
236 | bool oom_killer_disabled __read_mostly; |
237 | ||
13e7444b | 238 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 239 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 240 | { |
bdc8cb98 DH |
241 | int ret = 0; |
242 | unsigned seq; | |
243 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 244 | |
bdc8cb98 DH |
245 | do { |
246 | seq = zone_span_seqbegin(zone); | |
247 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
248 | ret = 1; | |
249 | else if (pfn < zone->zone_start_pfn) | |
250 | ret = 1; | |
251 | } while (zone_span_seqretry(zone, seq)); | |
252 | ||
253 | return ret; | |
c6a57e19 DH |
254 | } |
255 | ||
256 | static int page_is_consistent(struct zone *zone, struct page *page) | |
257 | { | |
14e07298 | 258 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 259 | return 0; |
1da177e4 | 260 | if (zone != page_zone(page)) |
c6a57e19 DH |
261 | return 0; |
262 | ||
263 | return 1; | |
264 | } | |
265 | /* | |
266 | * Temporary debugging check for pages not lying within a given zone. | |
267 | */ | |
268 | static int bad_range(struct zone *zone, struct page *page) | |
269 | { | |
270 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 271 | return 1; |
c6a57e19 DH |
272 | if (!page_is_consistent(zone, page)) |
273 | return 1; | |
274 | ||
1da177e4 LT |
275 | return 0; |
276 | } | |
13e7444b NP |
277 | #else |
278 | static inline int bad_range(struct zone *zone, struct page *page) | |
279 | { | |
280 | return 0; | |
281 | } | |
282 | #endif | |
283 | ||
224abf92 | 284 | static void bad_page(struct page *page) |
1da177e4 | 285 | { |
d936cf9b HD |
286 | static unsigned long resume; |
287 | static unsigned long nr_shown; | |
288 | static unsigned long nr_unshown; | |
289 | ||
2a7684a2 WF |
290 | /* Don't complain about poisoned pages */ |
291 | if (PageHWPoison(page)) { | |
ef2b4b95 | 292 | reset_page_mapcount(page); /* remove PageBuddy */ |
2a7684a2 WF |
293 | return; |
294 | } | |
295 | ||
d936cf9b HD |
296 | /* |
297 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
298 | * or allow a steady drip of one report per second. | |
299 | */ | |
300 | if (nr_shown == 60) { | |
301 | if (time_before(jiffies, resume)) { | |
302 | nr_unshown++; | |
303 | goto out; | |
304 | } | |
305 | if (nr_unshown) { | |
1e9e6365 HD |
306 | printk(KERN_ALERT |
307 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
308 | nr_unshown); |
309 | nr_unshown = 0; | |
310 | } | |
311 | nr_shown = 0; | |
312 | } | |
313 | if (nr_shown++ == 0) | |
314 | resume = jiffies + 60 * HZ; | |
315 | ||
1e9e6365 | 316 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 317 | current->comm, page_to_pfn(page)); |
718a3821 | 318 | dump_page(page); |
3dc14741 | 319 | |
4f31888c | 320 | print_modules(); |
1da177e4 | 321 | dump_stack(); |
d936cf9b | 322 | out: |
8cc3b392 | 323 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
ef2b4b95 | 324 | reset_page_mapcount(page); /* remove PageBuddy */ |
9f158333 | 325 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
326 | } |
327 | ||
1da177e4 LT |
328 | /* |
329 | * Higher-order pages are called "compound pages". They are structured thusly: | |
330 | * | |
331 | * The first PAGE_SIZE page is called the "head page". | |
332 | * | |
333 | * The remaining PAGE_SIZE pages are called "tail pages". | |
334 | * | |
6416b9fa WSH |
335 | * All pages have PG_compound set. All tail pages have their ->first_page |
336 | * pointing at the head page. | |
1da177e4 | 337 | * |
41d78ba5 HD |
338 | * The first tail page's ->lru.next holds the address of the compound page's |
339 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
340 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 341 | */ |
d98c7a09 HD |
342 | |
343 | static void free_compound_page(struct page *page) | |
344 | { | |
d85f3385 | 345 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
346 | } |
347 | ||
01ad1c08 | 348 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
349 | { |
350 | int i; | |
351 | int nr_pages = 1 << order; | |
352 | ||
353 | set_compound_page_dtor(page, free_compound_page); | |
354 | set_compound_order(page, order); | |
355 | __SetPageHead(page); | |
356 | for (i = 1; i < nr_pages; i++) { | |
357 | struct page *p = page + i; | |
18229df5 | 358 | __SetPageTail(p); |
58a84aa9 | 359 | set_page_count(p, 0); |
18229df5 AW |
360 | p->first_page = page; |
361 | } | |
362 | } | |
363 | ||
59ff4216 | 364 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 365 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
366 | { |
367 | int i; | |
368 | int nr_pages = 1 << order; | |
8cc3b392 | 369 | int bad = 0; |
1da177e4 | 370 | |
8cc3b392 HD |
371 | if (unlikely(compound_order(page) != order) || |
372 | unlikely(!PageHead(page))) { | |
224abf92 | 373 | bad_page(page); |
8cc3b392 HD |
374 | bad++; |
375 | } | |
1da177e4 | 376 | |
6d777953 | 377 | __ClearPageHead(page); |
8cc3b392 | 378 | |
18229df5 AW |
379 | for (i = 1; i < nr_pages; i++) { |
380 | struct page *p = page + i; | |
1da177e4 | 381 | |
e713a21d | 382 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 383 | bad_page(page); |
8cc3b392 HD |
384 | bad++; |
385 | } | |
d85f3385 | 386 | __ClearPageTail(p); |
1da177e4 | 387 | } |
8cc3b392 HD |
388 | |
389 | return bad; | |
1da177e4 | 390 | } |
1da177e4 | 391 | |
17cf4406 NP |
392 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
393 | { | |
394 | int i; | |
395 | ||
6626c5d5 AM |
396 | /* |
397 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
398 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
399 | */ | |
725d704e | 400 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
401 | for (i = 0; i < (1 << order); i++) |
402 | clear_highpage(page + i); | |
403 | } | |
404 | ||
c0a32fc5 SG |
405 | #ifdef CONFIG_DEBUG_PAGEALLOC |
406 | unsigned int _debug_guardpage_minorder; | |
407 | ||
408 | static int __init debug_guardpage_minorder_setup(char *buf) | |
409 | { | |
410 | unsigned long res; | |
411 | ||
412 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
413 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
414 | return 0; | |
415 | } | |
416 | _debug_guardpage_minorder = res; | |
417 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
418 | return 0; | |
419 | } | |
420 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
421 | ||
422 | static inline void set_page_guard_flag(struct page *page) | |
423 | { | |
424 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
425 | } | |
426 | ||
427 | static inline void clear_page_guard_flag(struct page *page) | |
428 | { | |
429 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
430 | } | |
431 | #else | |
432 | static inline void set_page_guard_flag(struct page *page) { } | |
433 | static inline void clear_page_guard_flag(struct page *page) { } | |
434 | #endif | |
435 | ||
6aa3001b AM |
436 | static inline void set_page_order(struct page *page, int order) |
437 | { | |
4c21e2f2 | 438 | set_page_private(page, order); |
676165a8 | 439 | __SetPageBuddy(page); |
1da177e4 LT |
440 | } |
441 | ||
442 | static inline void rmv_page_order(struct page *page) | |
443 | { | |
676165a8 | 444 | __ClearPageBuddy(page); |
4c21e2f2 | 445 | set_page_private(page, 0); |
1da177e4 LT |
446 | } |
447 | ||
448 | /* | |
449 | * Locate the struct page for both the matching buddy in our | |
450 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
451 | * | |
452 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
453 | * the following equation: | |
454 | * B2 = B1 ^ (1 << O) | |
455 | * For example, if the starting buddy (buddy2) is #8 its order | |
456 | * 1 buddy is #10: | |
457 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
458 | * | |
459 | * 2) Any buddy B will have an order O+1 parent P which | |
460 | * satisfies the following equation: | |
461 | * P = B & ~(1 << O) | |
462 | * | |
d6e05edc | 463 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 464 | */ |
1da177e4 | 465 | static inline unsigned long |
43506fad | 466 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 467 | { |
43506fad | 468 | return page_idx ^ (1 << order); |
1da177e4 LT |
469 | } |
470 | ||
471 | /* | |
472 | * This function checks whether a page is free && is the buddy | |
473 | * we can do coalesce a page and its buddy if | |
13e7444b | 474 | * (a) the buddy is not in a hole && |
676165a8 | 475 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
476 | * (c) a page and its buddy have the same order && |
477 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 478 | * |
5f24ce5f AA |
479 | * For recording whether a page is in the buddy system, we set ->_mapcount -2. |
480 | * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock. | |
1da177e4 | 481 | * |
676165a8 | 482 | * For recording page's order, we use page_private(page). |
1da177e4 | 483 | */ |
cb2b95e1 AW |
484 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
485 | int order) | |
1da177e4 | 486 | { |
14e07298 | 487 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 488 | return 0; |
13e7444b | 489 | |
cb2b95e1 AW |
490 | if (page_zone_id(page) != page_zone_id(buddy)) |
491 | return 0; | |
492 | ||
c0a32fc5 SG |
493 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
494 | VM_BUG_ON(page_count(buddy) != 0); | |
495 | return 1; | |
496 | } | |
497 | ||
cb2b95e1 | 498 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
a3af9c38 | 499 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 500 | return 1; |
676165a8 | 501 | } |
6aa3001b | 502 | return 0; |
1da177e4 LT |
503 | } |
504 | ||
505 | /* | |
506 | * Freeing function for a buddy system allocator. | |
507 | * | |
508 | * The concept of a buddy system is to maintain direct-mapped table | |
509 | * (containing bit values) for memory blocks of various "orders". | |
510 | * The bottom level table contains the map for the smallest allocatable | |
511 | * units of memory (here, pages), and each level above it describes | |
512 | * pairs of units from the levels below, hence, "buddies". | |
513 | * At a high level, all that happens here is marking the table entry | |
514 | * at the bottom level available, and propagating the changes upward | |
515 | * as necessary, plus some accounting needed to play nicely with other | |
516 | * parts of the VM system. | |
517 | * At each level, we keep a list of pages, which are heads of continuous | |
5f24ce5f | 518 | * free pages of length of (1 << order) and marked with _mapcount -2. Page's |
4c21e2f2 | 519 | * order is recorded in page_private(page) field. |
1da177e4 | 520 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
521 | * other. That is, if we allocate a small block, and both were |
522 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 523 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 524 | * triggers coalescing into a block of larger size. |
1da177e4 LT |
525 | * |
526 | * -- wli | |
527 | */ | |
528 | ||
48db57f8 | 529 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
530 | struct zone *zone, unsigned int order, |
531 | int migratetype) | |
1da177e4 LT |
532 | { |
533 | unsigned long page_idx; | |
6dda9d55 | 534 | unsigned long combined_idx; |
43506fad | 535 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 536 | struct page *buddy; |
1da177e4 | 537 | |
224abf92 | 538 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
539 | if (unlikely(destroy_compound_page(page, order))) |
540 | return; | |
1da177e4 | 541 | |
ed0ae21d MG |
542 | VM_BUG_ON(migratetype == -1); |
543 | ||
1da177e4 LT |
544 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
545 | ||
f2260e6b | 546 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 547 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 548 | |
1da177e4 | 549 | while (order < MAX_ORDER-1) { |
43506fad KC |
550 | buddy_idx = __find_buddy_index(page_idx, order); |
551 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 552 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 553 | break; |
c0a32fc5 SG |
554 | /* |
555 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
556 | * merge with it and move up one order. | |
557 | */ | |
558 | if (page_is_guard(buddy)) { | |
559 | clear_page_guard_flag(buddy); | |
560 | set_page_private(page, 0); | |
d1ce749a BZ |
561 | __mod_zone_freepage_state(zone, 1 << order, |
562 | migratetype); | |
c0a32fc5 SG |
563 | } else { |
564 | list_del(&buddy->lru); | |
565 | zone->free_area[order].nr_free--; | |
566 | rmv_page_order(buddy); | |
567 | } | |
43506fad | 568 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
569 | page = page + (combined_idx - page_idx); |
570 | page_idx = combined_idx; | |
571 | order++; | |
572 | } | |
573 | set_page_order(page, order); | |
6dda9d55 CZ |
574 | |
575 | /* | |
576 | * If this is not the largest possible page, check if the buddy | |
577 | * of the next-highest order is free. If it is, it's possible | |
578 | * that pages are being freed that will coalesce soon. In case, | |
579 | * that is happening, add the free page to the tail of the list | |
580 | * so it's less likely to be used soon and more likely to be merged | |
581 | * as a higher order page | |
582 | */ | |
b7f50cfa | 583 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 584 | struct page *higher_page, *higher_buddy; |
43506fad KC |
585 | combined_idx = buddy_idx & page_idx; |
586 | higher_page = page + (combined_idx - page_idx); | |
587 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 588 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
589 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
590 | list_add_tail(&page->lru, | |
591 | &zone->free_area[order].free_list[migratetype]); | |
592 | goto out; | |
593 | } | |
594 | } | |
595 | ||
596 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
597 | out: | |
1da177e4 LT |
598 | zone->free_area[order].nr_free++; |
599 | } | |
600 | ||
224abf92 | 601 | static inline int free_pages_check(struct page *page) |
1da177e4 | 602 | { |
92be2e33 NP |
603 | if (unlikely(page_mapcount(page) | |
604 | (page->mapping != NULL) | | |
a3af9c38 | 605 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
606 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
607 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 608 | bad_page(page); |
79f4b7bf | 609 | return 1; |
8cc3b392 | 610 | } |
79f4b7bf HD |
611 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
612 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
613 | return 0; | |
1da177e4 LT |
614 | } |
615 | ||
616 | /* | |
5f8dcc21 | 617 | * Frees a number of pages from the PCP lists |
1da177e4 | 618 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 619 | * count is the number of pages to free. |
1da177e4 LT |
620 | * |
621 | * If the zone was previously in an "all pages pinned" state then look to | |
622 | * see if this freeing clears that state. | |
623 | * | |
624 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
625 | * pinned" detection logic. | |
626 | */ | |
5f8dcc21 MG |
627 | static void free_pcppages_bulk(struct zone *zone, int count, |
628 | struct per_cpu_pages *pcp) | |
1da177e4 | 629 | { |
5f8dcc21 | 630 | int migratetype = 0; |
a6f9edd6 | 631 | int batch_free = 0; |
72853e29 | 632 | int to_free = count; |
5f8dcc21 | 633 | |
c54ad30c | 634 | spin_lock(&zone->lock); |
93e4a89a | 635 | zone->all_unreclaimable = 0; |
1da177e4 | 636 | zone->pages_scanned = 0; |
f2260e6b | 637 | |
72853e29 | 638 | while (to_free) { |
48db57f8 | 639 | struct page *page; |
5f8dcc21 MG |
640 | struct list_head *list; |
641 | ||
642 | /* | |
a6f9edd6 MG |
643 | * Remove pages from lists in a round-robin fashion. A |
644 | * batch_free count is maintained that is incremented when an | |
645 | * empty list is encountered. This is so more pages are freed | |
646 | * off fuller lists instead of spinning excessively around empty | |
647 | * lists | |
5f8dcc21 MG |
648 | */ |
649 | do { | |
a6f9edd6 | 650 | batch_free++; |
5f8dcc21 MG |
651 | if (++migratetype == MIGRATE_PCPTYPES) |
652 | migratetype = 0; | |
653 | list = &pcp->lists[migratetype]; | |
654 | } while (list_empty(list)); | |
48db57f8 | 655 | |
1d16871d NK |
656 | /* This is the only non-empty list. Free them all. */ |
657 | if (batch_free == MIGRATE_PCPTYPES) | |
658 | batch_free = to_free; | |
659 | ||
a6f9edd6 | 660 | do { |
770c8aaa BZ |
661 | int mt; /* migratetype of the to-be-freed page */ |
662 | ||
a6f9edd6 MG |
663 | page = list_entry(list->prev, struct page, lru); |
664 | /* must delete as __free_one_page list manipulates */ | |
665 | list_del(&page->lru); | |
b12c4ad1 | 666 | mt = get_freepage_migratetype(page); |
a7016235 | 667 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
770c8aaa BZ |
668 | __free_one_page(page, zone, 0, mt); |
669 | trace_mm_page_pcpu_drain(page, 0, mt); | |
97d0da22 WC |
670 | if (likely(get_pageblock_migratetype(page) != MIGRATE_ISOLATE)) { |
671 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); | |
672 | if (is_migrate_cma(mt)) | |
673 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
674 | } | |
72853e29 | 675 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 676 | } |
c54ad30c | 677 | spin_unlock(&zone->lock); |
1da177e4 LT |
678 | } |
679 | ||
ed0ae21d MG |
680 | static void free_one_page(struct zone *zone, struct page *page, int order, |
681 | int migratetype) | |
1da177e4 | 682 | { |
006d22d9 | 683 | spin_lock(&zone->lock); |
93e4a89a | 684 | zone->all_unreclaimable = 0; |
006d22d9 | 685 | zone->pages_scanned = 0; |
f2260e6b | 686 | |
ed0ae21d | 687 | __free_one_page(page, zone, order, migratetype); |
2139cbe6 | 688 | if (unlikely(migratetype != MIGRATE_ISOLATE)) |
d1ce749a | 689 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 690 | spin_unlock(&zone->lock); |
48db57f8 NP |
691 | } |
692 | ||
ec95f53a | 693 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 694 | { |
1da177e4 | 695 | int i; |
8cc3b392 | 696 | int bad = 0; |
1da177e4 | 697 | |
b413d48a | 698 | trace_mm_page_free(page, order); |
b1eeab67 VN |
699 | kmemcheck_free_shadow(page, order); |
700 | ||
8dd60a3a AA |
701 | if (PageAnon(page)) |
702 | page->mapping = NULL; | |
703 | for (i = 0; i < (1 << order); i++) | |
704 | bad += free_pages_check(page + i); | |
8cc3b392 | 705 | if (bad) |
ec95f53a | 706 | return false; |
689bcebf | 707 | |
3ac7fe5a | 708 | if (!PageHighMem(page)) { |
9858db50 | 709 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
710 | debug_check_no_obj_freed(page_address(page), |
711 | PAGE_SIZE << order); | |
712 | } | |
dafb1367 | 713 | arch_free_page(page, order); |
48db57f8 | 714 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 715 | |
ec95f53a KM |
716 | return true; |
717 | } | |
718 | ||
719 | static void __free_pages_ok(struct page *page, unsigned int order) | |
720 | { | |
721 | unsigned long flags; | |
95e34412 | 722 | int migratetype; |
ec95f53a KM |
723 | |
724 | if (!free_pages_prepare(page, order)) | |
725 | return; | |
726 | ||
c54ad30c | 727 | local_irq_save(flags); |
f8891e5e | 728 | __count_vm_events(PGFREE, 1 << order); |
95e34412 MK |
729 | migratetype = get_pageblock_migratetype(page); |
730 | set_freepage_migratetype(page, migratetype); | |
731 | free_one_page(page_zone(page), page, order, migratetype); | |
c54ad30c | 732 | local_irq_restore(flags); |
1da177e4 LT |
733 | } |
734 | ||
af370fb8 | 735 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 736 | { |
c3993076 JW |
737 | unsigned int nr_pages = 1 << order; |
738 | unsigned int loop; | |
a226f6c8 | 739 | |
c3993076 JW |
740 | prefetchw(page); |
741 | for (loop = 0; loop < nr_pages; loop++) { | |
742 | struct page *p = &page[loop]; | |
743 | ||
744 | if (loop + 1 < nr_pages) | |
745 | prefetchw(p + 1); | |
746 | __ClearPageReserved(p); | |
747 | set_page_count(p, 0); | |
a226f6c8 | 748 | } |
c3993076 JW |
749 | |
750 | set_page_refcounted(page); | |
751 | __free_pages(page, order); | |
a226f6c8 DH |
752 | } |
753 | ||
47118af0 MN |
754 | #ifdef CONFIG_CMA |
755 | /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */ | |
756 | void __init init_cma_reserved_pageblock(struct page *page) | |
757 | { | |
758 | unsigned i = pageblock_nr_pages; | |
759 | struct page *p = page; | |
760 | ||
761 | do { | |
762 | __ClearPageReserved(p); | |
763 | set_page_count(p, 0); | |
764 | } while (++p, --i); | |
765 | ||
766 | set_page_refcounted(page); | |
767 | set_pageblock_migratetype(page, MIGRATE_CMA); | |
768 | __free_pages(page, pageblock_order); | |
769 | totalram_pages += pageblock_nr_pages; | |
770 | } | |
771 | #endif | |
1da177e4 LT |
772 | |
773 | /* | |
774 | * The order of subdivision here is critical for the IO subsystem. | |
775 | * Please do not alter this order without good reasons and regression | |
776 | * testing. Specifically, as large blocks of memory are subdivided, | |
777 | * the order in which smaller blocks are delivered depends on the order | |
778 | * they're subdivided in this function. This is the primary factor | |
779 | * influencing the order in which pages are delivered to the IO | |
780 | * subsystem according to empirical testing, and this is also justified | |
781 | * by considering the behavior of a buddy system containing a single | |
782 | * large block of memory acted on by a series of small allocations. | |
783 | * This behavior is a critical factor in sglist merging's success. | |
784 | * | |
785 | * -- wli | |
786 | */ | |
085cc7d5 | 787 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
788 | int low, int high, struct free_area *area, |
789 | int migratetype) | |
1da177e4 LT |
790 | { |
791 | unsigned long size = 1 << high; | |
792 | ||
793 | while (high > low) { | |
794 | area--; | |
795 | high--; | |
796 | size >>= 1; | |
725d704e | 797 | VM_BUG_ON(bad_range(zone, &page[size])); |
c0a32fc5 SG |
798 | |
799 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
800 | if (high < debug_guardpage_minorder()) { | |
801 | /* | |
802 | * Mark as guard pages (or page), that will allow to | |
803 | * merge back to allocator when buddy will be freed. | |
804 | * Corresponding page table entries will not be touched, | |
805 | * pages will stay not present in virtual address space | |
806 | */ | |
807 | INIT_LIST_HEAD(&page[size].lru); | |
808 | set_page_guard_flag(&page[size]); | |
809 | set_page_private(&page[size], high); | |
810 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
811 | __mod_zone_freepage_state(zone, -(1 << high), |
812 | migratetype); | |
c0a32fc5 SG |
813 | continue; |
814 | } | |
815 | #endif | |
b2a0ac88 | 816 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
817 | area->nr_free++; |
818 | set_page_order(&page[size], high); | |
819 | } | |
1da177e4 LT |
820 | } |
821 | ||
1da177e4 LT |
822 | /* |
823 | * This page is about to be returned from the page allocator | |
824 | */ | |
2a7684a2 | 825 | static inline int check_new_page(struct page *page) |
1da177e4 | 826 | { |
92be2e33 NP |
827 | if (unlikely(page_mapcount(page) | |
828 | (page->mapping != NULL) | | |
a3af9c38 | 829 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
830 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
831 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 832 | bad_page(page); |
689bcebf | 833 | return 1; |
8cc3b392 | 834 | } |
2a7684a2 WF |
835 | return 0; |
836 | } | |
837 | ||
838 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
839 | { | |
840 | int i; | |
841 | ||
842 | for (i = 0; i < (1 << order); i++) { | |
843 | struct page *p = page + i; | |
844 | if (unlikely(check_new_page(p))) | |
845 | return 1; | |
846 | } | |
689bcebf | 847 | |
4c21e2f2 | 848 | set_page_private(page, 0); |
7835e98b | 849 | set_page_refcounted(page); |
cc102509 NP |
850 | |
851 | arch_alloc_page(page, order); | |
1da177e4 | 852 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
853 | |
854 | if (gfp_flags & __GFP_ZERO) | |
855 | prep_zero_page(page, order, gfp_flags); | |
856 | ||
857 | if (order && (gfp_flags & __GFP_COMP)) | |
858 | prep_compound_page(page, order); | |
859 | ||
689bcebf | 860 | return 0; |
1da177e4 LT |
861 | } |
862 | ||
56fd56b8 MG |
863 | /* |
864 | * Go through the free lists for the given migratetype and remove | |
865 | * the smallest available page from the freelists | |
866 | */ | |
728ec980 MG |
867 | static inline |
868 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
869 | int migratetype) |
870 | { | |
871 | unsigned int current_order; | |
872 | struct free_area * area; | |
873 | struct page *page; | |
874 | ||
875 | /* Find a page of the appropriate size in the preferred list */ | |
876 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
877 | area = &(zone->free_area[current_order]); | |
878 | if (list_empty(&area->free_list[migratetype])) | |
879 | continue; | |
880 | ||
881 | page = list_entry(area->free_list[migratetype].next, | |
882 | struct page, lru); | |
883 | list_del(&page->lru); | |
884 | rmv_page_order(page); | |
885 | area->nr_free--; | |
56fd56b8 MG |
886 | expand(zone, page, order, current_order, area, migratetype); |
887 | return page; | |
888 | } | |
889 | ||
890 | return NULL; | |
891 | } | |
892 | ||
893 | ||
b2a0ac88 MG |
894 | /* |
895 | * This array describes the order lists are fallen back to when | |
896 | * the free lists for the desirable migrate type are depleted | |
897 | */ | |
47118af0 MN |
898 | static int fallbacks[MIGRATE_TYPES][4] = { |
899 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
900 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
901 | #ifdef CONFIG_CMA | |
902 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
903 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
904 | #else | |
905 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
906 | #endif | |
6d4a4916 MN |
907 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
908 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
909 | }; |
910 | ||
c361be55 MG |
911 | /* |
912 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 913 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
914 | * boundary. If alignment is required, use move_freepages_block() |
915 | */ | |
435b405c | 916 | int move_freepages(struct zone *zone, |
b69a7288 AB |
917 | struct page *start_page, struct page *end_page, |
918 | int migratetype) | |
c361be55 MG |
919 | { |
920 | struct page *page; | |
921 | unsigned long order; | |
d100313f | 922 | int pages_moved = 0; |
c361be55 MG |
923 | |
924 | #ifndef CONFIG_HOLES_IN_ZONE | |
925 | /* | |
926 | * page_zone is not safe to call in this context when | |
927 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
928 | * anyway as we check zone boundaries in move_freepages_block(). | |
929 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 930 | * grouping pages by mobility |
c361be55 MG |
931 | */ |
932 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
933 | #endif | |
934 | ||
935 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
936 | /* Make sure we are not inadvertently changing nodes */ |
937 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
938 | ||
c361be55 MG |
939 | if (!pfn_valid_within(page_to_pfn(page))) { |
940 | page++; | |
941 | continue; | |
942 | } | |
943 | ||
944 | if (!PageBuddy(page)) { | |
945 | page++; | |
946 | continue; | |
947 | } | |
948 | ||
949 | order = page_order(page); | |
84be48d8 KS |
950 | list_move(&page->lru, |
951 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 952 | set_freepage_migratetype(page, migratetype); |
c361be55 | 953 | page += 1 << order; |
d100313f | 954 | pages_moved += 1 << order; |
c361be55 MG |
955 | } |
956 | ||
d100313f | 957 | return pages_moved; |
c361be55 MG |
958 | } |
959 | ||
ee6f509c | 960 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 961 | int migratetype) |
c361be55 MG |
962 | { |
963 | unsigned long start_pfn, end_pfn; | |
964 | struct page *start_page, *end_page; | |
965 | ||
966 | start_pfn = page_to_pfn(page); | |
d9c23400 | 967 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 968 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
969 | end_page = start_page + pageblock_nr_pages - 1; |
970 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
971 | |
972 | /* Do not cross zone boundaries */ | |
973 | if (start_pfn < zone->zone_start_pfn) | |
974 | start_page = page; | |
975 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
976 | return 0; | |
977 | ||
978 | return move_freepages(zone, start_page, end_page, migratetype); | |
979 | } | |
980 | ||
2f66a68f MG |
981 | static void change_pageblock_range(struct page *pageblock_page, |
982 | int start_order, int migratetype) | |
983 | { | |
984 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
985 | ||
986 | while (nr_pageblocks--) { | |
987 | set_pageblock_migratetype(pageblock_page, migratetype); | |
988 | pageblock_page += pageblock_nr_pages; | |
989 | } | |
990 | } | |
991 | ||
b2a0ac88 | 992 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
993 | static inline struct page * |
994 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
995 | { |
996 | struct free_area * area; | |
997 | int current_order; | |
998 | struct page *page; | |
999 | int migratetype, i; | |
1000 | ||
1001 | /* Find the largest possible block of pages in the other list */ | |
1002 | for (current_order = MAX_ORDER-1; current_order >= order; | |
1003 | --current_order) { | |
6d4a4916 | 1004 | for (i = 0;; i++) { |
b2a0ac88 MG |
1005 | migratetype = fallbacks[start_migratetype][i]; |
1006 | ||
56fd56b8 MG |
1007 | /* MIGRATE_RESERVE handled later if necessary */ |
1008 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1009 | break; |
e010487d | 1010 | |
b2a0ac88 MG |
1011 | area = &(zone->free_area[current_order]); |
1012 | if (list_empty(&area->free_list[migratetype])) | |
1013 | continue; | |
1014 | ||
1015 | page = list_entry(area->free_list[migratetype].next, | |
1016 | struct page, lru); | |
1017 | area->nr_free--; | |
1018 | ||
1019 | /* | |
c361be55 | 1020 | * If breaking a large block of pages, move all free |
46dafbca MG |
1021 | * pages to the preferred allocation list. If falling |
1022 | * back for a reclaimable kernel allocation, be more | |
25985edc | 1023 | * aggressive about taking ownership of free pages |
47118af0 MN |
1024 | * |
1025 | * On the other hand, never change migration | |
1026 | * type of MIGRATE_CMA pageblocks nor move CMA | |
1027 | * pages on different free lists. We don't | |
1028 | * want unmovable pages to be allocated from | |
1029 | * MIGRATE_CMA areas. | |
b2a0ac88 | 1030 | */ |
47118af0 MN |
1031 | if (!is_migrate_cma(migratetype) && |
1032 | (unlikely(current_order >= pageblock_order / 2) || | |
1033 | start_migratetype == MIGRATE_RECLAIMABLE || | |
1034 | page_group_by_mobility_disabled)) { | |
1035 | int pages; | |
46dafbca MG |
1036 | pages = move_freepages_block(zone, page, |
1037 | start_migratetype); | |
1038 | ||
1039 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
1040 | if (pages >= (1 << (pageblock_order-1)) || |
1041 | page_group_by_mobility_disabled) | |
46dafbca MG |
1042 | set_pageblock_migratetype(page, |
1043 | start_migratetype); | |
1044 | ||
b2a0ac88 | 1045 | migratetype = start_migratetype; |
c361be55 | 1046 | } |
b2a0ac88 MG |
1047 | |
1048 | /* Remove the page from the freelists */ | |
1049 | list_del(&page->lru); | |
1050 | rmv_page_order(page); | |
b2a0ac88 | 1051 | |
2f66a68f | 1052 | /* Take ownership for orders >= pageblock_order */ |
47118af0 MN |
1053 | if (current_order >= pageblock_order && |
1054 | !is_migrate_cma(migratetype)) | |
2f66a68f | 1055 | change_pageblock_range(page, current_order, |
b2a0ac88 MG |
1056 | start_migratetype); |
1057 | ||
47118af0 MN |
1058 | expand(zone, page, order, current_order, area, |
1059 | is_migrate_cma(migratetype) | |
1060 | ? migratetype : start_migratetype); | |
e0fff1bd MG |
1061 | |
1062 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
1063 | start_migratetype, migratetype); | |
1064 | ||
b2a0ac88 MG |
1065 | return page; |
1066 | } | |
1067 | } | |
1068 | ||
728ec980 | 1069 | return NULL; |
b2a0ac88 MG |
1070 | } |
1071 | ||
56fd56b8 | 1072 | /* |
1da177e4 LT |
1073 | * Do the hard work of removing an element from the buddy allocator. |
1074 | * Call me with the zone->lock already held. | |
1075 | */ | |
b2a0ac88 MG |
1076 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1077 | int migratetype) | |
1da177e4 | 1078 | { |
1da177e4 LT |
1079 | struct page *page; |
1080 | ||
728ec980 | 1081 | retry_reserve: |
56fd56b8 | 1082 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1083 | |
728ec980 | 1084 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1085 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1086 | |
728ec980 MG |
1087 | /* |
1088 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1089 | * is used because __rmqueue_smallest is an inline function | |
1090 | * and we want just one call site | |
1091 | */ | |
1092 | if (!page) { | |
1093 | migratetype = MIGRATE_RESERVE; | |
1094 | goto retry_reserve; | |
1095 | } | |
1096 | } | |
1097 | ||
0d3d062a | 1098 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1099 | return page; |
1da177e4 LT |
1100 | } |
1101 | ||
5f63b720 | 1102 | /* |
1da177e4 LT |
1103 | * Obtain a specified number of elements from the buddy allocator, all under |
1104 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1105 | * Returns the number of new pages which were placed at *list. | |
1106 | */ | |
5f63b720 | 1107 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1108 | unsigned long count, struct list_head *list, |
e084b2d9 | 1109 | int migratetype, int cold) |
1da177e4 | 1110 | { |
47118af0 | 1111 | int mt = migratetype, i; |
5f63b720 | 1112 | |
c54ad30c | 1113 | spin_lock(&zone->lock); |
1da177e4 | 1114 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1115 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1116 | if (unlikely(page == NULL)) |
1da177e4 | 1117 | break; |
81eabcbe MG |
1118 | |
1119 | /* | |
1120 | * Split buddy pages returned by expand() are received here | |
1121 | * in physical page order. The page is added to the callers and | |
1122 | * list and the list head then moves forward. From the callers | |
1123 | * perspective, the linked list is ordered by page number in | |
1124 | * some conditions. This is useful for IO devices that can | |
1125 | * merge IO requests if the physical pages are ordered | |
1126 | * properly. | |
1127 | */ | |
e084b2d9 MG |
1128 | if (likely(cold == 0)) |
1129 | list_add(&page->lru, list); | |
1130 | else | |
1131 | list_add_tail(&page->lru, list); | |
47118af0 MN |
1132 | if (IS_ENABLED(CONFIG_CMA)) { |
1133 | mt = get_pageblock_migratetype(page); | |
1134 | if (!is_migrate_cma(mt) && mt != MIGRATE_ISOLATE) | |
1135 | mt = migratetype; | |
1136 | } | |
b12c4ad1 | 1137 | set_freepage_migratetype(page, mt); |
81eabcbe | 1138 | list = &page->lru; |
d1ce749a BZ |
1139 | if (is_migrate_cma(mt)) |
1140 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, | |
1141 | -(1 << order)); | |
1da177e4 | 1142 | } |
f2260e6b | 1143 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1144 | spin_unlock(&zone->lock); |
085cc7d5 | 1145 | return i; |
1da177e4 LT |
1146 | } |
1147 | ||
4ae7c039 | 1148 | #ifdef CONFIG_NUMA |
8fce4d8e | 1149 | /* |
4037d452 CL |
1150 | * Called from the vmstat counter updater to drain pagesets of this |
1151 | * currently executing processor on remote nodes after they have | |
1152 | * expired. | |
1153 | * | |
879336c3 CL |
1154 | * Note that this function must be called with the thread pinned to |
1155 | * a single processor. | |
8fce4d8e | 1156 | */ |
4037d452 | 1157 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1158 | { |
4ae7c039 | 1159 | unsigned long flags; |
4037d452 | 1160 | int to_drain; |
4ae7c039 | 1161 | |
4037d452 CL |
1162 | local_irq_save(flags); |
1163 | if (pcp->count >= pcp->batch) | |
1164 | to_drain = pcp->batch; | |
1165 | else | |
1166 | to_drain = pcp->count; | |
2a13515c KM |
1167 | if (to_drain > 0) { |
1168 | free_pcppages_bulk(zone, to_drain, pcp); | |
1169 | pcp->count -= to_drain; | |
1170 | } | |
4037d452 | 1171 | local_irq_restore(flags); |
4ae7c039 CL |
1172 | } |
1173 | #endif | |
1174 | ||
9f8f2172 CL |
1175 | /* |
1176 | * Drain pages of the indicated processor. | |
1177 | * | |
1178 | * The processor must either be the current processor and the | |
1179 | * thread pinned to the current processor or a processor that | |
1180 | * is not online. | |
1181 | */ | |
1182 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1183 | { |
c54ad30c | 1184 | unsigned long flags; |
1da177e4 | 1185 | struct zone *zone; |
1da177e4 | 1186 | |
ee99c71c | 1187 | for_each_populated_zone(zone) { |
1da177e4 | 1188 | struct per_cpu_pageset *pset; |
3dfa5721 | 1189 | struct per_cpu_pages *pcp; |
1da177e4 | 1190 | |
99dcc3e5 CL |
1191 | local_irq_save(flags); |
1192 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1193 | |
1194 | pcp = &pset->pcp; | |
2ff754fa DR |
1195 | if (pcp->count) { |
1196 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1197 | pcp->count = 0; | |
1198 | } | |
3dfa5721 | 1199 | local_irq_restore(flags); |
1da177e4 LT |
1200 | } |
1201 | } | |
1da177e4 | 1202 | |
9f8f2172 CL |
1203 | /* |
1204 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1205 | */ | |
1206 | void drain_local_pages(void *arg) | |
1207 | { | |
1208 | drain_pages(smp_processor_id()); | |
1209 | } | |
1210 | ||
1211 | /* | |
74046494 GBY |
1212 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1213 | * | |
1214 | * Note that this code is protected against sending an IPI to an offline | |
1215 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1216 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1217 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1218 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1219 | */ |
1220 | void drain_all_pages(void) | |
1221 | { | |
74046494 GBY |
1222 | int cpu; |
1223 | struct per_cpu_pageset *pcp; | |
1224 | struct zone *zone; | |
1225 | ||
1226 | /* | |
1227 | * Allocate in the BSS so we wont require allocation in | |
1228 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1229 | */ | |
1230 | static cpumask_t cpus_with_pcps; | |
1231 | ||
1232 | /* | |
1233 | * We don't care about racing with CPU hotplug event | |
1234 | * as offline notification will cause the notified | |
1235 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1236 | * disables preemption as part of its processing | |
1237 | */ | |
1238 | for_each_online_cpu(cpu) { | |
1239 | bool has_pcps = false; | |
1240 | for_each_populated_zone(zone) { | |
1241 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1242 | if (pcp->pcp.count) { | |
1243 | has_pcps = true; | |
1244 | break; | |
1245 | } | |
1246 | } | |
1247 | if (has_pcps) | |
1248 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1249 | else | |
1250 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1251 | } | |
1252 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1253 | } |
1254 | ||
296699de | 1255 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1256 | |
1257 | void mark_free_pages(struct zone *zone) | |
1258 | { | |
f623f0db RW |
1259 | unsigned long pfn, max_zone_pfn; |
1260 | unsigned long flags; | |
b2a0ac88 | 1261 | int order, t; |
1da177e4 LT |
1262 | struct list_head *curr; |
1263 | ||
1264 | if (!zone->spanned_pages) | |
1265 | return; | |
1266 | ||
1267 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1268 | |
1269 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1270 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1271 | if (pfn_valid(pfn)) { | |
1272 | struct page *page = pfn_to_page(pfn); | |
1273 | ||
7be98234 RW |
1274 | if (!swsusp_page_is_forbidden(page)) |
1275 | swsusp_unset_page_free(page); | |
f623f0db | 1276 | } |
1da177e4 | 1277 | |
b2a0ac88 MG |
1278 | for_each_migratetype_order(order, t) { |
1279 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1280 | unsigned long i; |
1da177e4 | 1281 | |
f623f0db RW |
1282 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1283 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1284 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1285 | } |
b2a0ac88 | 1286 | } |
1da177e4 LT |
1287 | spin_unlock_irqrestore(&zone->lock, flags); |
1288 | } | |
e2c55dc8 | 1289 | #endif /* CONFIG_PM */ |
1da177e4 | 1290 | |
1da177e4 LT |
1291 | /* |
1292 | * Free a 0-order page | |
fc91668e | 1293 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1294 | */ |
fc91668e | 1295 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1296 | { |
1297 | struct zone *zone = page_zone(page); | |
1298 | struct per_cpu_pages *pcp; | |
1299 | unsigned long flags; | |
5f8dcc21 | 1300 | int migratetype; |
1da177e4 | 1301 | |
ec95f53a | 1302 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1303 | return; |
1304 | ||
5f8dcc21 | 1305 | migratetype = get_pageblock_migratetype(page); |
b12c4ad1 | 1306 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1307 | local_irq_save(flags); |
f8891e5e | 1308 | __count_vm_event(PGFREE); |
da456f14 | 1309 | |
5f8dcc21 MG |
1310 | /* |
1311 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1312 | * Free ISOLATE pages back to the allocator because they are being | |
1313 | * offlined but treat RESERVE as movable pages so we can get those | |
1314 | * areas back if necessary. Otherwise, we may have to free | |
1315 | * excessively into the page allocator | |
1316 | */ | |
1317 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1318 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1319 | free_one_page(zone, page, 0, migratetype); | |
1320 | goto out; | |
1321 | } | |
1322 | migratetype = MIGRATE_MOVABLE; | |
1323 | } | |
1324 | ||
99dcc3e5 | 1325 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1326 | if (cold) |
5f8dcc21 | 1327 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1328 | else |
5f8dcc21 | 1329 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1330 | pcp->count++; |
48db57f8 | 1331 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1332 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1333 | pcp->count -= pcp->batch; |
1334 | } | |
5f8dcc21 MG |
1335 | |
1336 | out: | |
1da177e4 | 1337 | local_irq_restore(flags); |
1da177e4 LT |
1338 | } |
1339 | ||
cc59850e KK |
1340 | /* |
1341 | * Free a list of 0-order pages | |
1342 | */ | |
1343 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1344 | { | |
1345 | struct page *page, *next; | |
1346 | ||
1347 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1348 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1349 | free_hot_cold_page(page, cold); |
1350 | } | |
1351 | } | |
1352 | ||
8dfcc9ba NP |
1353 | /* |
1354 | * split_page takes a non-compound higher-order page, and splits it into | |
1355 | * n (1<<order) sub-pages: page[0..n] | |
1356 | * Each sub-page must be freed individually. | |
1357 | * | |
1358 | * Note: this is probably too low level an operation for use in drivers. | |
1359 | * Please consult with lkml before using this in your driver. | |
1360 | */ | |
1361 | void split_page(struct page *page, unsigned int order) | |
1362 | { | |
1363 | int i; | |
1364 | ||
725d704e NP |
1365 | VM_BUG_ON(PageCompound(page)); |
1366 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1367 | |
1368 | #ifdef CONFIG_KMEMCHECK | |
1369 | /* | |
1370 | * Split shadow pages too, because free(page[0]) would | |
1371 | * otherwise free the whole shadow. | |
1372 | */ | |
1373 | if (kmemcheck_page_is_tracked(page)) | |
1374 | split_page(virt_to_page(page[0].shadow), order); | |
1375 | #endif | |
1376 | ||
7835e98b NP |
1377 | for (i = 1; i < (1 << order); i++) |
1378 | set_page_refcounted(page + i); | |
8dfcc9ba | 1379 | } |
8dfcc9ba | 1380 | |
748446bb | 1381 | /* |
1fb3f8ca MG |
1382 | * Similar to the split_page family of functions except that the page |
1383 | * required at the given order and being isolated now to prevent races | |
1384 | * with parallel allocators | |
748446bb | 1385 | */ |
1fb3f8ca | 1386 | int capture_free_page(struct page *page, int alloc_order, int migratetype) |
748446bb MG |
1387 | { |
1388 | unsigned int order; | |
1389 | unsigned long watermark; | |
1390 | struct zone *zone; | |
2139cbe6 | 1391 | int mt; |
748446bb MG |
1392 | |
1393 | BUG_ON(!PageBuddy(page)); | |
1394 | ||
1395 | zone = page_zone(page); | |
1396 | order = page_order(page); | |
2e30abd1 | 1397 | mt = get_pageblock_migratetype(page); |
748446bb | 1398 | |
2e30abd1 MS |
1399 | if (mt != MIGRATE_ISOLATE) { |
1400 | /* Obey watermarks as if the page was being allocated */ | |
1401 | watermark = low_wmark_pages(zone) + (1 << order); | |
1402 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1403 | return 0; | |
1404 | ||
1405 | __mod_zone_freepage_state(zone, -(1UL << alloc_order), mt); | |
1406 | } | |
748446bb MG |
1407 | |
1408 | /* Remove page from free list */ | |
1409 | list_del(&page->lru); | |
1410 | zone->free_area[order].nr_free--; | |
1411 | rmv_page_order(page); | |
2139cbe6 | 1412 | |
1fb3f8ca MG |
1413 | if (alloc_order != order) |
1414 | expand(zone, page, alloc_order, order, | |
1415 | &zone->free_area[order], migratetype); | |
748446bb | 1416 | |
1fb3f8ca | 1417 | /* Set the pageblock if the captured page is at least a pageblock */ |
748446bb MG |
1418 | if (order >= pageblock_order - 1) { |
1419 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1420 | for (; page < endpage; page += pageblock_nr_pages) { |
1421 | int mt = get_pageblock_migratetype(page); | |
1422 | if (mt != MIGRATE_ISOLATE && !is_migrate_cma(mt)) | |
1423 | set_pageblock_migratetype(page, | |
1424 | MIGRATE_MOVABLE); | |
1425 | } | |
748446bb MG |
1426 | } |
1427 | ||
58d00209 | 1428 | return 1UL << alloc_order; |
1fb3f8ca MG |
1429 | } |
1430 | ||
1431 | /* | |
1432 | * Similar to split_page except the page is already free. As this is only | |
1433 | * being used for migration, the migratetype of the block also changes. | |
1434 | * As this is called with interrupts disabled, the caller is responsible | |
1435 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1436 | * are enabled. | |
1437 | * | |
1438 | * Note: this is probably too low level an operation for use in drivers. | |
1439 | * Please consult with lkml before using this in your driver. | |
1440 | */ | |
1441 | int split_free_page(struct page *page) | |
1442 | { | |
1443 | unsigned int order; | |
1444 | int nr_pages; | |
1445 | ||
1446 | BUG_ON(!PageBuddy(page)); | |
1447 | order = page_order(page); | |
1448 | ||
1449 | nr_pages = capture_free_page(page, order, 0); | |
1450 | if (!nr_pages) | |
1451 | return 0; | |
1452 | ||
1453 | /* Split into individual pages */ | |
1454 | set_page_refcounted(page); | |
1455 | split_page(page, order); | |
1456 | return nr_pages; | |
748446bb MG |
1457 | } |
1458 | ||
1da177e4 LT |
1459 | /* |
1460 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1461 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1462 | * or two. | |
1463 | */ | |
0a15c3e9 MG |
1464 | static inline |
1465 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1466 | struct zone *zone, int order, gfp_t gfp_flags, |
1467 | int migratetype) | |
1da177e4 LT |
1468 | { |
1469 | unsigned long flags; | |
689bcebf | 1470 | struct page *page; |
1da177e4 LT |
1471 | int cold = !!(gfp_flags & __GFP_COLD); |
1472 | ||
689bcebf | 1473 | again: |
48db57f8 | 1474 | if (likely(order == 0)) { |
1da177e4 | 1475 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1476 | struct list_head *list; |
1da177e4 | 1477 | |
1da177e4 | 1478 | local_irq_save(flags); |
99dcc3e5 CL |
1479 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1480 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1481 | if (list_empty(list)) { |
535131e6 | 1482 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1483 | pcp->batch, list, |
e084b2d9 | 1484 | migratetype, cold); |
5f8dcc21 | 1485 | if (unlikely(list_empty(list))) |
6fb332fa | 1486 | goto failed; |
535131e6 | 1487 | } |
b92a6edd | 1488 | |
5f8dcc21 MG |
1489 | if (cold) |
1490 | page = list_entry(list->prev, struct page, lru); | |
1491 | else | |
1492 | page = list_entry(list->next, struct page, lru); | |
1493 | ||
b92a6edd MG |
1494 | list_del(&page->lru); |
1495 | pcp->count--; | |
7fb1d9fc | 1496 | } else { |
dab48dab AM |
1497 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1498 | /* | |
1499 | * __GFP_NOFAIL is not to be used in new code. | |
1500 | * | |
1501 | * All __GFP_NOFAIL callers should be fixed so that they | |
1502 | * properly detect and handle allocation failures. | |
1503 | * | |
1504 | * We most definitely don't want callers attempting to | |
4923abf9 | 1505 | * allocate greater than order-1 page units with |
dab48dab AM |
1506 | * __GFP_NOFAIL. |
1507 | */ | |
4923abf9 | 1508 | WARN_ON_ONCE(order > 1); |
dab48dab | 1509 | } |
1da177e4 | 1510 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1511 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1512 | spin_unlock(&zone->lock); |
1513 | if (!page) | |
1514 | goto failed; | |
d1ce749a BZ |
1515 | __mod_zone_freepage_state(zone, -(1 << order), |
1516 | get_pageblock_migratetype(page)); | |
1da177e4 LT |
1517 | } |
1518 | ||
f8891e5e | 1519 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1520 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1521 | local_irq_restore(flags); |
1da177e4 | 1522 | |
725d704e | 1523 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1524 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1525 | goto again; |
1da177e4 | 1526 | return page; |
a74609fa NP |
1527 | |
1528 | failed: | |
1529 | local_irq_restore(flags); | |
a74609fa | 1530 | return NULL; |
1da177e4 LT |
1531 | } |
1532 | ||
933e312e AM |
1533 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1534 | ||
b2588c4b | 1535 | static struct { |
933e312e AM |
1536 | struct fault_attr attr; |
1537 | ||
1538 | u32 ignore_gfp_highmem; | |
1539 | u32 ignore_gfp_wait; | |
54114994 | 1540 | u32 min_order; |
933e312e AM |
1541 | } fail_page_alloc = { |
1542 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1543 | .ignore_gfp_wait = 1, |
1544 | .ignore_gfp_highmem = 1, | |
54114994 | 1545 | .min_order = 1, |
933e312e AM |
1546 | }; |
1547 | ||
1548 | static int __init setup_fail_page_alloc(char *str) | |
1549 | { | |
1550 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1551 | } | |
1552 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1553 | ||
deaf386e | 1554 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1555 | { |
54114994 | 1556 | if (order < fail_page_alloc.min_order) |
deaf386e | 1557 | return false; |
933e312e | 1558 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1559 | return false; |
933e312e | 1560 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1561 | return false; |
933e312e | 1562 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1563 | return false; |
933e312e AM |
1564 | |
1565 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1566 | } | |
1567 | ||
1568 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1569 | ||
1570 | static int __init fail_page_alloc_debugfs(void) | |
1571 | { | |
f4ae40a6 | 1572 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1573 | struct dentry *dir; |
933e312e | 1574 | |
dd48c085 AM |
1575 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1576 | &fail_page_alloc.attr); | |
1577 | if (IS_ERR(dir)) | |
1578 | return PTR_ERR(dir); | |
933e312e | 1579 | |
b2588c4b AM |
1580 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1581 | &fail_page_alloc.ignore_gfp_wait)) | |
1582 | goto fail; | |
1583 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1584 | &fail_page_alloc.ignore_gfp_highmem)) | |
1585 | goto fail; | |
1586 | if (!debugfs_create_u32("min-order", mode, dir, | |
1587 | &fail_page_alloc.min_order)) | |
1588 | goto fail; | |
1589 | ||
1590 | return 0; | |
1591 | fail: | |
dd48c085 | 1592 | debugfs_remove_recursive(dir); |
933e312e | 1593 | |
b2588c4b | 1594 | return -ENOMEM; |
933e312e AM |
1595 | } |
1596 | ||
1597 | late_initcall(fail_page_alloc_debugfs); | |
1598 | ||
1599 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1600 | ||
1601 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1602 | ||
deaf386e | 1603 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1604 | { |
deaf386e | 1605 | return false; |
933e312e AM |
1606 | } |
1607 | ||
1608 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1609 | ||
1da177e4 | 1610 | /* |
88f5acf8 | 1611 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1612 | * of the allocation. |
1613 | */ | |
88f5acf8 MG |
1614 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1615 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1616 | { |
1617 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1618 | long min = mark; |
2cfed075 | 1619 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 LT |
1620 | int o; |
1621 | ||
df0a6daa | 1622 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1623 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1624 | min -= min / 2; |
7fb1d9fc | 1625 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1626 | min -= min / 4; |
d95ea5d1 BZ |
1627 | #ifdef CONFIG_CMA |
1628 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1629 | if (!(alloc_flags & ALLOC_CMA)) | |
1630 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
1631 | #endif | |
2cfed075 | 1632 | if (free_pages <= min + lowmem_reserve) |
88f5acf8 | 1633 | return false; |
1da177e4 LT |
1634 | for (o = 0; o < order; o++) { |
1635 | /* At the next order, this order's pages become unavailable */ | |
1636 | free_pages -= z->free_area[o].nr_free << o; | |
1637 | ||
1638 | /* Require fewer higher order pages to be free */ | |
1639 | min >>= 1; | |
1640 | ||
1641 | if (free_pages <= min) | |
88f5acf8 | 1642 | return false; |
1da177e4 | 1643 | } |
88f5acf8 MG |
1644 | return true; |
1645 | } | |
1646 | ||
702d1a6e MK |
1647 | #ifdef CONFIG_MEMORY_ISOLATION |
1648 | static inline unsigned long nr_zone_isolate_freepages(struct zone *zone) | |
1649 | { | |
1650 | if (unlikely(zone->nr_pageblock_isolate)) | |
1651 | return zone->nr_pageblock_isolate * pageblock_nr_pages; | |
1652 | return 0; | |
1653 | } | |
1654 | #else | |
1655 | static inline unsigned long nr_zone_isolate_freepages(struct zone *zone) | |
1656 | { | |
1657 | return 0; | |
1658 | } | |
1659 | #endif | |
1660 | ||
88f5acf8 MG |
1661 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1662 | int classzone_idx, int alloc_flags) | |
1663 | { | |
1664 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1665 | zone_page_state(z, NR_FREE_PAGES)); | |
1666 | } | |
1667 | ||
1668 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1669 | int classzone_idx, int alloc_flags) | |
1670 | { | |
1671 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1672 | ||
1673 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1674 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1675 | ||
702d1a6e MK |
1676 | /* |
1677 | * If the zone has MIGRATE_ISOLATE type free pages, we should consider | |
1678 | * it. nr_zone_isolate_freepages is never accurate so kswapd might not | |
1679 | * sleep although it could do so. But this is more desirable for memory | |
1680 | * hotplug than sleeping which can cause a livelock in the direct | |
1681 | * reclaim path. | |
1682 | */ | |
1683 | free_pages -= nr_zone_isolate_freepages(z); | |
88f5acf8 MG |
1684 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
1685 | free_pages); | |
1da177e4 LT |
1686 | } |
1687 | ||
9276b1bc PJ |
1688 | #ifdef CONFIG_NUMA |
1689 | /* | |
1690 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1691 | * skip over zones that are not allowed by the cpuset, or that have | |
1692 | * been recently (in last second) found to be nearly full. See further | |
1693 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1694 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1695 | * |
1696 | * If the zonelist cache is present in the passed in zonelist, then | |
1697 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1698 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1699 | * |
1700 | * If the zonelist cache is not available for this zonelist, does | |
1701 | * nothing and returns NULL. | |
1702 | * | |
1703 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1704 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1705 | * | |
1706 | * We hold off even calling zlc_setup, until after we've checked the | |
1707 | * first zone in the zonelist, on the theory that most allocations will | |
1708 | * be satisfied from that first zone, so best to examine that zone as | |
1709 | * quickly as we can. | |
1710 | */ | |
1711 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1712 | { | |
1713 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1714 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1715 | ||
1716 | zlc = zonelist->zlcache_ptr; | |
1717 | if (!zlc) | |
1718 | return NULL; | |
1719 | ||
f05111f5 | 1720 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1721 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1722 | zlc->last_full_zap = jiffies; | |
1723 | } | |
1724 | ||
1725 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1726 | &cpuset_current_mems_allowed : | |
37b07e41 | 1727 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1728 | return allowednodes; |
1729 | } | |
1730 | ||
1731 | /* | |
1732 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1733 | * if it is worth looking at further for free memory: | |
1734 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1735 | * bit set in the zonelist_cache fullzones BITMAP). | |
1736 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1737 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1738 | * Return true (non-zero) if zone is worth looking at further, or | |
1739 | * else return false (zero) if it is not. | |
1740 | * | |
1741 | * This check -ignores- the distinction between various watermarks, | |
1742 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1743 | * found to be full for any variation of these watermarks, it will | |
1744 | * be considered full for up to one second by all requests, unless | |
1745 | * we are so low on memory on all allowed nodes that we are forced | |
1746 | * into the second scan of the zonelist. | |
1747 | * | |
1748 | * In the second scan we ignore this zonelist cache and exactly | |
1749 | * apply the watermarks to all zones, even it is slower to do so. | |
1750 | * We are low on memory in the second scan, and should leave no stone | |
1751 | * unturned looking for a free page. | |
1752 | */ | |
dd1a239f | 1753 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1754 | nodemask_t *allowednodes) |
1755 | { | |
1756 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1757 | int i; /* index of *z in zonelist zones */ | |
1758 | int n; /* node that zone *z is on */ | |
1759 | ||
1760 | zlc = zonelist->zlcache_ptr; | |
1761 | if (!zlc) | |
1762 | return 1; | |
1763 | ||
dd1a239f | 1764 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1765 | n = zlc->z_to_n[i]; |
1766 | ||
1767 | /* This zone is worth trying if it is allowed but not full */ | |
1768 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1769 | } | |
1770 | ||
1771 | /* | |
1772 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1773 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1774 | * from that zone don't waste time re-examining it. | |
1775 | */ | |
dd1a239f | 1776 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1777 | { |
1778 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1779 | int i; /* index of *z in zonelist zones */ | |
1780 | ||
1781 | zlc = zonelist->zlcache_ptr; | |
1782 | if (!zlc) | |
1783 | return; | |
1784 | ||
dd1a239f | 1785 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1786 | |
1787 | set_bit(i, zlc->fullzones); | |
1788 | } | |
1789 | ||
76d3fbf8 MG |
1790 | /* |
1791 | * clear all zones full, called after direct reclaim makes progress so that | |
1792 | * a zone that was recently full is not skipped over for up to a second | |
1793 | */ | |
1794 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1795 | { | |
1796 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1797 | ||
1798 | zlc = zonelist->zlcache_ptr; | |
1799 | if (!zlc) | |
1800 | return; | |
1801 | ||
1802 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1803 | } | |
1804 | ||
957f822a DR |
1805 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1806 | { | |
1807 | return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes); | |
1808 | } | |
1809 | ||
1810 | static void __paginginit init_zone_allows_reclaim(int nid) | |
1811 | { | |
1812 | int i; | |
1813 | ||
1814 | for_each_online_node(i) | |
6b187d02 | 1815 | if (node_distance(nid, i) <= RECLAIM_DISTANCE) |
957f822a | 1816 | node_set(i, NODE_DATA(nid)->reclaim_nodes); |
6b187d02 | 1817 | else |
957f822a | 1818 | zone_reclaim_mode = 1; |
957f822a DR |
1819 | } |
1820 | ||
9276b1bc PJ |
1821 | #else /* CONFIG_NUMA */ |
1822 | ||
1823 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1824 | { | |
1825 | return NULL; | |
1826 | } | |
1827 | ||
dd1a239f | 1828 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1829 | nodemask_t *allowednodes) |
1830 | { | |
1831 | return 1; | |
1832 | } | |
1833 | ||
dd1a239f | 1834 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1835 | { |
1836 | } | |
76d3fbf8 MG |
1837 | |
1838 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1839 | { | |
1840 | } | |
957f822a DR |
1841 | |
1842 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) | |
1843 | { | |
1844 | return true; | |
1845 | } | |
1846 | ||
1847 | static inline void init_zone_allows_reclaim(int nid) | |
1848 | { | |
1849 | } | |
9276b1bc PJ |
1850 | #endif /* CONFIG_NUMA */ |
1851 | ||
7fb1d9fc | 1852 | /* |
0798e519 | 1853 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1854 | * a page. |
1855 | */ | |
1856 | static struct page * | |
19770b32 | 1857 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1858 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1859 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1860 | { |
dd1a239f | 1861 | struct zoneref *z; |
7fb1d9fc | 1862 | struct page *page = NULL; |
54a6eb5c | 1863 | int classzone_idx; |
5117f45d | 1864 | struct zone *zone; |
9276b1bc PJ |
1865 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1866 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1867 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1868 | |
19770b32 | 1869 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1870 | zonelist_scan: |
7fb1d9fc | 1871 | /* |
9276b1bc | 1872 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1873 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1874 | */ | |
19770b32 MG |
1875 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1876 | high_zoneidx, nodemask) { | |
e5adfffc | 1877 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1878 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1879 | continue; | |
7fb1d9fc | 1880 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1881 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1882 | continue; |
a756cf59 JW |
1883 | /* |
1884 | * When allocating a page cache page for writing, we | |
1885 | * want to get it from a zone that is within its dirty | |
1886 | * limit, such that no single zone holds more than its | |
1887 | * proportional share of globally allowed dirty pages. | |
1888 | * The dirty limits take into account the zone's | |
1889 | * lowmem reserves and high watermark so that kswapd | |
1890 | * should be able to balance it without having to | |
1891 | * write pages from its LRU list. | |
1892 | * | |
1893 | * This may look like it could increase pressure on | |
1894 | * lower zones by failing allocations in higher zones | |
1895 | * before they are full. But the pages that do spill | |
1896 | * over are limited as the lower zones are protected | |
1897 | * by this very same mechanism. It should not become | |
1898 | * a practical burden to them. | |
1899 | * | |
1900 | * XXX: For now, allow allocations to potentially | |
1901 | * exceed the per-zone dirty limit in the slowpath | |
1902 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1903 | * which is important when on a NUMA setup the allowed | |
1904 | * zones are together not big enough to reach the | |
1905 | * global limit. The proper fix for these situations | |
1906 | * will require awareness of zones in the | |
1907 | * dirty-throttling and the flusher threads. | |
1908 | */ | |
1909 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1910 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
1911 | goto this_zone_full; | |
7fb1d9fc | 1912 | |
41858966 | 1913 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1914 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1915 | unsigned long mark; |
fa5e084e MG |
1916 | int ret; |
1917 | ||
41858966 | 1918 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1919 | if (zone_watermark_ok(zone, order, mark, |
1920 | classzone_idx, alloc_flags)) | |
1921 | goto try_this_zone; | |
1922 | ||
e5adfffc KS |
1923 | if (IS_ENABLED(CONFIG_NUMA) && |
1924 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
1925 | /* |
1926 | * we do zlc_setup if there are multiple nodes | |
1927 | * and before considering the first zone allowed | |
1928 | * by the cpuset. | |
1929 | */ | |
1930 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1931 | zlc_active = 1; | |
1932 | did_zlc_setup = 1; | |
1933 | } | |
1934 | ||
957f822a DR |
1935 | if (zone_reclaim_mode == 0 || |
1936 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
1937 | goto this_zone_full; |
1938 | ||
cd38b115 MG |
1939 | /* |
1940 | * As we may have just activated ZLC, check if the first | |
1941 | * eligible zone has failed zone_reclaim recently. | |
1942 | */ | |
e5adfffc | 1943 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
1944 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1945 | continue; | |
1946 | ||
fa5e084e MG |
1947 | ret = zone_reclaim(zone, gfp_mask, order); |
1948 | switch (ret) { | |
1949 | case ZONE_RECLAIM_NOSCAN: | |
1950 | /* did not scan */ | |
cd38b115 | 1951 | continue; |
fa5e084e MG |
1952 | case ZONE_RECLAIM_FULL: |
1953 | /* scanned but unreclaimable */ | |
cd38b115 | 1954 | continue; |
fa5e084e MG |
1955 | default: |
1956 | /* did we reclaim enough */ | |
1957 | if (!zone_watermark_ok(zone, order, mark, | |
1958 | classzone_idx, alloc_flags)) | |
9276b1bc | 1959 | goto this_zone_full; |
0798e519 | 1960 | } |
7fb1d9fc RS |
1961 | } |
1962 | ||
fa5e084e | 1963 | try_this_zone: |
3dd28266 MG |
1964 | page = buffered_rmqueue(preferred_zone, zone, order, |
1965 | gfp_mask, migratetype); | |
0798e519 | 1966 | if (page) |
7fb1d9fc | 1967 | break; |
9276b1bc | 1968 | this_zone_full: |
e5adfffc | 1969 | if (IS_ENABLED(CONFIG_NUMA)) |
9276b1bc | 1970 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 1971 | } |
9276b1bc | 1972 | |
e5adfffc | 1973 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
1974 | /* Disable zlc cache for second zonelist scan */ |
1975 | zlc_active = 0; | |
1976 | goto zonelist_scan; | |
1977 | } | |
b121186a AS |
1978 | |
1979 | if (page) | |
1980 | /* | |
1981 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
1982 | * necessary to allocate the page. The expectation is | |
1983 | * that the caller is taking steps that will free more | |
1984 | * memory. The caller should avoid the page being used | |
1985 | * for !PFMEMALLOC purposes. | |
1986 | */ | |
1987 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
1988 | ||
7fb1d9fc | 1989 | return page; |
753ee728 MH |
1990 | } |
1991 | ||
29423e77 DR |
1992 | /* |
1993 | * Large machines with many possible nodes should not always dump per-node | |
1994 | * meminfo in irq context. | |
1995 | */ | |
1996 | static inline bool should_suppress_show_mem(void) | |
1997 | { | |
1998 | bool ret = false; | |
1999 | ||
2000 | #if NODES_SHIFT > 8 | |
2001 | ret = in_interrupt(); | |
2002 | #endif | |
2003 | return ret; | |
2004 | } | |
2005 | ||
a238ab5b DH |
2006 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2007 | DEFAULT_RATELIMIT_INTERVAL, | |
2008 | DEFAULT_RATELIMIT_BURST); | |
2009 | ||
2010 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2011 | { | |
a238ab5b DH |
2012 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2013 | ||
c0a32fc5 SG |
2014 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2015 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2016 | return; |
2017 | ||
2018 | /* | |
2019 | * This documents exceptions given to allocations in certain | |
2020 | * contexts that are allowed to allocate outside current's set | |
2021 | * of allowed nodes. | |
2022 | */ | |
2023 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2024 | if (test_thread_flag(TIF_MEMDIE) || | |
2025 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2026 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2027 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2028 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2029 | ||
2030 | if (fmt) { | |
3ee9a4f0 JP |
2031 | struct va_format vaf; |
2032 | va_list args; | |
2033 | ||
a238ab5b | 2034 | va_start(args, fmt); |
3ee9a4f0 JP |
2035 | |
2036 | vaf.fmt = fmt; | |
2037 | vaf.va = &args; | |
2038 | ||
2039 | pr_warn("%pV", &vaf); | |
2040 | ||
a238ab5b DH |
2041 | va_end(args); |
2042 | } | |
2043 | ||
3ee9a4f0 JP |
2044 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2045 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2046 | |
2047 | dump_stack(); | |
2048 | if (!should_suppress_show_mem()) | |
2049 | show_mem(filter); | |
2050 | } | |
2051 | ||
11e33f6a MG |
2052 | static inline int |
2053 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2054 | unsigned long did_some_progress, |
11e33f6a | 2055 | unsigned long pages_reclaimed) |
1da177e4 | 2056 | { |
11e33f6a MG |
2057 | /* Do not loop if specifically requested */ |
2058 | if (gfp_mask & __GFP_NORETRY) | |
2059 | return 0; | |
1da177e4 | 2060 | |
f90ac398 MG |
2061 | /* Always retry if specifically requested */ |
2062 | if (gfp_mask & __GFP_NOFAIL) | |
2063 | return 1; | |
2064 | ||
2065 | /* | |
2066 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2067 | * making forward progress without invoking OOM. Suspend also disables | |
2068 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2069 | */ | |
2070 | if (!did_some_progress && pm_suspended_storage()) | |
2071 | return 0; | |
2072 | ||
11e33f6a MG |
2073 | /* |
2074 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2075 | * means __GFP_NOFAIL, but that may not be true in other | |
2076 | * implementations. | |
2077 | */ | |
2078 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2079 | return 1; | |
2080 | ||
2081 | /* | |
2082 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2083 | * specified, then we retry until we no longer reclaim any pages | |
2084 | * (above), or we've reclaimed an order of pages at least as | |
2085 | * large as the allocation's order. In both cases, if the | |
2086 | * allocation still fails, we stop retrying. | |
2087 | */ | |
2088 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2089 | return 1; | |
cf40bd16 | 2090 | |
11e33f6a MG |
2091 | return 0; |
2092 | } | |
933e312e | 2093 | |
11e33f6a MG |
2094 | static inline struct page * |
2095 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2096 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2097 | nodemask_t *nodemask, struct zone *preferred_zone, |
2098 | int migratetype) | |
11e33f6a MG |
2099 | { |
2100 | struct page *page; | |
2101 | ||
2102 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2103 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2104 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2105 | return NULL; |
2106 | } | |
6b1de916 | 2107 | |
11e33f6a MG |
2108 | /* |
2109 | * Go through the zonelist yet one more time, keep very high watermark | |
2110 | * here, this is only to catch a parallel oom killing, we must fail if | |
2111 | * we're still under heavy pressure. | |
2112 | */ | |
2113 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2114 | order, zonelist, high_zoneidx, | |
5117f45d | 2115 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 2116 | preferred_zone, migratetype); |
7fb1d9fc | 2117 | if (page) |
11e33f6a MG |
2118 | goto out; |
2119 | ||
4365a567 KH |
2120 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2121 | /* The OOM killer will not help higher order allocs */ | |
2122 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2123 | goto out; | |
03668b3c DR |
2124 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2125 | if (high_zoneidx < ZONE_NORMAL) | |
2126 | goto out; | |
4365a567 KH |
2127 | /* |
2128 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2129 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2130 | * The caller should handle page allocation failure by itself if | |
2131 | * it specifies __GFP_THISNODE. | |
2132 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2133 | */ | |
2134 | if (gfp_mask & __GFP_THISNODE) | |
2135 | goto out; | |
2136 | } | |
11e33f6a | 2137 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2138 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2139 | |
2140 | out: | |
2141 | clear_zonelist_oom(zonelist, gfp_mask); | |
2142 | return page; | |
2143 | } | |
2144 | ||
56de7263 MG |
2145 | #ifdef CONFIG_COMPACTION |
2146 | /* Try memory compaction for high-order allocations before reclaim */ | |
2147 | static struct page * | |
2148 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2149 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2150 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2151 | int migratetype, bool sync_migration, |
c67fe375 | 2152 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2153 | unsigned long *did_some_progress) |
56de7263 | 2154 | { |
1fb3f8ca | 2155 | struct page *page = NULL; |
56de7263 | 2156 | |
66199712 | 2157 | if (!order) |
56de7263 MG |
2158 | return NULL; |
2159 | ||
aff62249 | 2160 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2161 | *deferred_compaction = true; |
2162 | return NULL; | |
2163 | } | |
2164 | ||
c06b1fca | 2165 | current->flags |= PF_MEMALLOC; |
56de7263 | 2166 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
c67fe375 | 2167 | nodemask, sync_migration, |
1fb3f8ca | 2168 | contended_compaction, &page); |
c06b1fca | 2169 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2170 | |
1fb3f8ca MG |
2171 | /* If compaction captured a page, prep and use it */ |
2172 | if (page) { | |
2173 | prep_new_page(page, order, gfp_mask); | |
2174 | goto got_page; | |
2175 | } | |
2176 | ||
2177 | if (*did_some_progress != COMPACT_SKIPPED) { | |
56de7263 MG |
2178 | /* Page migration frees to the PCP lists but we want merging */ |
2179 | drain_pages(get_cpu()); | |
2180 | put_cpu(); | |
2181 | ||
2182 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2183 | order, zonelist, high_zoneidx, | |
cfd19c5a MG |
2184 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2185 | preferred_zone, migratetype); | |
56de7263 | 2186 | if (page) { |
1fb3f8ca | 2187 | got_page: |
62997027 | 2188 | preferred_zone->compact_blockskip_flush = false; |
4f92e258 MG |
2189 | preferred_zone->compact_considered = 0; |
2190 | preferred_zone->compact_defer_shift = 0; | |
aff62249 RR |
2191 | if (order >= preferred_zone->compact_order_failed) |
2192 | preferred_zone->compact_order_failed = order + 1; | |
56de7263 MG |
2193 | count_vm_event(COMPACTSUCCESS); |
2194 | return page; | |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * It's bad if compaction run occurs and fails. | |
2199 | * The most likely reason is that pages exist, | |
2200 | * but not enough to satisfy watermarks. | |
2201 | */ | |
2202 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2203 | |
2204 | /* | |
2205 | * As async compaction considers a subset of pageblocks, only | |
2206 | * defer if the failure was a sync compaction failure. | |
2207 | */ | |
2208 | if (sync_migration) | |
aff62249 | 2209 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2210 | |
2211 | cond_resched(); | |
2212 | } | |
2213 | ||
2214 | return NULL; | |
2215 | } | |
2216 | #else | |
2217 | static inline struct page * | |
2218 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2219 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2220 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2221 | int migratetype, bool sync_migration, |
c67fe375 | 2222 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2223 | unsigned long *did_some_progress) |
56de7263 MG |
2224 | { |
2225 | return NULL; | |
2226 | } | |
2227 | #endif /* CONFIG_COMPACTION */ | |
2228 | ||
bba90710 MS |
2229 | /* Perform direct synchronous page reclaim */ |
2230 | static int | |
2231 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2232 | nodemask_t *nodemask) | |
11e33f6a | 2233 | { |
11e33f6a | 2234 | struct reclaim_state reclaim_state; |
bba90710 | 2235 | int progress; |
11e33f6a MG |
2236 | |
2237 | cond_resched(); | |
2238 | ||
2239 | /* We now go into synchronous reclaim */ | |
2240 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2241 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2242 | lockdep_set_current_reclaim_state(gfp_mask); |
2243 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2244 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2245 | |
bba90710 | 2246 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2247 | |
c06b1fca | 2248 | current->reclaim_state = NULL; |
11e33f6a | 2249 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2250 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2251 | |
2252 | cond_resched(); | |
2253 | ||
bba90710 MS |
2254 | return progress; |
2255 | } | |
2256 | ||
2257 | /* The really slow allocator path where we enter direct reclaim */ | |
2258 | static inline struct page * | |
2259 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2260 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2261 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
2262 | int migratetype, unsigned long *did_some_progress) | |
2263 | { | |
2264 | struct page *page = NULL; | |
2265 | bool drained = false; | |
2266 | ||
2267 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2268 | nodemask); | |
9ee493ce MG |
2269 | if (unlikely(!(*did_some_progress))) |
2270 | return NULL; | |
11e33f6a | 2271 | |
76d3fbf8 | 2272 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2273 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2274 | zlc_clear_zones_full(zonelist); |
2275 | ||
9ee493ce MG |
2276 | retry: |
2277 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2278 | zonelist, high_zoneidx, |
cfd19c5a MG |
2279 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2280 | preferred_zone, migratetype); | |
9ee493ce MG |
2281 | |
2282 | /* | |
2283 | * If an allocation failed after direct reclaim, it could be because | |
2284 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2285 | */ | |
2286 | if (!page && !drained) { | |
2287 | drain_all_pages(); | |
2288 | drained = true; | |
2289 | goto retry; | |
2290 | } | |
2291 | ||
11e33f6a MG |
2292 | return page; |
2293 | } | |
2294 | ||
1da177e4 | 2295 | /* |
11e33f6a MG |
2296 | * This is called in the allocator slow-path if the allocation request is of |
2297 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2298 | */ |
11e33f6a MG |
2299 | static inline struct page * |
2300 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2301 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2302 | nodemask_t *nodemask, struct zone *preferred_zone, |
2303 | int migratetype) | |
11e33f6a MG |
2304 | { |
2305 | struct page *page; | |
2306 | ||
2307 | do { | |
2308 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2309 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2310 | preferred_zone, migratetype); |
11e33f6a MG |
2311 | |
2312 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2313 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2314 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2315 | ||
2316 | return page; | |
2317 | } | |
2318 | ||
2319 | static inline | |
2320 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
99504748 MG |
2321 | enum zone_type high_zoneidx, |
2322 | enum zone_type classzone_idx) | |
1da177e4 | 2323 | { |
dd1a239f MG |
2324 | struct zoneref *z; |
2325 | struct zone *zone; | |
1da177e4 | 2326 | |
11e33f6a | 2327 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
99504748 | 2328 | wakeup_kswapd(zone, order, classzone_idx); |
11e33f6a | 2329 | } |
cf40bd16 | 2330 | |
341ce06f PZ |
2331 | static inline int |
2332 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2333 | { | |
341ce06f PZ |
2334 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2335 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2336 | |
a56f57ff | 2337 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2338 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2339 | |
341ce06f PZ |
2340 | /* |
2341 | * The caller may dip into page reserves a bit more if the caller | |
2342 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2343 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2344 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2345 | */ | |
e6223a3b | 2346 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2347 | |
341ce06f | 2348 | if (!wait) { |
5c3240d9 AA |
2349 | /* |
2350 | * Not worth trying to allocate harder for | |
2351 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2352 | */ | |
2353 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2354 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2355 | /* |
341ce06f PZ |
2356 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2357 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2358 | */ |
341ce06f | 2359 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2360 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2361 | alloc_flags |= ALLOC_HARDER; |
2362 | ||
b37f1dd0 MG |
2363 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2364 | if (gfp_mask & __GFP_MEMALLOC) | |
2365 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2366 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2367 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2368 | else if (!in_interrupt() && | |
2369 | ((current->flags & PF_MEMALLOC) || | |
2370 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2371 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2372 | } |
d95ea5d1 BZ |
2373 | #ifdef CONFIG_CMA |
2374 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2375 | alloc_flags |= ALLOC_CMA; | |
2376 | #endif | |
341ce06f PZ |
2377 | return alloc_flags; |
2378 | } | |
2379 | ||
072bb0aa MG |
2380 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2381 | { | |
b37f1dd0 | 2382 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2383 | } |
2384 | ||
11e33f6a MG |
2385 | static inline struct page * |
2386 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2387 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2388 | nodemask_t *nodemask, struct zone *preferred_zone, |
2389 | int migratetype) | |
11e33f6a MG |
2390 | { |
2391 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2392 | struct page *page = NULL; | |
2393 | int alloc_flags; | |
2394 | unsigned long pages_reclaimed = 0; | |
2395 | unsigned long did_some_progress; | |
77f1fe6b | 2396 | bool sync_migration = false; |
66199712 | 2397 | bool deferred_compaction = false; |
c67fe375 | 2398 | bool contended_compaction = false; |
1da177e4 | 2399 | |
72807a74 MG |
2400 | /* |
2401 | * In the slowpath, we sanity check order to avoid ever trying to | |
2402 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2403 | * be using allocators in order of preference for an area that is | |
2404 | * too large. | |
2405 | */ | |
1fc28b70 MG |
2406 | if (order >= MAX_ORDER) { |
2407 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2408 | return NULL; |
1fc28b70 | 2409 | } |
1da177e4 | 2410 | |
952f3b51 CL |
2411 | /* |
2412 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2413 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2414 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2415 | * using a larger set of nodes after it has established that the | |
2416 | * allowed per node queues are empty and that nodes are | |
2417 | * over allocated. | |
2418 | */ | |
e5adfffc KS |
2419 | if (IS_ENABLED(CONFIG_NUMA) && |
2420 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2421 | goto nopage; |
2422 | ||
cc4a6851 | 2423 | restart: |
caf49191 LT |
2424 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2425 | wake_all_kswapd(order, zonelist, high_zoneidx, | |
2426 | zone_idx(preferred_zone)); | |
1da177e4 | 2427 | |
9bf2229f | 2428 | /* |
7fb1d9fc RS |
2429 | * OK, we're below the kswapd watermark and have kicked background |
2430 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2431 | * to how we want to proceed. | |
9bf2229f | 2432 | */ |
341ce06f | 2433 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2434 | |
f33261d7 DR |
2435 | /* |
2436 | * Find the true preferred zone if the allocation is unconstrained by | |
2437 | * cpusets. | |
2438 | */ | |
2439 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2440 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2441 | &preferred_zone); | |
2442 | ||
cfa54a0f | 2443 | rebalance: |
341ce06f | 2444 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2445 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2446 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2447 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2448 | if (page) |
2449 | goto got_pg; | |
1da177e4 | 2450 | |
11e33f6a | 2451 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2452 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2453 | /* |
2454 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2455 | * the allocation is high priority and these type of | |
2456 | * allocations are system rather than user orientated | |
2457 | */ | |
2458 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2459 | ||
341ce06f PZ |
2460 | page = __alloc_pages_high_priority(gfp_mask, order, |
2461 | zonelist, high_zoneidx, nodemask, | |
2462 | preferred_zone, migratetype); | |
cfd19c5a | 2463 | if (page) { |
341ce06f | 2464 | goto got_pg; |
cfd19c5a | 2465 | } |
1da177e4 LT |
2466 | } |
2467 | ||
2468 | /* Atomic allocations - we can't balance anything */ | |
2469 | if (!wait) | |
2470 | goto nopage; | |
2471 | ||
341ce06f | 2472 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2473 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2474 | goto nopage; |
2475 | ||
6583bb64 DR |
2476 | /* Avoid allocations with no watermarks from looping endlessly */ |
2477 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2478 | goto nopage; | |
2479 | ||
77f1fe6b MG |
2480 | /* |
2481 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2482 | * attempts after direct reclaim are synchronous | |
2483 | */ | |
56de7263 MG |
2484 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2485 | zonelist, high_zoneidx, | |
2486 | nodemask, | |
2487 | alloc_flags, preferred_zone, | |
66199712 | 2488 | migratetype, sync_migration, |
c67fe375 | 2489 | &contended_compaction, |
66199712 MG |
2490 | &deferred_compaction, |
2491 | &did_some_progress); | |
56de7263 MG |
2492 | if (page) |
2493 | goto got_pg; | |
c6a140bf | 2494 | sync_migration = true; |
56de7263 | 2495 | |
31f8d42d LT |
2496 | /* |
2497 | * If compaction is deferred for high-order allocations, it is because | |
2498 | * sync compaction recently failed. In this is the case and the caller | |
2499 | * requested a movable allocation that does not heavily disrupt the | |
2500 | * system then fail the allocation instead of entering direct reclaim. | |
2501 | */ | |
2502 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2503 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2504 | goto nopage; |
66199712 | 2505 | |
11e33f6a MG |
2506 | /* Try direct reclaim and then allocating */ |
2507 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2508 | zonelist, high_zoneidx, | |
2509 | nodemask, | |
5117f45d | 2510 | alloc_flags, preferred_zone, |
3dd28266 | 2511 | migratetype, &did_some_progress); |
11e33f6a MG |
2512 | if (page) |
2513 | goto got_pg; | |
1da177e4 | 2514 | |
e33c3b5e | 2515 | /* |
11e33f6a MG |
2516 | * If we failed to make any progress reclaiming, then we are |
2517 | * running out of options and have to consider going OOM | |
e33c3b5e | 2518 | */ |
11e33f6a MG |
2519 | if (!did_some_progress) { |
2520 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
2521 | if (oom_killer_disabled) |
2522 | goto nopage; | |
29fd66d2 DR |
2523 | /* Coredumps can quickly deplete all memory reserves */ |
2524 | if ((current->flags & PF_DUMPCORE) && | |
2525 | !(gfp_mask & __GFP_NOFAIL)) | |
2526 | goto nopage; | |
11e33f6a MG |
2527 | page = __alloc_pages_may_oom(gfp_mask, order, |
2528 | zonelist, high_zoneidx, | |
3dd28266 MG |
2529 | nodemask, preferred_zone, |
2530 | migratetype); | |
11e33f6a MG |
2531 | if (page) |
2532 | goto got_pg; | |
1da177e4 | 2533 | |
03668b3c DR |
2534 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2535 | /* | |
2536 | * The oom killer is not called for high-order | |
2537 | * allocations that may fail, so if no progress | |
2538 | * is being made, there are no other options and | |
2539 | * retrying is unlikely to help. | |
2540 | */ | |
2541 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2542 | goto nopage; | |
2543 | /* | |
2544 | * The oom killer is not called for lowmem | |
2545 | * allocations to prevent needlessly killing | |
2546 | * innocent tasks. | |
2547 | */ | |
2548 | if (high_zoneidx < ZONE_NORMAL) | |
2549 | goto nopage; | |
2550 | } | |
e2c55dc8 | 2551 | |
ff0ceb9d DR |
2552 | goto restart; |
2553 | } | |
1da177e4 LT |
2554 | } |
2555 | ||
11e33f6a | 2556 | /* Check if we should retry the allocation */ |
a41f24ea | 2557 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2558 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2559 | pages_reclaimed)) { | |
11e33f6a | 2560 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2561 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2562 | goto rebalance; |
3e7d3449 MG |
2563 | } else { |
2564 | /* | |
2565 | * High-order allocations do not necessarily loop after | |
2566 | * direct reclaim and reclaim/compaction depends on compaction | |
2567 | * being called after reclaim so call directly if necessary | |
2568 | */ | |
2569 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2570 | zonelist, high_zoneidx, | |
2571 | nodemask, | |
2572 | alloc_flags, preferred_zone, | |
66199712 | 2573 | migratetype, sync_migration, |
c67fe375 | 2574 | &contended_compaction, |
66199712 MG |
2575 | &deferred_compaction, |
2576 | &did_some_progress); | |
3e7d3449 MG |
2577 | if (page) |
2578 | goto got_pg; | |
1da177e4 LT |
2579 | } |
2580 | ||
2581 | nopage: | |
a238ab5b | 2582 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2583 | return page; |
1da177e4 | 2584 | got_pg: |
b1eeab67 VN |
2585 | if (kmemcheck_enabled) |
2586 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2587 | |
072bb0aa | 2588 | return page; |
1da177e4 | 2589 | } |
11e33f6a MG |
2590 | |
2591 | /* | |
2592 | * This is the 'heart' of the zoned buddy allocator. | |
2593 | */ | |
2594 | struct page * | |
2595 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2596 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2597 | { | |
2598 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2599 | struct zone *preferred_zone; |
cc9a6c87 | 2600 | struct page *page = NULL; |
3dd28266 | 2601 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2602 | unsigned int cpuset_mems_cookie; |
d95ea5d1 | 2603 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET; |
11e33f6a | 2604 | |
dcce284a BH |
2605 | gfp_mask &= gfp_allowed_mask; |
2606 | ||
11e33f6a MG |
2607 | lockdep_trace_alloc(gfp_mask); |
2608 | ||
2609 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2610 | ||
2611 | if (should_fail_alloc_page(gfp_mask, order)) | |
2612 | return NULL; | |
2613 | ||
2614 | /* | |
2615 | * Check the zones suitable for the gfp_mask contain at least one | |
2616 | * valid zone. It's possible to have an empty zonelist as a result | |
2617 | * of GFP_THISNODE and a memoryless node | |
2618 | */ | |
2619 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2620 | return NULL; | |
2621 | ||
cc9a6c87 MG |
2622 | retry_cpuset: |
2623 | cpuset_mems_cookie = get_mems_allowed(); | |
2624 | ||
5117f45d | 2625 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2626 | first_zones_zonelist(zonelist, high_zoneidx, |
2627 | nodemask ? : &cpuset_current_mems_allowed, | |
2628 | &preferred_zone); | |
cc9a6c87 MG |
2629 | if (!preferred_zone) |
2630 | goto out; | |
5117f45d | 2631 | |
d95ea5d1 BZ |
2632 | #ifdef CONFIG_CMA |
2633 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2634 | alloc_flags |= ALLOC_CMA; | |
2635 | #endif | |
5117f45d | 2636 | /* First allocation attempt */ |
11e33f6a | 2637 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2638 | zonelist, high_zoneidx, alloc_flags, |
3dd28266 | 2639 | preferred_zone, migratetype); |
11e33f6a MG |
2640 | if (unlikely(!page)) |
2641 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 2642 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2643 | preferred_zone, migratetype); |
11e33f6a | 2644 | |
4b4f278c | 2645 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2646 | |
2647 | out: | |
2648 | /* | |
2649 | * When updating a task's mems_allowed, it is possible to race with | |
2650 | * parallel threads in such a way that an allocation can fail while | |
2651 | * the mask is being updated. If a page allocation is about to fail, | |
2652 | * check if the cpuset changed during allocation and if so, retry. | |
2653 | */ | |
2654 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2655 | goto retry_cpuset; | |
2656 | ||
11e33f6a | 2657 | return page; |
1da177e4 | 2658 | } |
d239171e | 2659 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2660 | |
2661 | /* | |
2662 | * Common helper functions. | |
2663 | */ | |
920c7a5d | 2664 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2665 | { |
945a1113 AM |
2666 | struct page *page; |
2667 | ||
2668 | /* | |
2669 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2670 | * a highmem page | |
2671 | */ | |
2672 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2673 | ||
1da177e4 LT |
2674 | page = alloc_pages(gfp_mask, order); |
2675 | if (!page) | |
2676 | return 0; | |
2677 | return (unsigned long) page_address(page); | |
2678 | } | |
1da177e4 LT |
2679 | EXPORT_SYMBOL(__get_free_pages); |
2680 | ||
920c7a5d | 2681 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2682 | { |
945a1113 | 2683 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2684 | } |
1da177e4 LT |
2685 | EXPORT_SYMBOL(get_zeroed_page); |
2686 | ||
920c7a5d | 2687 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2688 | { |
b5810039 | 2689 | if (put_page_testzero(page)) { |
1da177e4 | 2690 | if (order == 0) |
fc91668e | 2691 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2692 | else |
2693 | __free_pages_ok(page, order); | |
2694 | } | |
2695 | } | |
2696 | ||
2697 | EXPORT_SYMBOL(__free_pages); | |
2698 | ||
920c7a5d | 2699 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2700 | { |
2701 | if (addr != 0) { | |
725d704e | 2702 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2703 | __free_pages(virt_to_page((void *)addr), order); |
2704 | } | |
2705 | } | |
2706 | ||
2707 | EXPORT_SYMBOL(free_pages); | |
2708 | ||
ee85c2e1 AK |
2709 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2710 | { | |
2711 | if (addr) { | |
2712 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2713 | unsigned long used = addr + PAGE_ALIGN(size); | |
2714 | ||
2715 | split_page(virt_to_page((void *)addr), order); | |
2716 | while (used < alloc_end) { | |
2717 | free_page(used); | |
2718 | used += PAGE_SIZE; | |
2719 | } | |
2720 | } | |
2721 | return (void *)addr; | |
2722 | } | |
2723 | ||
2be0ffe2 TT |
2724 | /** |
2725 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2726 | * @size: the number of bytes to allocate | |
2727 | * @gfp_mask: GFP flags for the allocation | |
2728 | * | |
2729 | * This function is similar to alloc_pages(), except that it allocates the | |
2730 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2731 | * allocate memory in power-of-two pages. | |
2732 | * | |
2733 | * This function is also limited by MAX_ORDER. | |
2734 | * | |
2735 | * Memory allocated by this function must be released by free_pages_exact(). | |
2736 | */ | |
2737 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2738 | { | |
2739 | unsigned int order = get_order(size); | |
2740 | unsigned long addr; | |
2741 | ||
2742 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2743 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2744 | } |
2745 | EXPORT_SYMBOL(alloc_pages_exact); | |
2746 | ||
ee85c2e1 AK |
2747 | /** |
2748 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2749 | * pages on a node. | |
b5e6ab58 | 2750 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2751 | * @size: the number of bytes to allocate |
2752 | * @gfp_mask: GFP flags for the allocation | |
2753 | * | |
2754 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2755 | * back. | |
2756 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2757 | * but is not exact. | |
2758 | */ | |
2759 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2760 | { | |
2761 | unsigned order = get_order(size); | |
2762 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2763 | if (!p) | |
2764 | return NULL; | |
2765 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2766 | } | |
2767 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2768 | ||
2be0ffe2 TT |
2769 | /** |
2770 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2771 | * @virt: the value returned by alloc_pages_exact. | |
2772 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2773 | * | |
2774 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2775 | */ | |
2776 | void free_pages_exact(void *virt, size_t size) | |
2777 | { | |
2778 | unsigned long addr = (unsigned long)virt; | |
2779 | unsigned long end = addr + PAGE_ALIGN(size); | |
2780 | ||
2781 | while (addr < end) { | |
2782 | free_page(addr); | |
2783 | addr += PAGE_SIZE; | |
2784 | } | |
2785 | } | |
2786 | EXPORT_SYMBOL(free_pages_exact); | |
2787 | ||
1da177e4 LT |
2788 | static unsigned int nr_free_zone_pages(int offset) |
2789 | { | |
dd1a239f | 2790 | struct zoneref *z; |
54a6eb5c MG |
2791 | struct zone *zone; |
2792 | ||
e310fd43 | 2793 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2794 | unsigned int sum = 0; |
2795 | ||
0e88460d | 2796 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2797 | |
54a6eb5c | 2798 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2799 | unsigned long size = zone->present_pages; |
41858966 | 2800 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2801 | if (size > high) |
2802 | sum += size - high; | |
1da177e4 LT |
2803 | } |
2804 | ||
2805 | return sum; | |
2806 | } | |
2807 | ||
2808 | /* | |
2809 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2810 | */ | |
2811 | unsigned int nr_free_buffer_pages(void) | |
2812 | { | |
af4ca457 | 2813 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2814 | } |
c2f1a551 | 2815 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2816 | |
2817 | /* | |
2818 | * Amount of free RAM allocatable within all zones | |
2819 | */ | |
2820 | unsigned int nr_free_pagecache_pages(void) | |
2821 | { | |
2a1e274a | 2822 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2823 | } |
08e0f6a9 CL |
2824 | |
2825 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2826 | { |
e5adfffc | 2827 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 2828 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2829 | } |
1da177e4 | 2830 | |
1da177e4 LT |
2831 | void si_meminfo(struct sysinfo *val) |
2832 | { | |
2833 | val->totalram = totalram_pages; | |
2834 | val->sharedram = 0; | |
d23ad423 | 2835 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2836 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2837 | val->totalhigh = totalhigh_pages; |
2838 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2839 | val->mem_unit = PAGE_SIZE; |
2840 | } | |
2841 | ||
2842 | EXPORT_SYMBOL(si_meminfo); | |
2843 | ||
2844 | #ifdef CONFIG_NUMA | |
2845 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2846 | { | |
2847 | pg_data_t *pgdat = NODE_DATA(nid); | |
2848 | ||
2849 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2850 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2851 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2852 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2853 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2854 | NR_FREE_PAGES); | |
98d2b0eb CL |
2855 | #else |
2856 | val->totalhigh = 0; | |
2857 | val->freehigh = 0; | |
2858 | #endif | |
1da177e4 LT |
2859 | val->mem_unit = PAGE_SIZE; |
2860 | } | |
2861 | #endif | |
2862 | ||
ddd588b5 | 2863 | /* |
7bf02ea2 DR |
2864 | * Determine whether the node should be displayed or not, depending on whether |
2865 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 2866 | */ |
7bf02ea2 | 2867 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
2868 | { |
2869 | bool ret = false; | |
cc9a6c87 | 2870 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
2871 | |
2872 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
2873 | goto out; | |
2874 | ||
cc9a6c87 MG |
2875 | do { |
2876 | cpuset_mems_cookie = get_mems_allowed(); | |
2877 | ret = !node_isset(nid, cpuset_current_mems_allowed); | |
2878 | } while (!put_mems_allowed(cpuset_mems_cookie)); | |
ddd588b5 DR |
2879 | out: |
2880 | return ret; | |
2881 | } | |
2882 | ||
1da177e4 LT |
2883 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
2884 | ||
377e4f16 RV |
2885 | static void show_migration_types(unsigned char type) |
2886 | { | |
2887 | static const char types[MIGRATE_TYPES] = { | |
2888 | [MIGRATE_UNMOVABLE] = 'U', | |
2889 | [MIGRATE_RECLAIMABLE] = 'E', | |
2890 | [MIGRATE_MOVABLE] = 'M', | |
2891 | [MIGRATE_RESERVE] = 'R', | |
2892 | #ifdef CONFIG_CMA | |
2893 | [MIGRATE_CMA] = 'C', | |
2894 | #endif | |
2895 | [MIGRATE_ISOLATE] = 'I', | |
2896 | }; | |
2897 | char tmp[MIGRATE_TYPES + 1]; | |
2898 | char *p = tmp; | |
2899 | int i; | |
2900 | ||
2901 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
2902 | if (type & (1 << i)) | |
2903 | *p++ = types[i]; | |
2904 | } | |
2905 | ||
2906 | *p = '\0'; | |
2907 | printk("(%s) ", tmp); | |
2908 | } | |
2909 | ||
1da177e4 LT |
2910 | /* |
2911 | * Show free area list (used inside shift_scroll-lock stuff) | |
2912 | * We also calculate the percentage fragmentation. We do this by counting the | |
2913 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
2914 | * Suppresses nodes that are not allowed by current's cpuset if |
2915 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 2916 | */ |
7bf02ea2 | 2917 | void show_free_areas(unsigned int filter) |
1da177e4 | 2918 | { |
c7241913 | 2919 | int cpu; |
1da177e4 LT |
2920 | struct zone *zone; |
2921 | ||
ee99c71c | 2922 | for_each_populated_zone(zone) { |
7bf02ea2 | 2923 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2924 | continue; |
c7241913 JS |
2925 | show_node(zone); |
2926 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2927 | |
6b482c67 | 2928 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2929 | struct per_cpu_pageset *pageset; |
2930 | ||
99dcc3e5 | 2931 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 2932 | |
3dfa5721 CL |
2933 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2934 | cpu, pageset->pcp.high, | |
2935 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2936 | } |
2937 | } | |
2938 | ||
a731286d KM |
2939 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2940 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2941 | " unevictable:%lu" |
b76146ed | 2942 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2943 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
2944 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
2945 | " free_cma:%lu\n", | |
4f98a2fe | 2946 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2947 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2948 | global_page_state(NR_ISOLATED_ANON), |
2949 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2950 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2951 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2952 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2953 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2954 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2955 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2956 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2957 | global_page_state(NR_SLAB_RECLAIMABLE), |
2958 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2959 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2960 | global_page_state(NR_SHMEM), |
a25700a5 | 2961 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
2962 | global_page_state(NR_BOUNCE), |
2963 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 2964 | |
ee99c71c | 2965 | for_each_populated_zone(zone) { |
1da177e4 LT |
2966 | int i; |
2967 | ||
7bf02ea2 | 2968 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2969 | continue; |
1da177e4 LT |
2970 | show_node(zone); |
2971 | printk("%s" | |
2972 | " free:%lukB" | |
2973 | " min:%lukB" | |
2974 | " low:%lukB" | |
2975 | " high:%lukB" | |
4f98a2fe RR |
2976 | " active_anon:%lukB" |
2977 | " inactive_anon:%lukB" | |
2978 | " active_file:%lukB" | |
2979 | " inactive_file:%lukB" | |
7b854121 | 2980 | " unevictable:%lukB" |
a731286d KM |
2981 | " isolated(anon):%lukB" |
2982 | " isolated(file):%lukB" | |
1da177e4 | 2983 | " present:%lukB" |
4a0aa73f KM |
2984 | " mlocked:%lukB" |
2985 | " dirty:%lukB" | |
2986 | " writeback:%lukB" | |
2987 | " mapped:%lukB" | |
4b02108a | 2988 | " shmem:%lukB" |
4a0aa73f KM |
2989 | " slab_reclaimable:%lukB" |
2990 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 2991 | " kernel_stack:%lukB" |
4a0aa73f KM |
2992 | " pagetables:%lukB" |
2993 | " unstable:%lukB" | |
2994 | " bounce:%lukB" | |
d1ce749a | 2995 | " free_cma:%lukB" |
4a0aa73f | 2996 | " writeback_tmp:%lukB" |
1da177e4 LT |
2997 | " pages_scanned:%lu" |
2998 | " all_unreclaimable? %s" | |
2999 | "\n", | |
3000 | zone->name, | |
88f5acf8 | 3001 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3002 | K(min_wmark_pages(zone)), |
3003 | K(low_wmark_pages(zone)), | |
3004 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3005 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3006 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3007 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3008 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3009 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3010 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3011 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3012 | K(zone->present_pages), |
4a0aa73f KM |
3013 | K(zone_page_state(zone, NR_MLOCK)), |
3014 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3015 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3016 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3017 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3018 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3019 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3020 | zone_page_state(zone, NR_KERNEL_STACK) * |
3021 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3022 | K(zone_page_state(zone, NR_PAGETABLE)), |
3023 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3024 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3025 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3026 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3027 | zone->pages_scanned, |
93e4a89a | 3028 | (zone->all_unreclaimable ? "yes" : "no") |
1da177e4 LT |
3029 | ); |
3030 | printk("lowmem_reserve[]:"); | |
3031 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3032 | printk(" %lu", zone->lowmem_reserve[i]); | |
3033 | printk("\n"); | |
3034 | } | |
3035 | ||
ee99c71c | 3036 | for_each_populated_zone(zone) { |
8f9de51a | 3037 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3038 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3039 | |
7bf02ea2 | 3040 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3041 | continue; |
1da177e4 LT |
3042 | show_node(zone); |
3043 | printk("%s: ", zone->name); | |
1da177e4 LT |
3044 | |
3045 | spin_lock_irqsave(&zone->lock, flags); | |
3046 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3047 | struct free_area *area = &zone->free_area[order]; |
3048 | int type; | |
3049 | ||
3050 | nr[order] = area->nr_free; | |
8f9de51a | 3051 | total += nr[order] << order; |
377e4f16 RV |
3052 | |
3053 | types[order] = 0; | |
3054 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3055 | if (!list_empty(&area->free_list[type])) | |
3056 | types[order] |= 1 << type; | |
3057 | } | |
1da177e4 LT |
3058 | } |
3059 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3060 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3061 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3062 | if (nr[order]) |
3063 | show_migration_types(types[order]); | |
3064 | } | |
1da177e4 LT |
3065 | printk("= %lukB\n", K(total)); |
3066 | } | |
3067 | ||
e6f3602d LW |
3068 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3069 | ||
1da177e4 LT |
3070 | show_swap_cache_info(); |
3071 | } | |
3072 | ||
19770b32 MG |
3073 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3074 | { | |
3075 | zoneref->zone = zone; | |
3076 | zoneref->zone_idx = zone_idx(zone); | |
3077 | } | |
3078 | ||
1da177e4 LT |
3079 | /* |
3080 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3081 | * |
3082 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3083 | */ |
f0c0b2b8 KH |
3084 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
3085 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 3086 | { |
1a93205b CL |
3087 | struct zone *zone; |
3088 | ||
98d2b0eb | 3089 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 3090 | zone_type++; |
02a68a5e CL |
3091 | |
3092 | do { | |
2f6726e5 | 3093 | zone_type--; |
070f8032 | 3094 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3095 | if (populated_zone(zone)) { |
dd1a239f MG |
3096 | zoneref_set_zone(zone, |
3097 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3098 | check_highest_zone(zone_type); |
1da177e4 | 3099 | } |
02a68a5e | 3100 | |
2f6726e5 | 3101 | } while (zone_type); |
070f8032 | 3102 | return nr_zones; |
1da177e4 LT |
3103 | } |
3104 | ||
f0c0b2b8 KH |
3105 | |
3106 | /* | |
3107 | * zonelist_order: | |
3108 | * 0 = automatic detection of better ordering. | |
3109 | * 1 = order by ([node] distance, -zonetype) | |
3110 | * 2 = order by (-zonetype, [node] distance) | |
3111 | * | |
3112 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3113 | * the same zonelist. So only NUMA can configure this param. | |
3114 | */ | |
3115 | #define ZONELIST_ORDER_DEFAULT 0 | |
3116 | #define ZONELIST_ORDER_NODE 1 | |
3117 | #define ZONELIST_ORDER_ZONE 2 | |
3118 | ||
3119 | /* zonelist order in the kernel. | |
3120 | * set_zonelist_order() will set this to NODE or ZONE. | |
3121 | */ | |
3122 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3123 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3124 | ||
3125 | ||
1da177e4 | 3126 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3127 | /* The value user specified ....changed by config */ |
3128 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3129 | /* string for sysctl */ | |
3130 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3131 | char numa_zonelist_order[16] = "default"; | |
3132 | ||
3133 | /* | |
3134 | * interface for configure zonelist ordering. | |
3135 | * command line option "numa_zonelist_order" | |
3136 | * = "[dD]efault - default, automatic configuration. | |
3137 | * = "[nN]ode - order by node locality, then by zone within node | |
3138 | * = "[zZ]one - order by zone, then by locality within zone | |
3139 | */ | |
3140 | ||
3141 | static int __parse_numa_zonelist_order(char *s) | |
3142 | { | |
3143 | if (*s == 'd' || *s == 'D') { | |
3144 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3145 | } else if (*s == 'n' || *s == 'N') { | |
3146 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3147 | } else if (*s == 'z' || *s == 'Z') { | |
3148 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3149 | } else { | |
3150 | printk(KERN_WARNING | |
3151 | "Ignoring invalid numa_zonelist_order value: " | |
3152 | "%s\n", s); | |
3153 | return -EINVAL; | |
3154 | } | |
3155 | return 0; | |
3156 | } | |
3157 | ||
3158 | static __init int setup_numa_zonelist_order(char *s) | |
3159 | { | |
ecb256f8 VL |
3160 | int ret; |
3161 | ||
3162 | if (!s) | |
3163 | return 0; | |
3164 | ||
3165 | ret = __parse_numa_zonelist_order(s); | |
3166 | if (ret == 0) | |
3167 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3168 | ||
3169 | return ret; | |
f0c0b2b8 KH |
3170 | } |
3171 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3172 | ||
3173 | /* | |
3174 | * sysctl handler for numa_zonelist_order | |
3175 | */ | |
3176 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 3177 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3178 | loff_t *ppos) |
3179 | { | |
3180 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3181 | int ret; | |
443c6f14 | 3182 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3183 | |
443c6f14 | 3184 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 3185 | if (write) |
443c6f14 | 3186 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 3187 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3188 | if (ret) |
443c6f14 | 3189 | goto out; |
f0c0b2b8 KH |
3190 | if (write) { |
3191 | int oldval = user_zonelist_order; | |
3192 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
3193 | /* | |
3194 | * bogus value. restore saved string | |
3195 | */ | |
3196 | strncpy((char*)table->data, saved_string, | |
3197 | NUMA_ZONELIST_ORDER_LEN); | |
3198 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3199 | } else if (oldval != user_zonelist_order) { |
3200 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3201 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3202 | mutex_unlock(&zonelists_mutex); |
3203 | } | |
f0c0b2b8 | 3204 | } |
443c6f14 AK |
3205 | out: |
3206 | mutex_unlock(&zl_order_mutex); | |
3207 | return ret; | |
f0c0b2b8 KH |
3208 | } |
3209 | ||
3210 | ||
62bc62a8 | 3211 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3212 | static int node_load[MAX_NUMNODES]; |
3213 | ||
1da177e4 | 3214 | /** |
4dc3b16b | 3215 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3216 | * @node: node whose fallback list we're appending |
3217 | * @used_node_mask: nodemask_t of already used nodes | |
3218 | * | |
3219 | * We use a number of factors to determine which is the next node that should | |
3220 | * appear on a given node's fallback list. The node should not have appeared | |
3221 | * already in @node's fallback list, and it should be the next closest node | |
3222 | * according to the distance array (which contains arbitrary distance values | |
3223 | * from each node to each node in the system), and should also prefer nodes | |
3224 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3225 | * on them otherwise. | |
3226 | * It returns -1 if no node is found. | |
3227 | */ | |
f0c0b2b8 | 3228 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3229 | { |
4cf808eb | 3230 | int n, val; |
1da177e4 LT |
3231 | int min_val = INT_MAX; |
3232 | int best_node = -1; | |
a70f7302 | 3233 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3234 | |
4cf808eb LT |
3235 | /* Use the local node if we haven't already */ |
3236 | if (!node_isset(node, *used_node_mask)) { | |
3237 | node_set(node, *used_node_mask); | |
3238 | return node; | |
3239 | } | |
1da177e4 | 3240 | |
37b07e41 | 3241 | for_each_node_state(n, N_HIGH_MEMORY) { |
1da177e4 LT |
3242 | |
3243 | /* Don't want a node to appear more than once */ | |
3244 | if (node_isset(n, *used_node_mask)) | |
3245 | continue; | |
3246 | ||
1da177e4 LT |
3247 | /* Use the distance array to find the distance */ |
3248 | val = node_distance(node, n); | |
3249 | ||
4cf808eb LT |
3250 | /* Penalize nodes under us ("prefer the next node") */ |
3251 | val += (n < node); | |
3252 | ||
1da177e4 | 3253 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3254 | tmp = cpumask_of_node(n); |
3255 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3256 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3257 | ||
3258 | /* Slight preference for less loaded node */ | |
3259 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3260 | val += node_load[n]; | |
3261 | ||
3262 | if (val < min_val) { | |
3263 | min_val = val; | |
3264 | best_node = n; | |
3265 | } | |
3266 | } | |
3267 | ||
3268 | if (best_node >= 0) | |
3269 | node_set(best_node, *used_node_mask); | |
3270 | ||
3271 | return best_node; | |
3272 | } | |
3273 | ||
f0c0b2b8 KH |
3274 | |
3275 | /* | |
3276 | * Build zonelists ordered by node and zones within node. | |
3277 | * This results in maximum locality--normal zone overflows into local | |
3278 | * DMA zone, if any--but risks exhausting DMA zone. | |
3279 | */ | |
3280 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3281 | { |
f0c0b2b8 | 3282 | int j; |
1da177e4 | 3283 | struct zonelist *zonelist; |
f0c0b2b8 | 3284 | |
54a6eb5c | 3285 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3286 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
3287 | ; |
3288 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3289 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
3290 | zonelist->_zonerefs[j].zone = NULL; |
3291 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3292 | } |
3293 | ||
523b9458 CL |
3294 | /* |
3295 | * Build gfp_thisnode zonelists | |
3296 | */ | |
3297 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3298 | { | |
523b9458 CL |
3299 | int j; |
3300 | struct zonelist *zonelist; | |
3301 | ||
54a6eb5c MG |
3302 | zonelist = &pgdat->node_zonelists[1]; |
3303 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
3304 | zonelist->_zonerefs[j].zone = NULL; |
3305 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3306 | } |
3307 | ||
f0c0b2b8 KH |
3308 | /* |
3309 | * Build zonelists ordered by zone and nodes within zones. | |
3310 | * This results in conserving DMA zone[s] until all Normal memory is | |
3311 | * exhausted, but results in overflowing to remote node while memory | |
3312 | * may still exist in local DMA zone. | |
3313 | */ | |
3314 | static int node_order[MAX_NUMNODES]; | |
3315 | ||
3316 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3317 | { | |
f0c0b2b8 KH |
3318 | int pos, j, node; |
3319 | int zone_type; /* needs to be signed */ | |
3320 | struct zone *z; | |
3321 | struct zonelist *zonelist; | |
3322 | ||
54a6eb5c MG |
3323 | zonelist = &pgdat->node_zonelists[0]; |
3324 | pos = 0; | |
3325 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3326 | for (j = 0; j < nr_nodes; j++) { | |
3327 | node = node_order[j]; | |
3328 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3329 | if (populated_zone(z)) { | |
dd1a239f MG |
3330 | zoneref_set_zone(z, |
3331 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3332 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3333 | } |
3334 | } | |
f0c0b2b8 | 3335 | } |
dd1a239f MG |
3336 | zonelist->_zonerefs[pos].zone = NULL; |
3337 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3338 | } |
3339 | ||
3340 | static int default_zonelist_order(void) | |
3341 | { | |
3342 | int nid, zone_type; | |
3343 | unsigned long low_kmem_size,total_size; | |
3344 | struct zone *z; | |
3345 | int average_size; | |
3346 | /* | |
88393161 | 3347 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3348 | * If they are really small and used heavily, the system can fall |
3349 | * into OOM very easily. | |
e325c90f | 3350 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3351 | */ |
3352 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3353 | low_kmem_size = 0; | |
3354 | total_size = 0; | |
3355 | for_each_online_node(nid) { | |
3356 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3357 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3358 | if (populated_zone(z)) { | |
3359 | if (zone_type < ZONE_NORMAL) | |
3360 | low_kmem_size += z->present_pages; | |
3361 | total_size += z->present_pages; | |
e325c90f DR |
3362 | } else if (zone_type == ZONE_NORMAL) { |
3363 | /* | |
3364 | * If any node has only lowmem, then node order | |
3365 | * is preferred to allow kernel allocations | |
3366 | * locally; otherwise, they can easily infringe | |
3367 | * on other nodes when there is an abundance of | |
3368 | * lowmem available to allocate from. | |
3369 | */ | |
3370 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3371 | } |
3372 | } | |
3373 | } | |
3374 | if (!low_kmem_size || /* there are no DMA area. */ | |
3375 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3376 | return ZONELIST_ORDER_NODE; | |
3377 | /* | |
3378 | * look into each node's config. | |
3379 | * If there is a node whose DMA/DMA32 memory is very big area on | |
3380 | * local memory, NODE_ORDER may be suitable. | |
3381 | */ | |
37b07e41 LS |
3382 | average_size = total_size / |
3383 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
3384 | for_each_online_node(nid) { |
3385 | low_kmem_size = 0; | |
3386 | total_size = 0; | |
3387 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3388 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3389 | if (populated_zone(z)) { | |
3390 | if (zone_type < ZONE_NORMAL) | |
3391 | low_kmem_size += z->present_pages; | |
3392 | total_size += z->present_pages; | |
3393 | } | |
3394 | } | |
3395 | if (low_kmem_size && | |
3396 | total_size > average_size && /* ignore small node */ | |
3397 | low_kmem_size > total_size * 70/100) | |
3398 | return ZONELIST_ORDER_NODE; | |
3399 | } | |
3400 | return ZONELIST_ORDER_ZONE; | |
3401 | } | |
3402 | ||
3403 | static void set_zonelist_order(void) | |
3404 | { | |
3405 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3406 | current_zonelist_order = default_zonelist_order(); | |
3407 | else | |
3408 | current_zonelist_order = user_zonelist_order; | |
3409 | } | |
3410 | ||
3411 | static void build_zonelists(pg_data_t *pgdat) | |
3412 | { | |
3413 | int j, node, load; | |
3414 | enum zone_type i; | |
1da177e4 | 3415 | nodemask_t used_mask; |
f0c0b2b8 KH |
3416 | int local_node, prev_node; |
3417 | struct zonelist *zonelist; | |
3418 | int order = current_zonelist_order; | |
1da177e4 LT |
3419 | |
3420 | /* initialize zonelists */ | |
523b9458 | 3421 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3422 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3423 | zonelist->_zonerefs[0].zone = NULL; |
3424 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3425 | } |
3426 | ||
3427 | /* NUMA-aware ordering of nodes */ | |
3428 | local_node = pgdat->node_id; | |
62bc62a8 | 3429 | load = nr_online_nodes; |
1da177e4 LT |
3430 | prev_node = local_node; |
3431 | nodes_clear(used_mask); | |
f0c0b2b8 | 3432 | |
f0c0b2b8 KH |
3433 | memset(node_order, 0, sizeof(node_order)); |
3434 | j = 0; | |
3435 | ||
1da177e4 LT |
3436 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3437 | /* | |
3438 | * We don't want to pressure a particular node. | |
3439 | * So adding penalty to the first node in same | |
3440 | * distance group to make it round-robin. | |
3441 | */ | |
957f822a DR |
3442 | if (node_distance(local_node, node) != |
3443 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3444 | node_load[node] = load; |
3445 | ||
1da177e4 LT |
3446 | prev_node = node; |
3447 | load--; | |
f0c0b2b8 KH |
3448 | if (order == ZONELIST_ORDER_NODE) |
3449 | build_zonelists_in_node_order(pgdat, node); | |
3450 | else | |
3451 | node_order[j++] = node; /* remember order */ | |
3452 | } | |
1da177e4 | 3453 | |
f0c0b2b8 KH |
3454 | if (order == ZONELIST_ORDER_ZONE) { |
3455 | /* calculate node order -- i.e., DMA last! */ | |
3456 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3457 | } |
523b9458 CL |
3458 | |
3459 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3460 | } |
3461 | ||
9276b1bc | 3462 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3463 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3464 | { |
54a6eb5c MG |
3465 | struct zonelist *zonelist; |
3466 | struct zonelist_cache *zlc; | |
dd1a239f | 3467 | struct zoneref *z; |
9276b1bc | 3468 | |
54a6eb5c MG |
3469 | zonelist = &pgdat->node_zonelists[0]; |
3470 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3471 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3472 | for (z = zonelist->_zonerefs; z->zone; z++) |
3473 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3474 | } |
3475 | ||
7aac7898 LS |
3476 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3477 | /* | |
3478 | * Return node id of node used for "local" allocations. | |
3479 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3480 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3481 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3482 | */ | |
3483 | int local_memory_node(int node) | |
3484 | { | |
3485 | struct zone *zone; | |
3486 | ||
3487 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3488 | gfp_zone(GFP_KERNEL), | |
3489 | NULL, | |
3490 | &zone); | |
3491 | return zone->node; | |
3492 | } | |
3493 | #endif | |
f0c0b2b8 | 3494 | |
1da177e4 LT |
3495 | #else /* CONFIG_NUMA */ |
3496 | ||
f0c0b2b8 KH |
3497 | static void set_zonelist_order(void) |
3498 | { | |
3499 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3500 | } | |
3501 | ||
3502 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3503 | { |
19655d34 | 3504 | int node, local_node; |
54a6eb5c MG |
3505 | enum zone_type j; |
3506 | struct zonelist *zonelist; | |
1da177e4 LT |
3507 | |
3508 | local_node = pgdat->node_id; | |
1da177e4 | 3509 | |
54a6eb5c MG |
3510 | zonelist = &pgdat->node_zonelists[0]; |
3511 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 3512 | |
54a6eb5c MG |
3513 | /* |
3514 | * Now we build the zonelist so that it contains the zones | |
3515 | * of all the other nodes. | |
3516 | * We don't want to pressure a particular node, so when | |
3517 | * building the zones for node N, we make sure that the | |
3518 | * zones coming right after the local ones are those from | |
3519 | * node N+1 (modulo N) | |
3520 | */ | |
3521 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3522 | if (!node_online(node)) | |
3523 | continue; | |
3524 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3525 | MAX_NR_ZONES - 1); | |
1da177e4 | 3526 | } |
54a6eb5c MG |
3527 | for (node = 0; node < local_node; node++) { |
3528 | if (!node_online(node)) | |
3529 | continue; | |
3530 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3531 | MAX_NR_ZONES - 1); | |
3532 | } | |
3533 | ||
dd1a239f MG |
3534 | zonelist->_zonerefs[j].zone = NULL; |
3535 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3536 | } |
3537 | ||
9276b1bc | 3538 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3539 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3540 | { |
54a6eb5c | 3541 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3542 | } |
3543 | ||
1da177e4 LT |
3544 | #endif /* CONFIG_NUMA */ |
3545 | ||
99dcc3e5 CL |
3546 | /* |
3547 | * Boot pageset table. One per cpu which is going to be used for all | |
3548 | * zones and all nodes. The parameters will be set in such a way | |
3549 | * that an item put on a list will immediately be handed over to | |
3550 | * the buddy list. This is safe since pageset manipulation is done | |
3551 | * with interrupts disabled. | |
3552 | * | |
3553 | * The boot_pagesets must be kept even after bootup is complete for | |
3554 | * unused processors and/or zones. They do play a role for bootstrapping | |
3555 | * hotplugged processors. | |
3556 | * | |
3557 | * zoneinfo_show() and maybe other functions do | |
3558 | * not check if the processor is online before following the pageset pointer. | |
3559 | * Other parts of the kernel may not check if the zone is available. | |
3560 | */ | |
3561 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3562 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3563 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3564 | |
4eaf3f64 HL |
3565 | /* |
3566 | * Global mutex to protect against size modification of zonelists | |
3567 | * as well as to serialize pageset setup for the new populated zone. | |
3568 | */ | |
3569 | DEFINE_MUTEX(zonelists_mutex); | |
3570 | ||
9b1a4d38 | 3571 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3572 | static int __build_all_zonelists(void *data) |
1da177e4 | 3573 | { |
6811378e | 3574 | int nid; |
99dcc3e5 | 3575 | int cpu; |
9adb62a5 | 3576 | pg_data_t *self = data; |
9276b1bc | 3577 | |
7f9cfb31 BL |
3578 | #ifdef CONFIG_NUMA |
3579 | memset(node_load, 0, sizeof(node_load)); | |
3580 | #endif | |
9adb62a5 JL |
3581 | |
3582 | if (self && !node_online(self->node_id)) { | |
3583 | build_zonelists(self); | |
3584 | build_zonelist_cache(self); | |
3585 | } | |
3586 | ||
9276b1bc | 3587 | for_each_online_node(nid) { |
7ea1530a CL |
3588 | pg_data_t *pgdat = NODE_DATA(nid); |
3589 | ||
3590 | build_zonelists(pgdat); | |
3591 | build_zonelist_cache(pgdat); | |
9276b1bc | 3592 | } |
99dcc3e5 CL |
3593 | |
3594 | /* | |
3595 | * Initialize the boot_pagesets that are going to be used | |
3596 | * for bootstrapping processors. The real pagesets for | |
3597 | * each zone will be allocated later when the per cpu | |
3598 | * allocator is available. | |
3599 | * | |
3600 | * boot_pagesets are used also for bootstrapping offline | |
3601 | * cpus if the system is already booted because the pagesets | |
3602 | * are needed to initialize allocators on a specific cpu too. | |
3603 | * F.e. the percpu allocator needs the page allocator which | |
3604 | * needs the percpu allocator in order to allocate its pagesets | |
3605 | * (a chicken-egg dilemma). | |
3606 | */ | |
7aac7898 | 3607 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3608 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3609 | ||
7aac7898 LS |
3610 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3611 | /* | |
3612 | * We now know the "local memory node" for each node-- | |
3613 | * i.e., the node of the first zone in the generic zonelist. | |
3614 | * Set up numa_mem percpu variable for on-line cpus. During | |
3615 | * boot, only the boot cpu should be on-line; we'll init the | |
3616 | * secondary cpus' numa_mem as they come on-line. During | |
3617 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3618 | */ | |
3619 | if (cpu_online(cpu)) | |
3620 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3621 | #endif | |
3622 | } | |
3623 | ||
6811378e YG |
3624 | return 0; |
3625 | } | |
3626 | ||
4eaf3f64 HL |
3627 | /* |
3628 | * Called with zonelists_mutex held always | |
3629 | * unless system_state == SYSTEM_BOOTING. | |
3630 | */ | |
9adb62a5 | 3631 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3632 | { |
f0c0b2b8 KH |
3633 | set_zonelist_order(); |
3634 | ||
6811378e | 3635 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3636 | __build_all_zonelists(NULL); |
68ad8df4 | 3637 | mminit_verify_zonelist(); |
6811378e YG |
3638 | cpuset_init_current_mems_allowed(); |
3639 | } else { | |
183ff22b | 3640 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 3641 | of zonelist */ |
e9959f0f | 3642 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3643 | if (zone) |
3644 | setup_zone_pageset(zone); | |
e9959f0f | 3645 | #endif |
9adb62a5 | 3646 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3647 | /* cpuset refresh routine should be here */ |
3648 | } | |
bd1e22b8 | 3649 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3650 | /* |
3651 | * Disable grouping by mobility if the number of pages in the | |
3652 | * system is too low to allow the mechanism to work. It would be | |
3653 | * more accurate, but expensive to check per-zone. This check is | |
3654 | * made on memory-hotadd so a system can start with mobility | |
3655 | * disabled and enable it later | |
3656 | */ | |
d9c23400 | 3657 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3658 | page_group_by_mobility_disabled = 1; |
3659 | else | |
3660 | page_group_by_mobility_disabled = 0; | |
3661 | ||
3662 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3663 | "Total pages: %ld\n", | |
62bc62a8 | 3664 | nr_online_nodes, |
f0c0b2b8 | 3665 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3666 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3667 | vm_total_pages); |
3668 | #ifdef CONFIG_NUMA | |
3669 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3670 | #endif | |
1da177e4 LT |
3671 | } |
3672 | ||
3673 | /* | |
3674 | * Helper functions to size the waitqueue hash table. | |
3675 | * Essentially these want to choose hash table sizes sufficiently | |
3676 | * large so that collisions trying to wait on pages are rare. | |
3677 | * But in fact, the number of active page waitqueues on typical | |
3678 | * systems is ridiculously low, less than 200. So this is even | |
3679 | * conservative, even though it seems large. | |
3680 | * | |
3681 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3682 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3683 | */ | |
3684 | #define PAGES_PER_WAITQUEUE 256 | |
3685 | ||
cca448fe | 3686 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3687 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3688 | { |
3689 | unsigned long size = 1; | |
3690 | ||
3691 | pages /= PAGES_PER_WAITQUEUE; | |
3692 | ||
3693 | while (size < pages) | |
3694 | size <<= 1; | |
3695 | ||
3696 | /* | |
3697 | * Once we have dozens or even hundreds of threads sleeping | |
3698 | * on IO we've got bigger problems than wait queue collision. | |
3699 | * Limit the size of the wait table to a reasonable size. | |
3700 | */ | |
3701 | size = min(size, 4096UL); | |
3702 | ||
3703 | return max(size, 4UL); | |
3704 | } | |
cca448fe YG |
3705 | #else |
3706 | /* | |
3707 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3708 | * a suitable size for its wait_table. So we use the maximum size now. | |
3709 | * | |
3710 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3711 | * | |
3712 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3713 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3714 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3715 | * | |
3716 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3717 | * or more by the traditional way. (See above). It equals: | |
3718 | * | |
3719 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3720 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3721 | * powerpc (64K page size) : = (32G +16M)byte. | |
3722 | */ | |
3723 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3724 | { | |
3725 | return 4096UL; | |
3726 | } | |
3727 | #endif | |
1da177e4 LT |
3728 | |
3729 | /* | |
3730 | * This is an integer logarithm so that shifts can be used later | |
3731 | * to extract the more random high bits from the multiplicative | |
3732 | * hash function before the remainder is taken. | |
3733 | */ | |
3734 | static inline unsigned long wait_table_bits(unsigned long size) | |
3735 | { | |
3736 | return ffz(~size); | |
3737 | } | |
3738 | ||
3739 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
3740 | ||
6d3163ce AH |
3741 | /* |
3742 | * Check if a pageblock contains reserved pages | |
3743 | */ | |
3744 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3745 | { | |
3746 | unsigned long pfn; | |
3747 | ||
3748 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3749 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3750 | return 1; | |
3751 | } | |
3752 | return 0; | |
3753 | } | |
3754 | ||
56fd56b8 | 3755 | /* |
d9c23400 | 3756 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3757 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3758 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3759 | * higher will lead to a bigger reserve which will get freed as contiguous |
3760 | * blocks as reclaim kicks in | |
3761 | */ | |
3762 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3763 | { | |
6d3163ce | 3764 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3765 | struct page *page; |
78986a67 MG |
3766 | unsigned long block_migratetype; |
3767 | int reserve; | |
56fd56b8 | 3768 | |
d0215638 MH |
3769 | /* |
3770 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3771 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3772 | * make sure that we always check pfn_valid for the first page in | |
3773 | * the block. | |
3774 | */ | |
56fd56b8 MG |
3775 | start_pfn = zone->zone_start_pfn; |
3776 | end_pfn = start_pfn + zone->spanned_pages; | |
d0215638 | 3777 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3778 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3779 | pageblock_order; |
56fd56b8 | 3780 | |
78986a67 MG |
3781 | /* |
3782 | * Reserve blocks are generally in place to help high-order atomic | |
3783 | * allocations that are short-lived. A min_free_kbytes value that | |
3784 | * would result in more than 2 reserve blocks for atomic allocations | |
3785 | * is assumed to be in place to help anti-fragmentation for the | |
3786 | * future allocation of hugepages at runtime. | |
3787 | */ | |
3788 | reserve = min(2, reserve); | |
3789 | ||
d9c23400 | 3790 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3791 | if (!pfn_valid(pfn)) |
3792 | continue; | |
3793 | page = pfn_to_page(pfn); | |
3794 | ||
344c790e AL |
3795 | /* Watch out for overlapping nodes */ |
3796 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3797 | continue; | |
3798 | ||
56fd56b8 MG |
3799 | block_migratetype = get_pageblock_migratetype(page); |
3800 | ||
938929f1 MG |
3801 | /* Only test what is necessary when the reserves are not met */ |
3802 | if (reserve > 0) { | |
3803 | /* | |
3804 | * Blocks with reserved pages will never free, skip | |
3805 | * them. | |
3806 | */ | |
3807 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
3808 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
3809 | continue; | |
56fd56b8 | 3810 | |
938929f1 MG |
3811 | /* If this block is reserved, account for it */ |
3812 | if (block_migratetype == MIGRATE_RESERVE) { | |
3813 | reserve--; | |
3814 | continue; | |
3815 | } | |
3816 | ||
3817 | /* Suitable for reserving if this block is movable */ | |
3818 | if (block_migratetype == MIGRATE_MOVABLE) { | |
3819 | set_pageblock_migratetype(page, | |
3820 | MIGRATE_RESERVE); | |
3821 | move_freepages_block(zone, page, | |
3822 | MIGRATE_RESERVE); | |
3823 | reserve--; | |
3824 | continue; | |
3825 | } | |
56fd56b8 MG |
3826 | } |
3827 | ||
3828 | /* | |
3829 | * If the reserve is met and this is a previous reserved block, | |
3830 | * take it back | |
3831 | */ | |
3832 | if (block_migratetype == MIGRATE_RESERVE) { | |
3833 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3834 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3835 | } | |
3836 | } | |
3837 | } | |
ac0e5b7a | 3838 | |
1da177e4 LT |
3839 | /* |
3840 | * Initially all pages are reserved - free ones are freed | |
3841 | * up by free_all_bootmem() once the early boot process is | |
3842 | * done. Non-atomic initialization, single-pass. | |
3843 | */ | |
c09b4240 | 3844 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3845 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3846 | { |
1da177e4 | 3847 | struct page *page; |
29751f69 AW |
3848 | unsigned long end_pfn = start_pfn + size; |
3849 | unsigned long pfn; | |
86051ca5 | 3850 | struct zone *z; |
1da177e4 | 3851 | |
22b31eec HD |
3852 | if (highest_memmap_pfn < end_pfn - 1) |
3853 | highest_memmap_pfn = end_pfn - 1; | |
3854 | ||
86051ca5 | 3855 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3856 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3857 | /* |
3858 | * There can be holes in boot-time mem_map[]s | |
3859 | * handed to this function. They do not | |
3860 | * exist on hotplugged memory. | |
3861 | */ | |
3862 | if (context == MEMMAP_EARLY) { | |
3863 | if (!early_pfn_valid(pfn)) | |
3864 | continue; | |
3865 | if (!early_pfn_in_nid(pfn, nid)) | |
3866 | continue; | |
3867 | } | |
d41dee36 AW |
3868 | page = pfn_to_page(pfn); |
3869 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 3870 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 3871 | init_page_count(page); |
1da177e4 LT |
3872 | reset_page_mapcount(page); |
3873 | SetPageReserved(page); | |
b2a0ac88 MG |
3874 | /* |
3875 | * Mark the block movable so that blocks are reserved for | |
3876 | * movable at startup. This will force kernel allocations | |
3877 | * to reserve their blocks rather than leaking throughout | |
3878 | * the address space during boot when many long-lived | |
56fd56b8 MG |
3879 | * kernel allocations are made. Later some blocks near |
3880 | * the start are marked MIGRATE_RESERVE by | |
3881 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
3882 | * |
3883 | * bitmap is created for zone's valid pfn range. but memmap | |
3884 | * can be created for invalid pages (for alignment) | |
3885 | * check here not to call set_pageblock_migratetype() against | |
3886 | * pfn out of zone. | |
b2a0ac88 | 3887 | */ |
86051ca5 KH |
3888 | if ((z->zone_start_pfn <= pfn) |
3889 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
3890 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 3891 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 3892 | |
1da177e4 LT |
3893 | INIT_LIST_HEAD(&page->lru); |
3894 | #ifdef WANT_PAGE_VIRTUAL | |
3895 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3896 | if (!is_highmem_idx(zone)) | |
3212c6be | 3897 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 3898 | #endif |
1da177e4 LT |
3899 | } |
3900 | } | |
3901 | ||
1e548deb | 3902 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3903 | { |
b2a0ac88 MG |
3904 | int order, t; |
3905 | for_each_migratetype_order(order, t) { | |
3906 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3907 | zone->free_area[order].nr_free = 0; |
3908 | } | |
3909 | } | |
3910 | ||
3911 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3912 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3913 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3914 | #endif |
3915 | ||
4ed7e022 | 3916 | static int __meminit zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3917 | { |
3a6be87f | 3918 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3919 | int batch; |
3920 | ||
3921 | /* | |
3922 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3923 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3924 | * |
3925 | * OK, so we don't know how big the cache is. So guess. | |
3926 | */ | |
3927 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3928 | if (batch * PAGE_SIZE > 512 * 1024) |
3929 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3930 | batch /= 4; /* We effectively *= 4 below */ |
3931 | if (batch < 1) | |
3932 | batch = 1; | |
3933 | ||
3934 | /* | |
0ceaacc9 NP |
3935 | * Clamp the batch to a 2^n - 1 value. Having a power |
3936 | * of 2 value was found to be more likely to have | |
3937 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3938 | * |
0ceaacc9 NP |
3939 | * For example if 2 tasks are alternately allocating |
3940 | * batches of pages, one task can end up with a lot | |
3941 | * of pages of one half of the possible page colors | |
3942 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3943 | */ |
9155203a | 3944 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3945 | |
e7c8d5c9 | 3946 | return batch; |
3a6be87f DH |
3947 | |
3948 | #else | |
3949 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3950 | * conditions. | |
3951 | * | |
3952 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3953 | * of contiguous memory as there's no hardware page translation to | |
3954 | * assemble apparent contiguous memory from discontiguous pages. | |
3955 | * | |
3956 | * Queueing large contiguous runs of pages for batching, however, | |
3957 | * causes the pages to actually be freed in smaller chunks. As there | |
3958 | * can be a significant delay between the individual batches being | |
3959 | * recycled, this leads to the once large chunks of space being | |
3960 | * fragmented and becoming unavailable for high-order allocations. | |
3961 | */ | |
3962 | return 0; | |
3963 | #endif | |
e7c8d5c9 CL |
3964 | } |
3965 | ||
b69a7288 | 3966 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3967 | { |
3968 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3969 | int migratetype; |
2caaad41 | 3970 | |
1c6fe946 MD |
3971 | memset(p, 0, sizeof(*p)); |
3972 | ||
3dfa5721 | 3973 | pcp = &p->pcp; |
2caaad41 | 3974 | pcp->count = 0; |
2caaad41 CL |
3975 | pcp->high = 6 * batch; |
3976 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3977 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3978 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3979 | } |
3980 | ||
8ad4b1fb RS |
3981 | /* |
3982 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3983 | * to the value high for the pageset p. | |
3984 | */ | |
3985 | ||
3986 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3987 | unsigned long high) | |
3988 | { | |
3989 | struct per_cpu_pages *pcp; | |
3990 | ||
3dfa5721 | 3991 | pcp = &p->pcp; |
8ad4b1fb RS |
3992 | pcp->high = high; |
3993 | pcp->batch = max(1UL, high/4); | |
3994 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3995 | pcp->batch = PAGE_SHIFT * 8; | |
3996 | } | |
3997 | ||
4ed7e022 | 3998 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
3999 | { |
4000 | int cpu; | |
4001 | ||
4002 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
4003 | ||
4004 | for_each_possible_cpu(cpu) { | |
4005 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4006 | ||
4007 | setup_pageset(pcp, zone_batchsize(zone)); | |
4008 | ||
4009 | if (percpu_pagelist_fraction) | |
4010 | setup_pagelist_highmark(pcp, | |
4011 | (zone->present_pages / | |
4012 | percpu_pagelist_fraction)); | |
4013 | } | |
4014 | } | |
4015 | ||
2caaad41 | 4016 | /* |
99dcc3e5 CL |
4017 | * Allocate per cpu pagesets and initialize them. |
4018 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4019 | */ |
99dcc3e5 | 4020 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4021 | { |
99dcc3e5 | 4022 | struct zone *zone; |
e7c8d5c9 | 4023 | |
319774e2 WF |
4024 | for_each_populated_zone(zone) |
4025 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4026 | } |
4027 | ||
577a32f6 | 4028 | static noinline __init_refok |
cca448fe | 4029 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4030 | { |
4031 | int i; | |
4032 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 4033 | size_t alloc_size; |
ed8ece2e DH |
4034 | |
4035 | /* | |
4036 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4037 | * per zone. | |
4038 | */ | |
02b694de YG |
4039 | zone->wait_table_hash_nr_entries = |
4040 | wait_table_hash_nr_entries(zone_size_pages); | |
4041 | zone->wait_table_bits = | |
4042 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4043 | alloc_size = zone->wait_table_hash_nr_entries |
4044 | * sizeof(wait_queue_head_t); | |
4045 | ||
cd94b9db | 4046 | if (!slab_is_available()) { |
cca448fe | 4047 | zone->wait_table = (wait_queue_head_t *) |
8f389a99 | 4048 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
cca448fe YG |
4049 | } else { |
4050 | /* | |
4051 | * This case means that a zone whose size was 0 gets new memory | |
4052 | * via memory hot-add. | |
4053 | * But it may be the case that a new node was hot-added. In | |
4054 | * this case vmalloc() will not be able to use this new node's | |
4055 | * memory - this wait_table must be initialized to use this new | |
4056 | * node itself as well. | |
4057 | * To use this new node's memory, further consideration will be | |
4058 | * necessary. | |
4059 | */ | |
8691f3a7 | 4060 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4061 | } |
4062 | if (!zone->wait_table) | |
4063 | return -ENOMEM; | |
ed8ece2e | 4064 | |
02b694de | 4065 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4066 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4067 | |
4068 | return 0; | |
ed8ece2e DH |
4069 | } |
4070 | ||
c09b4240 | 4071 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4072 | { |
99dcc3e5 CL |
4073 | /* |
4074 | * per cpu subsystem is not up at this point. The following code | |
4075 | * relies on the ability of the linker to provide the | |
4076 | * offset of a (static) per cpu variable into the per cpu area. | |
4077 | */ | |
4078 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4079 | |
f5335c0f | 4080 | if (zone->present_pages) |
99dcc3e5 CL |
4081 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4082 | zone->name, zone->present_pages, | |
4083 | zone_batchsize(zone)); | |
ed8ece2e DH |
4084 | } |
4085 | ||
4ed7e022 | 4086 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4087 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4088 | unsigned long size, |
4089 | enum memmap_context context) | |
ed8ece2e DH |
4090 | { |
4091 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4092 | int ret; |
4093 | ret = zone_wait_table_init(zone, size); | |
4094 | if (ret) | |
4095 | return ret; | |
ed8ece2e DH |
4096 | pgdat->nr_zones = zone_idx(zone) + 1; |
4097 | ||
ed8ece2e DH |
4098 | zone->zone_start_pfn = zone_start_pfn; |
4099 | ||
708614e6 MG |
4100 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4101 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4102 | pgdat->node_id, | |
4103 | (unsigned long)zone_idx(zone), | |
4104 | zone_start_pfn, (zone_start_pfn + size)); | |
4105 | ||
1e548deb | 4106 | zone_init_free_lists(zone); |
718127cc YG |
4107 | |
4108 | return 0; | |
ed8ece2e DH |
4109 | } |
4110 | ||
0ee332c1 | 4111 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4112 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4113 | /* | |
4114 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
4115 | * Architectures may implement their own version but if add_active_range() | |
4116 | * was used and there are no special requirements, this is a convenient | |
4117 | * alternative | |
4118 | */ | |
f2dbcfa7 | 4119 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4120 | { |
c13291a5 TH |
4121 | unsigned long start_pfn, end_pfn; |
4122 | int i, nid; | |
c713216d | 4123 | |
c13291a5 | 4124 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
c713216d | 4125 | if (start_pfn <= pfn && pfn < end_pfn) |
c13291a5 | 4126 | return nid; |
cc2559bc KH |
4127 | /* This is a memory hole */ |
4128 | return -1; | |
c713216d MG |
4129 | } |
4130 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4131 | ||
f2dbcfa7 KH |
4132 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4133 | { | |
cc2559bc KH |
4134 | int nid; |
4135 | ||
4136 | nid = __early_pfn_to_nid(pfn); | |
4137 | if (nid >= 0) | |
4138 | return nid; | |
4139 | /* just returns 0 */ | |
4140 | return 0; | |
f2dbcfa7 KH |
4141 | } |
4142 | ||
cc2559bc KH |
4143 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4144 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4145 | { | |
4146 | int nid; | |
4147 | ||
4148 | nid = __early_pfn_to_nid(pfn); | |
4149 | if (nid >= 0 && nid != node) | |
4150 | return false; | |
4151 | return true; | |
4152 | } | |
4153 | #endif | |
f2dbcfa7 | 4154 | |
c713216d MG |
4155 | /** |
4156 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
4157 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
4158 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
4159 | * |
4160 | * If an architecture guarantees that all ranges registered with | |
4161 | * add_active_ranges() contain no holes and may be freed, this | |
4162 | * this function may be used instead of calling free_bootmem() manually. | |
4163 | */ | |
c13291a5 | 4164 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4165 | { |
c13291a5 TH |
4166 | unsigned long start_pfn, end_pfn; |
4167 | int i, this_nid; | |
edbe7d23 | 4168 | |
c13291a5 TH |
4169 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4170 | start_pfn = min(start_pfn, max_low_pfn); | |
4171 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4172 | |
c13291a5 TH |
4173 | if (start_pfn < end_pfn) |
4174 | free_bootmem_node(NODE_DATA(this_nid), | |
4175 | PFN_PHYS(start_pfn), | |
4176 | (end_pfn - start_pfn) << PAGE_SHIFT); | |
edbe7d23 | 4177 | } |
edbe7d23 | 4178 | } |
edbe7d23 | 4179 | |
c713216d MG |
4180 | /** |
4181 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4182 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
4183 | * |
4184 | * If an architecture guarantees that all ranges registered with | |
4185 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 4186 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
4187 | */ |
4188 | void __init sparse_memory_present_with_active_regions(int nid) | |
4189 | { | |
c13291a5 TH |
4190 | unsigned long start_pfn, end_pfn; |
4191 | int i, this_nid; | |
c713216d | 4192 | |
c13291a5 TH |
4193 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4194 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4195 | } |
4196 | ||
4197 | /** | |
4198 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4199 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4200 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4201 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4202 | * |
4203 | * It returns the start and end page frame of a node based on information | |
4204 | * provided by an arch calling add_active_range(). If called for a node | |
4205 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 4206 | * PFNs will be 0. |
c713216d | 4207 | */ |
a3142c8e | 4208 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4209 | unsigned long *start_pfn, unsigned long *end_pfn) |
4210 | { | |
c13291a5 | 4211 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4212 | int i; |
c13291a5 | 4213 | |
c713216d MG |
4214 | *start_pfn = -1UL; |
4215 | *end_pfn = 0; | |
4216 | ||
c13291a5 TH |
4217 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4218 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4219 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4220 | } |
4221 | ||
633c0666 | 4222 | if (*start_pfn == -1UL) |
c713216d | 4223 | *start_pfn = 0; |
c713216d MG |
4224 | } |
4225 | ||
2a1e274a MG |
4226 | /* |
4227 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4228 | * assumption is made that zones within a node are ordered in monotonic | |
4229 | * increasing memory addresses so that the "highest" populated zone is used | |
4230 | */ | |
b69a7288 | 4231 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4232 | { |
4233 | int zone_index; | |
4234 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4235 | if (zone_index == ZONE_MOVABLE) | |
4236 | continue; | |
4237 | ||
4238 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4239 | arch_zone_lowest_possible_pfn[zone_index]) | |
4240 | break; | |
4241 | } | |
4242 | ||
4243 | VM_BUG_ON(zone_index == -1); | |
4244 | movable_zone = zone_index; | |
4245 | } | |
4246 | ||
4247 | /* | |
4248 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4249 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4250 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4251 | * in each node depending on the size of each node and how evenly kernelcore | |
4252 | * is distributed. This helper function adjusts the zone ranges | |
4253 | * provided by the architecture for a given node by using the end of the | |
4254 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4255 | * zones within a node are in order of monotonic increases memory addresses | |
4256 | */ | |
b69a7288 | 4257 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4258 | unsigned long zone_type, |
4259 | unsigned long node_start_pfn, | |
4260 | unsigned long node_end_pfn, | |
4261 | unsigned long *zone_start_pfn, | |
4262 | unsigned long *zone_end_pfn) | |
4263 | { | |
4264 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4265 | if (zone_movable_pfn[nid]) { | |
4266 | /* Size ZONE_MOVABLE */ | |
4267 | if (zone_type == ZONE_MOVABLE) { | |
4268 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4269 | *zone_end_pfn = min(node_end_pfn, | |
4270 | arch_zone_highest_possible_pfn[movable_zone]); | |
4271 | ||
4272 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4273 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4274 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4275 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4276 | ||
4277 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4278 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4279 | *zone_start_pfn = *zone_end_pfn; | |
4280 | } | |
4281 | } | |
4282 | ||
c713216d MG |
4283 | /* |
4284 | * Return the number of pages a zone spans in a node, including holes | |
4285 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4286 | */ | |
6ea6e688 | 4287 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4288 | unsigned long zone_type, |
4289 | unsigned long *ignored) | |
4290 | { | |
4291 | unsigned long node_start_pfn, node_end_pfn; | |
4292 | unsigned long zone_start_pfn, zone_end_pfn; | |
4293 | ||
4294 | /* Get the start and end of the node and zone */ | |
4295 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
4296 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
4297 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4298 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4299 | node_start_pfn, node_end_pfn, | |
4300 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4301 | |
4302 | /* Check that this node has pages within the zone's required range */ | |
4303 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4304 | return 0; | |
4305 | ||
4306 | /* Move the zone boundaries inside the node if necessary */ | |
4307 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4308 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4309 | ||
4310 | /* Return the spanned pages */ | |
4311 | return zone_end_pfn - zone_start_pfn; | |
4312 | } | |
4313 | ||
4314 | /* | |
4315 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4316 | * then all holes in the requested range will be accounted for. |
c713216d | 4317 | */ |
32996250 | 4318 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4319 | unsigned long range_start_pfn, |
4320 | unsigned long range_end_pfn) | |
4321 | { | |
96e907d1 TH |
4322 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4323 | unsigned long start_pfn, end_pfn; | |
4324 | int i; | |
c713216d | 4325 | |
96e907d1 TH |
4326 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4327 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4328 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4329 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4330 | } |
96e907d1 | 4331 | return nr_absent; |
c713216d MG |
4332 | } |
4333 | ||
4334 | /** | |
4335 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4336 | * @start_pfn: The start PFN to start searching for holes | |
4337 | * @end_pfn: The end PFN to stop searching for holes | |
4338 | * | |
88ca3b94 | 4339 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4340 | */ |
4341 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4342 | unsigned long end_pfn) | |
4343 | { | |
4344 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4345 | } | |
4346 | ||
4347 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4348 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4349 | unsigned long zone_type, |
4350 | unsigned long *ignored) | |
4351 | { | |
96e907d1 TH |
4352 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4353 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4354 | unsigned long node_start_pfn, node_end_pfn; |
4355 | unsigned long zone_start_pfn, zone_end_pfn; | |
4356 | ||
4357 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
96e907d1 TH |
4358 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4359 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4360 | |
2a1e274a MG |
4361 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4362 | node_start_pfn, node_end_pfn, | |
4363 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4364 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4365 | } |
0e0b864e | 4366 | |
0ee332c1 | 4367 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4368 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4369 | unsigned long zone_type, |
4370 | unsigned long *zones_size) | |
4371 | { | |
4372 | return zones_size[zone_type]; | |
4373 | } | |
4374 | ||
6ea6e688 | 4375 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4376 | unsigned long zone_type, |
4377 | unsigned long *zholes_size) | |
4378 | { | |
4379 | if (!zholes_size) | |
4380 | return 0; | |
4381 | ||
4382 | return zholes_size[zone_type]; | |
4383 | } | |
0e0b864e | 4384 | |
0ee332c1 | 4385 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4386 | |
a3142c8e | 4387 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
4388 | unsigned long *zones_size, unsigned long *zholes_size) |
4389 | { | |
4390 | unsigned long realtotalpages, totalpages = 0; | |
4391 | enum zone_type i; | |
4392 | ||
4393 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4394 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
4395 | zones_size); | |
4396 | pgdat->node_spanned_pages = totalpages; | |
4397 | ||
4398 | realtotalpages = totalpages; | |
4399 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4400 | realtotalpages -= | |
4401 | zone_absent_pages_in_node(pgdat->node_id, i, | |
4402 | zholes_size); | |
4403 | pgdat->node_present_pages = realtotalpages; | |
4404 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4405 | realtotalpages); | |
4406 | } | |
4407 | ||
835c134e MG |
4408 | #ifndef CONFIG_SPARSEMEM |
4409 | /* | |
4410 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4411 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4412 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4413 | * round what is now in bits to nearest long in bits, then return it in |
4414 | * bytes. | |
4415 | */ | |
4416 | static unsigned long __init usemap_size(unsigned long zonesize) | |
4417 | { | |
4418 | unsigned long usemapsize; | |
4419 | ||
d9c23400 MG |
4420 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4421 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4422 | usemapsize *= NR_PAGEBLOCK_BITS; |
4423 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4424 | ||
4425 | return usemapsize / 8; | |
4426 | } | |
4427 | ||
4428 | static void __init setup_usemap(struct pglist_data *pgdat, | |
4429 | struct zone *zone, unsigned long zonesize) | |
4430 | { | |
4431 | unsigned long usemapsize = usemap_size(zonesize); | |
4432 | zone->pageblock_flags = NULL; | |
58a01a45 | 4433 | if (usemapsize) |
8f389a99 YL |
4434 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
4435 | usemapsize); | |
835c134e MG |
4436 | } |
4437 | #else | |
fa9f90be | 4438 | static inline void setup_usemap(struct pglist_data *pgdat, |
835c134e MG |
4439 | struct zone *zone, unsigned long zonesize) {} |
4440 | #endif /* CONFIG_SPARSEMEM */ | |
4441 | ||
d9c23400 | 4442 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4443 | |
d9c23400 | 4444 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
ca57df79 | 4445 | void __init set_pageblock_order(void) |
d9c23400 | 4446 | { |
955c1cd7 AM |
4447 | unsigned int order; |
4448 | ||
d9c23400 MG |
4449 | /* Check that pageblock_nr_pages has not already been setup */ |
4450 | if (pageblock_order) | |
4451 | return; | |
4452 | ||
955c1cd7 AM |
4453 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4454 | order = HUGETLB_PAGE_ORDER; | |
4455 | else | |
4456 | order = MAX_ORDER - 1; | |
4457 | ||
d9c23400 MG |
4458 | /* |
4459 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4460 | * This value may be variable depending on boot parameters on IA64 and |
4461 | * powerpc. | |
d9c23400 MG |
4462 | */ |
4463 | pageblock_order = order; | |
4464 | } | |
4465 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4466 | ||
ba72cb8c MG |
4467 | /* |
4468 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4469 | * is unused as pageblock_order is set at compile-time. See |
4470 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4471 | * the kernel config | |
ba72cb8c | 4472 | */ |
ca57df79 | 4473 | void __init set_pageblock_order(void) |
ba72cb8c | 4474 | { |
ba72cb8c | 4475 | } |
d9c23400 MG |
4476 | |
4477 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4478 | ||
1da177e4 LT |
4479 | /* |
4480 | * Set up the zone data structures: | |
4481 | * - mark all pages reserved | |
4482 | * - mark all memory queues empty | |
4483 | * - clear the memory bitmaps | |
6527af5d MK |
4484 | * |
4485 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4486 | */ |
b5a0e011 | 4487 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
4488 | unsigned long *zones_size, unsigned long *zholes_size) |
4489 | { | |
2f1b6248 | 4490 | enum zone_type j; |
ed8ece2e | 4491 | int nid = pgdat->node_id; |
1da177e4 | 4492 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4493 | int ret; |
1da177e4 | 4494 | |
208d54e5 | 4495 | pgdat_resize_init(pgdat); |
1da177e4 | 4496 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4497 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4498 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4499 | |
1da177e4 LT |
4500 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4501 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 4502 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 4503 | |
c713216d MG |
4504 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
4505 | realsize = size - zone_absent_pages_in_node(nid, j, | |
4506 | zholes_size); | |
1da177e4 | 4507 | |
0e0b864e MG |
4508 | /* |
4509 | * Adjust realsize so that it accounts for how much memory | |
4510 | * is used by this zone for memmap. This affects the watermark | |
4511 | * and per-cpu initialisations | |
4512 | */ | |
f7232154 JW |
4513 | memmap_pages = |
4514 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
0e0b864e MG |
4515 | if (realsize >= memmap_pages) { |
4516 | realsize -= memmap_pages; | |
5594c8c8 YL |
4517 | if (memmap_pages) |
4518 | printk(KERN_DEBUG | |
4519 | " %s zone: %lu pages used for memmap\n", | |
4520 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4521 | } else |
4522 | printk(KERN_WARNING | |
4523 | " %s zone: %lu pages exceeds realsize %lu\n", | |
4524 | zone_names[j], memmap_pages, realsize); | |
4525 | ||
6267276f CL |
4526 | /* Account for reserved pages */ |
4527 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 4528 | realsize -= dma_reserve; |
d903ef9f | 4529 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4530 | zone_names[0], dma_reserve); |
0e0b864e MG |
4531 | } |
4532 | ||
98d2b0eb | 4533 | if (!is_highmem_idx(j)) |
1da177e4 LT |
4534 | nr_kernel_pages += realsize; |
4535 | nr_all_pages += realsize; | |
4536 | ||
4537 | zone->spanned_pages = size; | |
4538 | zone->present_pages = realsize; | |
9614634f | 4539 | #ifdef CONFIG_NUMA |
d5f541ed | 4540 | zone->node = nid; |
8417bba4 | 4541 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 4542 | / 100; |
0ff38490 | 4543 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 4544 | #endif |
1da177e4 LT |
4545 | zone->name = zone_names[j]; |
4546 | spin_lock_init(&zone->lock); | |
4547 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4548 | zone_seqlock_init(zone); |
1da177e4 | 4549 | zone->zone_pgdat = pgdat; |
1da177e4 | 4550 | |
ed8ece2e | 4551 | zone_pcp_init(zone); |
bea8c150 | 4552 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4553 | if (!size) |
4554 | continue; | |
4555 | ||
955c1cd7 | 4556 | set_pageblock_order(); |
835c134e | 4557 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
4558 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4559 | size, MEMMAP_EARLY); | |
718127cc | 4560 | BUG_ON(ret); |
76cdd58e | 4561 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4562 | zone_start_pfn += size; |
1da177e4 LT |
4563 | } |
4564 | } | |
4565 | ||
577a32f6 | 4566 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4567 | { |
1da177e4 LT |
4568 | /* Skip empty nodes */ |
4569 | if (!pgdat->node_spanned_pages) | |
4570 | return; | |
4571 | ||
d41dee36 | 4572 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4573 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4574 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4575 | unsigned long size, start, end; |
d41dee36 AW |
4576 | struct page *map; |
4577 | ||
e984bb43 BP |
4578 | /* |
4579 | * The zone's endpoints aren't required to be MAX_ORDER | |
4580 | * aligned but the node_mem_map endpoints must be in order | |
4581 | * for the buddy allocator to function correctly. | |
4582 | */ | |
4583 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
4584 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
4585 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
4586 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4587 | map = alloc_remap(pgdat->node_id, size); |
4588 | if (!map) | |
8f389a99 | 4589 | map = alloc_bootmem_node_nopanic(pgdat, size); |
e984bb43 | 4590 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4591 | } |
12d810c1 | 4592 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4593 | /* |
4594 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4595 | */ | |
c713216d | 4596 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4597 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4598 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4599 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4600 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4601 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4602 | } |
1da177e4 | 4603 | #endif |
d41dee36 | 4604 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4605 | } |
4606 | ||
9109fb7b JW |
4607 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4608 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4609 | { |
9109fb7b JW |
4610 | pg_data_t *pgdat = NODE_DATA(nid); |
4611 | ||
88fdf75d | 4612 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4613 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4614 | |
1da177e4 LT |
4615 | pgdat->node_id = nid; |
4616 | pgdat->node_start_pfn = node_start_pfn; | |
957f822a | 4617 | init_zone_allows_reclaim(nid); |
c713216d | 4618 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
4619 | |
4620 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4621 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4622 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4623 | nid, (unsigned long)pgdat, | |
4624 | (unsigned long)pgdat->node_mem_map); | |
4625 | #endif | |
1da177e4 LT |
4626 | |
4627 | free_area_init_core(pgdat, zones_size, zholes_size); | |
4628 | } | |
4629 | ||
0ee332c1 | 4630 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4631 | |
4632 | #if MAX_NUMNODES > 1 | |
4633 | /* | |
4634 | * Figure out the number of possible node ids. | |
4635 | */ | |
4636 | static void __init setup_nr_node_ids(void) | |
4637 | { | |
4638 | unsigned int node; | |
4639 | unsigned int highest = 0; | |
4640 | ||
4641 | for_each_node_mask(node, node_possible_map) | |
4642 | highest = node; | |
4643 | nr_node_ids = highest + 1; | |
4644 | } | |
4645 | #else | |
4646 | static inline void setup_nr_node_ids(void) | |
4647 | { | |
4648 | } | |
4649 | #endif | |
4650 | ||
1e01979c TH |
4651 | /** |
4652 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4653 | * | |
4654 | * This function should be called after node map is populated and sorted. | |
4655 | * It calculates the maximum power of two alignment which can distinguish | |
4656 | * all the nodes. | |
4657 | * | |
4658 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4659 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4660 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4661 | * shifted, 1GiB is enough and this function will indicate so. | |
4662 | * | |
4663 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4664 | * model has fine enough granularity to avoid incorrect mapping for the | |
4665 | * populated node map. | |
4666 | * | |
4667 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4668 | * requirement (single node). | |
4669 | */ | |
4670 | unsigned long __init node_map_pfn_alignment(void) | |
4671 | { | |
4672 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4673 | unsigned long start, end, mask; |
1e01979c | 4674 | int last_nid = -1; |
c13291a5 | 4675 | int i, nid; |
1e01979c | 4676 | |
c13291a5 | 4677 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
4678 | if (!start || last_nid < 0 || last_nid == nid) { |
4679 | last_nid = nid; | |
4680 | last_end = end; | |
4681 | continue; | |
4682 | } | |
4683 | ||
4684 | /* | |
4685 | * Start with a mask granular enough to pin-point to the | |
4686 | * start pfn and tick off bits one-by-one until it becomes | |
4687 | * too coarse to separate the current node from the last. | |
4688 | */ | |
4689 | mask = ~((1 << __ffs(start)) - 1); | |
4690 | while (mask && last_end <= (start & (mask << 1))) | |
4691 | mask <<= 1; | |
4692 | ||
4693 | /* accumulate all internode masks */ | |
4694 | accl_mask |= mask; | |
4695 | } | |
4696 | ||
4697 | /* convert mask to number of pages */ | |
4698 | return ~accl_mask + 1; | |
4699 | } | |
4700 | ||
a6af2bc3 | 4701 | /* Find the lowest pfn for a node */ |
b69a7288 | 4702 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 4703 | { |
a6af2bc3 | 4704 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
4705 | unsigned long start_pfn; |
4706 | int i; | |
1abbfb41 | 4707 | |
c13291a5 TH |
4708 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
4709 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 4710 | |
a6af2bc3 MG |
4711 | if (min_pfn == ULONG_MAX) { |
4712 | printk(KERN_WARNING | |
2bc0d261 | 4713 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4714 | return 0; |
4715 | } | |
4716 | ||
4717 | return min_pfn; | |
c713216d MG |
4718 | } |
4719 | ||
4720 | /** | |
4721 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4722 | * | |
4723 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4724 | * add_active_range(). |
c713216d MG |
4725 | */ |
4726 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4727 | { | |
4728 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4729 | } | |
4730 | ||
37b07e41 LS |
4731 | /* |
4732 | * early_calculate_totalpages() | |
4733 | * Sum pages in active regions for movable zone. | |
4734 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4735 | */ | |
484f51f8 | 4736 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 4737 | { |
7e63efef | 4738 | unsigned long totalpages = 0; |
c13291a5 TH |
4739 | unsigned long start_pfn, end_pfn; |
4740 | int i, nid; | |
4741 | ||
4742 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4743 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 4744 | |
37b07e41 LS |
4745 | totalpages += pages; |
4746 | if (pages) | |
c13291a5 | 4747 | node_set_state(nid, N_HIGH_MEMORY); |
37b07e41 LS |
4748 | } |
4749 | return totalpages; | |
7e63efef MG |
4750 | } |
4751 | ||
2a1e274a MG |
4752 | /* |
4753 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4754 | * is spread evenly between nodes as long as the nodes have enough | |
4755 | * memory. When they don't, some nodes will have more kernelcore than | |
4756 | * others | |
4757 | */ | |
b224ef85 | 4758 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
4759 | { |
4760 | int i, nid; | |
4761 | unsigned long usable_startpfn; | |
4762 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd YL |
4763 | /* save the state before borrow the nodemask */ |
4764 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
37b07e41 LS |
4765 | unsigned long totalpages = early_calculate_totalpages(); |
4766 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 4767 | |
7e63efef MG |
4768 | /* |
4769 | * If movablecore was specified, calculate what size of | |
4770 | * kernelcore that corresponds so that memory usable for | |
4771 | * any allocation type is evenly spread. If both kernelcore | |
4772 | * and movablecore are specified, then the value of kernelcore | |
4773 | * will be used for required_kernelcore if it's greater than | |
4774 | * what movablecore would have allowed. | |
4775 | */ | |
4776 | if (required_movablecore) { | |
7e63efef MG |
4777 | unsigned long corepages; |
4778 | ||
4779 | /* | |
4780 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4781 | * was requested by the user | |
4782 | */ | |
4783 | required_movablecore = | |
4784 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4785 | corepages = totalpages - required_movablecore; | |
4786 | ||
4787 | required_kernelcore = max(required_kernelcore, corepages); | |
4788 | } | |
4789 | ||
2a1e274a MG |
4790 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
4791 | if (!required_kernelcore) | |
66918dcd | 4792 | goto out; |
2a1e274a MG |
4793 | |
4794 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4795 | find_usable_zone_for_movable(); | |
4796 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4797 | ||
4798 | restart: | |
4799 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4800 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 4801 | for_each_node_state(nid, N_HIGH_MEMORY) { |
c13291a5 TH |
4802 | unsigned long start_pfn, end_pfn; |
4803 | ||
2a1e274a MG |
4804 | /* |
4805 | * Recalculate kernelcore_node if the division per node | |
4806 | * now exceeds what is necessary to satisfy the requested | |
4807 | * amount of memory for the kernel | |
4808 | */ | |
4809 | if (required_kernelcore < kernelcore_node) | |
4810 | kernelcore_node = required_kernelcore / usable_nodes; | |
4811 | ||
4812 | /* | |
4813 | * As the map is walked, we track how much memory is usable | |
4814 | * by the kernel using kernelcore_remaining. When it is | |
4815 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4816 | */ | |
4817 | kernelcore_remaining = kernelcore_node; | |
4818 | ||
4819 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 4820 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
4821 | unsigned long size_pages; |
4822 | ||
c13291a5 | 4823 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
4824 | if (start_pfn >= end_pfn) |
4825 | continue; | |
4826 | ||
4827 | /* Account for what is only usable for kernelcore */ | |
4828 | if (start_pfn < usable_startpfn) { | |
4829 | unsigned long kernel_pages; | |
4830 | kernel_pages = min(end_pfn, usable_startpfn) | |
4831 | - start_pfn; | |
4832 | ||
4833 | kernelcore_remaining -= min(kernel_pages, | |
4834 | kernelcore_remaining); | |
4835 | required_kernelcore -= min(kernel_pages, | |
4836 | required_kernelcore); | |
4837 | ||
4838 | /* Continue if range is now fully accounted */ | |
4839 | if (end_pfn <= usable_startpfn) { | |
4840 | ||
4841 | /* | |
4842 | * Push zone_movable_pfn to the end so | |
4843 | * that if we have to rebalance | |
4844 | * kernelcore across nodes, we will | |
4845 | * not double account here | |
4846 | */ | |
4847 | zone_movable_pfn[nid] = end_pfn; | |
4848 | continue; | |
4849 | } | |
4850 | start_pfn = usable_startpfn; | |
4851 | } | |
4852 | ||
4853 | /* | |
4854 | * The usable PFN range for ZONE_MOVABLE is from | |
4855 | * start_pfn->end_pfn. Calculate size_pages as the | |
4856 | * number of pages used as kernelcore | |
4857 | */ | |
4858 | size_pages = end_pfn - start_pfn; | |
4859 | if (size_pages > kernelcore_remaining) | |
4860 | size_pages = kernelcore_remaining; | |
4861 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4862 | ||
4863 | /* | |
4864 | * Some kernelcore has been met, update counts and | |
4865 | * break if the kernelcore for this node has been | |
4866 | * satisified | |
4867 | */ | |
4868 | required_kernelcore -= min(required_kernelcore, | |
4869 | size_pages); | |
4870 | kernelcore_remaining -= size_pages; | |
4871 | if (!kernelcore_remaining) | |
4872 | break; | |
4873 | } | |
4874 | } | |
4875 | ||
4876 | /* | |
4877 | * If there is still required_kernelcore, we do another pass with one | |
4878 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4879 | * along on the nodes that still have memory until kernelcore is | |
4880 | * satisified | |
4881 | */ | |
4882 | usable_nodes--; | |
4883 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4884 | goto restart; | |
4885 | ||
4886 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4887 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4888 | zone_movable_pfn[nid] = | |
4889 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd YL |
4890 | |
4891 | out: | |
4892 | /* restore the node_state */ | |
4893 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
2a1e274a MG |
4894 | } |
4895 | ||
37b07e41 | 4896 | /* Any regular memory on that node ? */ |
4ed7e022 | 4897 | static void __init check_for_regular_memory(pg_data_t *pgdat) |
37b07e41 LS |
4898 | { |
4899 | #ifdef CONFIG_HIGHMEM | |
4900 | enum zone_type zone_type; | |
4901 | ||
4902 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4903 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
d0048b0e | 4904 | if (zone->present_pages) { |
37b07e41 | 4905 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); |
d0048b0e BL |
4906 | break; |
4907 | } | |
37b07e41 LS |
4908 | } |
4909 | #endif | |
4910 | } | |
4911 | ||
c713216d MG |
4912 | /** |
4913 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 4914 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
4915 | * |
4916 | * This will call free_area_init_node() for each active node in the system. | |
4917 | * Using the page ranges provided by add_active_range(), the size of each | |
4918 | * zone in each node and their holes is calculated. If the maximum PFN | |
4919 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4920 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4921 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4922 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4923 | * at arch_max_dma_pfn. | |
4924 | */ | |
4925 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4926 | { | |
c13291a5 TH |
4927 | unsigned long start_pfn, end_pfn; |
4928 | int i, nid; | |
a6af2bc3 | 4929 | |
c713216d MG |
4930 | /* Record where the zone boundaries are */ |
4931 | memset(arch_zone_lowest_possible_pfn, 0, | |
4932 | sizeof(arch_zone_lowest_possible_pfn)); | |
4933 | memset(arch_zone_highest_possible_pfn, 0, | |
4934 | sizeof(arch_zone_highest_possible_pfn)); | |
4935 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4936 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4937 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
4938 | if (i == ZONE_MOVABLE) |
4939 | continue; | |
c713216d MG |
4940 | arch_zone_lowest_possible_pfn[i] = |
4941 | arch_zone_highest_possible_pfn[i-1]; | |
4942 | arch_zone_highest_possible_pfn[i] = | |
4943 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4944 | } | |
2a1e274a MG |
4945 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
4946 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4947 | ||
4948 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4949 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 4950 | find_zone_movable_pfns_for_nodes(); |
c713216d | 4951 | |
c713216d | 4952 | /* Print out the zone ranges */ |
a62e2f4f | 4953 | printk("Zone ranges:\n"); |
2a1e274a MG |
4954 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4955 | if (i == ZONE_MOVABLE) | |
4956 | continue; | |
155cbfc8 | 4957 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
4958 | if (arch_zone_lowest_possible_pfn[i] == |
4959 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 4960 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 4961 | else |
a62e2f4f BH |
4962 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
4963 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
4964 | (arch_zone_highest_possible_pfn[i] | |
4965 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
4966 | } |
4967 | ||
4968 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 4969 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
4970 | for (i = 0; i < MAX_NUMNODES; i++) { |
4971 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
4972 | printk(" Node %d: %#010lx\n", i, |
4973 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 4974 | } |
c713216d | 4975 | |
f2d52fe5 | 4976 | /* Print out the early node map */ |
a62e2f4f | 4977 | printk("Early memory node ranges\n"); |
c13291a5 | 4978 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
4979 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
4980 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
4981 | |
4982 | /* Initialise every node */ | |
708614e6 | 4983 | mminit_verify_pageflags_layout(); |
8ef82866 | 4984 | setup_nr_node_ids(); |
c713216d MG |
4985 | for_each_online_node(nid) { |
4986 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 4987 | free_area_init_node(nid, NULL, |
c713216d | 4988 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
4989 | |
4990 | /* Any memory on that node */ | |
4991 | if (pgdat->node_present_pages) | |
4992 | node_set_state(nid, N_HIGH_MEMORY); | |
4993 | check_for_regular_memory(pgdat); | |
c713216d MG |
4994 | } |
4995 | } | |
2a1e274a | 4996 | |
7e63efef | 4997 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
4998 | { |
4999 | unsigned long long coremem; | |
5000 | if (!p) | |
5001 | return -EINVAL; | |
5002 | ||
5003 | coremem = memparse(p, &p); | |
7e63efef | 5004 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5005 | |
7e63efef | 5006 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5007 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5008 | ||
5009 | return 0; | |
5010 | } | |
ed7ed365 | 5011 | |
7e63efef MG |
5012 | /* |
5013 | * kernelcore=size sets the amount of memory for use for allocations that | |
5014 | * cannot be reclaimed or migrated. | |
5015 | */ | |
5016 | static int __init cmdline_parse_kernelcore(char *p) | |
5017 | { | |
5018 | return cmdline_parse_core(p, &required_kernelcore); | |
5019 | } | |
5020 | ||
5021 | /* | |
5022 | * movablecore=size sets the amount of memory for use for allocations that | |
5023 | * can be reclaimed or migrated. | |
5024 | */ | |
5025 | static int __init cmdline_parse_movablecore(char *p) | |
5026 | { | |
5027 | return cmdline_parse_core(p, &required_movablecore); | |
5028 | } | |
5029 | ||
ed7ed365 | 5030 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5031 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5032 | |
0ee332c1 | 5033 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5034 | |
0e0b864e | 5035 | /** |
88ca3b94 RD |
5036 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5037 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5038 | * |
5039 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5040 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5041 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5042 | * function may optionally be used to account for unfreeable pages in the |
5043 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5044 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5045 | */ |
5046 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5047 | { | |
5048 | dma_reserve = new_dma_reserve; | |
5049 | } | |
5050 | ||
1da177e4 LT |
5051 | void __init free_area_init(unsigned long *zones_size) |
5052 | { | |
9109fb7b | 5053 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5054 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5055 | } | |
1da177e4 | 5056 | |
1da177e4 LT |
5057 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5058 | unsigned long action, void *hcpu) | |
5059 | { | |
5060 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5061 | |
8bb78442 | 5062 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5063 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5064 | drain_pages(cpu); |
5065 | ||
5066 | /* | |
5067 | * Spill the event counters of the dead processor | |
5068 | * into the current processors event counters. | |
5069 | * This artificially elevates the count of the current | |
5070 | * processor. | |
5071 | */ | |
f8891e5e | 5072 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5073 | |
5074 | /* | |
5075 | * Zero the differential counters of the dead processor | |
5076 | * so that the vm statistics are consistent. | |
5077 | * | |
5078 | * This is only okay since the processor is dead and cannot | |
5079 | * race with what we are doing. | |
5080 | */ | |
2244b95a | 5081 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
5082 | } |
5083 | return NOTIFY_OK; | |
5084 | } | |
1da177e4 LT |
5085 | |
5086 | void __init page_alloc_init(void) | |
5087 | { | |
5088 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5089 | } | |
5090 | ||
cb45b0e9 HA |
5091 | /* |
5092 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5093 | * or min_free_kbytes changes. | |
5094 | */ | |
5095 | static void calculate_totalreserve_pages(void) | |
5096 | { | |
5097 | struct pglist_data *pgdat; | |
5098 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5099 | enum zone_type i, j; |
cb45b0e9 HA |
5100 | |
5101 | for_each_online_pgdat(pgdat) { | |
5102 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5103 | struct zone *zone = pgdat->node_zones + i; | |
5104 | unsigned long max = 0; | |
5105 | ||
5106 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5107 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5108 | if (zone->lowmem_reserve[j] > max) | |
5109 | max = zone->lowmem_reserve[j]; | |
5110 | } | |
5111 | ||
41858966 MG |
5112 | /* we treat the high watermark as reserved pages. */ |
5113 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
5114 | |
5115 | if (max > zone->present_pages) | |
5116 | max = zone->present_pages; | |
5117 | reserve_pages += max; | |
ab8fabd4 JW |
5118 | /* |
5119 | * Lowmem reserves are not available to | |
5120 | * GFP_HIGHUSER page cache allocations and | |
5121 | * kswapd tries to balance zones to their high | |
5122 | * watermark. As a result, neither should be | |
5123 | * regarded as dirtyable memory, to prevent a | |
5124 | * situation where reclaim has to clean pages | |
5125 | * in order to balance the zones. | |
5126 | */ | |
5127 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5128 | } |
5129 | } | |
ab8fabd4 | 5130 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5131 | totalreserve_pages = reserve_pages; |
5132 | } | |
5133 | ||
1da177e4 LT |
5134 | /* |
5135 | * setup_per_zone_lowmem_reserve - called whenever | |
5136 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5137 | * has a correct pages reserved value, so an adequate number of | |
5138 | * pages are left in the zone after a successful __alloc_pages(). | |
5139 | */ | |
5140 | static void setup_per_zone_lowmem_reserve(void) | |
5141 | { | |
5142 | struct pglist_data *pgdat; | |
2f6726e5 | 5143 | enum zone_type j, idx; |
1da177e4 | 5144 | |
ec936fc5 | 5145 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5146 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5147 | struct zone *zone = pgdat->node_zones + j; | |
5148 | unsigned long present_pages = zone->present_pages; | |
5149 | ||
5150 | zone->lowmem_reserve[j] = 0; | |
5151 | ||
2f6726e5 CL |
5152 | idx = j; |
5153 | while (idx) { | |
1da177e4 LT |
5154 | struct zone *lower_zone; |
5155 | ||
2f6726e5 CL |
5156 | idx--; |
5157 | ||
1da177e4 LT |
5158 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5159 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5160 | ||
5161 | lower_zone = pgdat->node_zones + idx; | |
5162 | lower_zone->lowmem_reserve[j] = present_pages / | |
5163 | sysctl_lowmem_reserve_ratio[idx]; | |
5164 | present_pages += lower_zone->present_pages; | |
5165 | } | |
5166 | } | |
5167 | } | |
cb45b0e9 HA |
5168 | |
5169 | /* update totalreserve_pages */ | |
5170 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5171 | } |
5172 | ||
cfd3da1e | 5173 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5174 | { |
5175 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5176 | unsigned long lowmem_pages = 0; | |
5177 | struct zone *zone; | |
5178 | unsigned long flags; | |
5179 | ||
5180 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5181 | for_each_zone(zone) { | |
5182 | if (!is_highmem(zone)) | |
5183 | lowmem_pages += zone->present_pages; | |
5184 | } | |
5185 | ||
5186 | for_each_zone(zone) { | |
ac924c60 AM |
5187 | u64 tmp; |
5188 | ||
1125b4e3 | 5189 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
5190 | tmp = (u64)pages_min * zone->present_pages; |
5191 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
5192 | if (is_highmem(zone)) { |
5193 | /* | |
669ed175 NP |
5194 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5195 | * need highmem pages, so cap pages_min to a small | |
5196 | * value here. | |
5197 | * | |
41858966 | 5198 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5199 | * deltas controls asynch page reclaim, and so should |
5200 | * not be capped for highmem. | |
1da177e4 LT |
5201 | */ |
5202 | int min_pages; | |
5203 | ||
5204 | min_pages = zone->present_pages / 1024; | |
5205 | if (min_pages < SWAP_CLUSTER_MAX) | |
5206 | min_pages = SWAP_CLUSTER_MAX; | |
5207 | if (min_pages > 128) | |
5208 | min_pages = 128; | |
41858966 | 5209 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5210 | } else { |
669ed175 NP |
5211 | /* |
5212 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5213 | * proportionate to the zone's size. |
5214 | */ | |
41858966 | 5215 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5216 | } |
5217 | ||
41858966 MG |
5218 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5219 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5220 | |
56fd56b8 | 5221 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5222 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5223 | } |
cb45b0e9 HA |
5224 | |
5225 | /* update totalreserve_pages */ | |
5226 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5227 | } |
5228 | ||
cfd3da1e MG |
5229 | /** |
5230 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5231 | * or when memory is hot-{added|removed} | |
5232 | * | |
5233 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5234 | * correctly with respect to min_free_kbytes. | |
5235 | */ | |
5236 | void setup_per_zone_wmarks(void) | |
5237 | { | |
5238 | mutex_lock(&zonelists_mutex); | |
5239 | __setup_per_zone_wmarks(); | |
5240 | mutex_unlock(&zonelists_mutex); | |
5241 | } | |
5242 | ||
55a4462a | 5243 | /* |
556adecb RR |
5244 | * The inactive anon list should be small enough that the VM never has to |
5245 | * do too much work, but large enough that each inactive page has a chance | |
5246 | * to be referenced again before it is swapped out. | |
5247 | * | |
5248 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5249 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5250 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5251 | * the anonymous pages are kept on the inactive list. | |
5252 | * | |
5253 | * total target max | |
5254 | * memory ratio inactive anon | |
5255 | * ------------------------------------- | |
5256 | * 10MB 1 5MB | |
5257 | * 100MB 1 50MB | |
5258 | * 1GB 3 250MB | |
5259 | * 10GB 10 0.9GB | |
5260 | * 100GB 31 3GB | |
5261 | * 1TB 101 10GB | |
5262 | * 10TB 320 32GB | |
5263 | */ | |
1b79acc9 | 5264 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5265 | { |
96cb4df5 | 5266 | unsigned int gb, ratio; |
556adecb | 5267 | |
96cb4df5 MK |
5268 | /* Zone size in gigabytes */ |
5269 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
5270 | if (gb) | |
556adecb | 5271 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5272 | else |
5273 | ratio = 1; | |
556adecb | 5274 | |
96cb4df5 MK |
5275 | zone->inactive_ratio = ratio; |
5276 | } | |
556adecb | 5277 | |
839a4fcc | 5278 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5279 | { |
5280 | struct zone *zone; | |
5281 | ||
5282 | for_each_zone(zone) | |
5283 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5284 | } |
5285 | ||
1da177e4 LT |
5286 | /* |
5287 | * Initialise min_free_kbytes. | |
5288 | * | |
5289 | * For small machines we want it small (128k min). For large machines | |
5290 | * we want it large (64MB max). But it is not linear, because network | |
5291 | * bandwidth does not increase linearly with machine size. We use | |
5292 | * | |
5293 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
5294 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
5295 | * | |
5296 | * which yields | |
5297 | * | |
5298 | * 16MB: 512k | |
5299 | * 32MB: 724k | |
5300 | * 64MB: 1024k | |
5301 | * 128MB: 1448k | |
5302 | * 256MB: 2048k | |
5303 | * 512MB: 2896k | |
5304 | * 1024MB: 4096k | |
5305 | * 2048MB: 5792k | |
5306 | * 4096MB: 8192k | |
5307 | * 8192MB: 11584k | |
5308 | * 16384MB: 16384k | |
5309 | */ | |
1b79acc9 | 5310 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5311 | { |
5312 | unsigned long lowmem_kbytes; | |
5313 | ||
5314 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5315 | ||
5316 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
5317 | if (min_free_kbytes < 128) | |
5318 | min_free_kbytes = 128; | |
5319 | if (min_free_kbytes > 65536) | |
5320 | min_free_kbytes = 65536; | |
bc75d33f | 5321 | setup_per_zone_wmarks(); |
a6cccdc3 | 5322 | refresh_zone_stat_thresholds(); |
1da177e4 | 5323 | setup_per_zone_lowmem_reserve(); |
556adecb | 5324 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5325 | return 0; |
5326 | } | |
bc75d33f | 5327 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5328 | |
5329 | /* | |
5330 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
5331 | * that we can call two helper functions whenever min_free_kbytes | |
5332 | * changes. | |
5333 | */ | |
5334 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5335 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5336 | { |
8d65af78 | 5337 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 5338 | if (write) |
bc75d33f | 5339 | setup_per_zone_wmarks(); |
1da177e4 LT |
5340 | return 0; |
5341 | } | |
5342 | ||
9614634f CL |
5343 | #ifdef CONFIG_NUMA |
5344 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5345 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5346 | { |
5347 | struct zone *zone; | |
5348 | int rc; | |
5349 | ||
8d65af78 | 5350 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5351 | if (rc) |
5352 | return rc; | |
5353 | ||
5354 | for_each_zone(zone) | |
8417bba4 | 5355 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
5356 | sysctl_min_unmapped_ratio) / 100; |
5357 | return 0; | |
5358 | } | |
0ff38490 CL |
5359 | |
5360 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5361 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5362 | { |
5363 | struct zone *zone; | |
5364 | int rc; | |
5365 | ||
8d65af78 | 5366 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5367 | if (rc) |
5368 | return rc; | |
5369 | ||
5370 | for_each_zone(zone) | |
5371 | zone->min_slab_pages = (zone->present_pages * | |
5372 | sysctl_min_slab_ratio) / 100; | |
5373 | return 0; | |
5374 | } | |
9614634f CL |
5375 | #endif |
5376 | ||
1da177e4 LT |
5377 | /* |
5378 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5379 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5380 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5381 | * | |
5382 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5383 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5384 | * if in function of the boot time zone sizes. |
5385 | */ | |
5386 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5387 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5388 | { |
8d65af78 | 5389 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5390 | setup_per_zone_lowmem_reserve(); |
5391 | return 0; | |
5392 | } | |
5393 | ||
8ad4b1fb RS |
5394 | /* |
5395 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
5396 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
5397 | * can have before it gets flushed back to buddy allocator. | |
5398 | */ | |
5399 | ||
5400 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5401 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5402 | { |
5403 | struct zone *zone; | |
5404 | unsigned int cpu; | |
5405 | int ret; | |
5406 | ||
8d65af78 | 5407 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
93278814 | 5408 | if (!write || (ret < 0)) |
8ad4b1fb | 5409 | return ret; |
364df0eb | 5410 | for_each_populated_zone(zone) { |
99dcc3e5 | 5411 | for_each_possible_cpu(cpu) { |
8ad4b1fb RS |
5412 | unsigned long high; |
5413 | high = zone->present_pages / percpu_pagelist_fraction; | |
99dcc3e5 CL |
5414 | setup_pagelist_highmark( |
5415 | per_cpu_ptr(zone->pageset, cpu), high); | |
8ad4b1fb RS |
5416 | } |
5417 | } | |
5418 | return 0; | |
5419 | } | |
5420 | ||
f034b5d4 | 5421 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5422 | |
5423 | #ifdef CONFIG_NUMA | |
5424 | static int __init set_hashdist(char *str) | |
5425 | { | |
5426 | if (!str) | |
5427 | return 0; | |
5428 | hashdist = simple_strtoul(str, &str, 0); | |
5429 | return 1; | |
5430 | } | |
5431 | __setup("hashdist=", set_hashdist); | |
5432 | #endif | |
5433 | ||
5434 | /* | |
5435 | * allocate a large system hash table from bootmem | |
5436 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5437 | * quantity of entries | |
5438 | * - limit is the number of hash buckets, not the total allocation size | |
5439 | */ | |
5440 | void *__init alloc_large_system_hash(const char *tablename, | |
5441 | unsigned long bucketsize, | |
5442 | unsigned long numentries, | |
5443 | int scale, | |
5444 | int flags, | |
5445 | unsigned int *_hash_shift, | |
5446 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5447 | unsigned long low_limit, |
5448 | unsigned long high_limit) | |
1da177e4 | 5449 | { |
31fe62b9 | 5450 | unsigned long long max = high_limit; |
1da177e4 LT |
5451 | unsigned long log2qty, size; |
5452 | void *table = NULL; | |
5453 | ||
5454 | /* allow the kernel cmdline to have a say */ | |
5455 | if (!numentries) { | |
5456 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5457 | numentries = nr_kernel_pages; |
1da177e4 LT |
5458 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
5459 | numentries >>= 20 - PAGE_SHIFT; | |
5460 | numentries <<= 20 - PAGE_SHIFT; | |
5461 | ||
5462 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5463 | if (scale > PAGE_SHIFT) | |
5464 | numentries >>= (scale - PAGE_SHIFT); | |
5465 | else | |
5466 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5467 | |
5468 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5469 | if (unlikely(flags & HASH_SMALL)) { |
5470 | /* Makes no sense without HASH_EARLY */ | |
5471 | WARN_ON(!(flags & HASH_EARLY)); | |
5472 | if (!(numentries >> *_hash_shift)) { | |
5473 | numentries = 1UL << *_hash_shift; | |
5474 | BUG_ON(!numentries); | |
5475 | } | |
5476 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5477 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5478 | } |
6e692ed3 | 5479 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5480 | |
5481 | /* limit allocation size to 1/16 total memory by default */ | |
5482 | if (max == 0) { | |
5483 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5484 | do_div(max, bucketsize); | |
5485 | } | |
074b8517 | 5486 | max = min(max, 0x80000000ULL); |
1da177e4 | 5487 | |
31fe62b9 TB |
5488 | if (numentries < low_limit) |
5489 | numentries = low_limit; | |
1da177e4 LT |
5490 | if (numentries > max) |
5491 | numentries = max; | |
5492 | ||
f0d1b0b3 | 5493 | log2qty = ilog2(numentries); |
1da177e4 LT |
5494 | |
5495 | do { | |
5496 | size = bucketsize << log2qty; | |
5497 | if (flags & HASH_EARLY) | |
74768ed8 | 5498 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5499 | else if (hashdist) |
5500 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5501 | else { | |
1037b83b ED |
5502 | /* |
5503 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5504 | * some pages at the end of hash table which |
5505 | * alloc_pages_exact() automatically does | |
1037b83b | 5506 | */ |
264ef8a9 | 5507 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5508 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5509 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5510 | } | |
1da177e4 LT |
5511 | } |
5512 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5513 | ||
5514 | if (!table) | |
5515 | panic("Failed to allocate %s hash table\n", tablename); | |
5516 | ||
f241e660 | 5517 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5518 | tablename, |
f241e660 | 5519 | (1UL << log2qty), |
f0d1b0b3 | 5520 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5521 | size); |
5522 | ||
5523 | if (_hash_shift) | |
5524 | *_hash_shift = log2qty; | |
5525 | if (_hash_mask) | |
5526 | *_hash_mask = (1 << log2qty) - 1; | |
5527 | ||
5528 | return table; | |
5529 | } | |
a117e66e | 5530 | |
835c134e MG |
5531 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5532 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5533 | unsigned long pfn) | |
5534 | { | |
5535 | #ifdef CONFIG_SPARSEMEM | |
5536 | return __pfn_to_section(pfn)->pageblock_flags; | |
5537 | #else | |
5538 | return zone->pageblock_flags; | |
5539 | #endif /* CONFIG_SPARSEMEM */ | |
5540 | } | |
5541 | ||
5542 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5543 | { | |
5544 | #ifdef CONFIG_SPARSEMEM | |
5545 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5546 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5547 | #else |
5548 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 5549 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5550 | #endif /* CONFIG_SPARSEMEM */ |
5551 | } | |
5552 | ||
5553 | /** | |
d9c23400 | 5554 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5555 | * @page: The page within the block of interest |
5556 | * @start_bitidx: The first bit of interest to retrieve | |
5557 | * @end_bitidx: The last bit of interest | |
5558 | * returns pageblock_bits flags | |
5559 | */ | |
5560 | unsigned long get_pageblock_flags_group(struct page *page, | |
5561 | int start_bitidx, int end_bitidx) | |
5562 | { | |
5563 | struct zone *zone; | |
5564 | unsigned long *bitmap; | |
5565 | unsigned long pfn, bitidx; | |
5566 | unsigned long flags = 0; | |
5567 | unsigned long value = 1; | |
5568 | ||
5569 | zone = page_zone(page); | |
5570 | pfn = page_to_pfn(page); | |
5571 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5572 | bitidx = pfn_to_bitidx(zone, pfn); | |
5573 | ||
5574 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5575 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5576 | flags |= value; | |
6220ec78 | 5577 | |
835c134e MG |
5578 | return flags; |
5579 | } | |
5580 | ||
5581 | /** | |
d9c23400 | 5582 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5583 | * @page: The page within the block of interest |
5584 | * @start_bitidx: The first bit of interest | |
5585 | * @end_bitidx: The last bit of interest | |
5586 | * @flags: The flags to set | |
5587 | */ | |
5588 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5589 | int start_bitidx, int end_bitidx) | |
5590 | { | |
5591 | struct zone *zone; | |
5592 | unsigned long *bitmap; | |
5593 | unsigned long pfn, bitidx; | |
5594 | unsigned long value = 1; | |
5595 | ||
5596 | zone = page_zone(page); | |
5597 | pfn = page_to_pfn(page); | |
5598 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5599 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
5600 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
5601 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
5602 | |
5603 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5604 | if (flags & value) | |
5605 | __set_bit(bitidx + start_bitidx, bitmap); | |
5606 | else | |
5607 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5608 | } | |
a5d76b54 KH |
5609 | |
5610 | /* | |
80934513 MK |
5611 | * This function checks whether pageblock includes unmovable pages or not. |
5612 | * If @count is not zero, it is okay to include less @count unmovable pages | |
5613 | * | |
5614 | * PageLRU check wihtout isolation or lru_lock could race so that | |
5615 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't | |
5616 | * expect this function should be exact. | |
a5d76b54 | 5617 | */ |
b023f468 WC |
5618 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
5619 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
5620 | { |
5621 | unsigned long pfn, iter, found; | |
47118af0 MN |
5622 | int mt; |
5623 | ||
49ac8255 KH |
5624 | /* |
5625 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 5626 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
5627 | */ |
5628 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 5629 | return false; |
47118af0 MN |
5630 | mt = get_pageblock_migratetype(page); |
5631 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 5632 | return false; |
49ac8255 KH |
5633 | |
5634 | pfn = page_to_pfn(page); | |
5635 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5636 | unsigned long check = pfn + iter; | |
5637 | ||
29723fcc | 5638 | if (!pfn_valid_within(check)) |
49ac8255 | 5639 | continue; |
29723fcc | 5640 | |
49ac8255 | 5641 | page = pfn_to_page(check); |
97d255c8 MK |
5642 | /* |
5643 | * We can't use page_count without pin a page | |
5644 | * because another CPU can free compound page. | |
5645 | * This check already skips compound tails of THP | |
5646 | * because their page->_count is zero at all time. | |
5647 | */ | |
5648 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
5649 | if (PageBuddy(page)) |
5650 | iter += (1 << page_order(page)) - 1; | |
5651 | continue; | |
5652 | } | |
97d255c8 | 5653 | |
b023f468 WC |
5654 | /* |
5655 | * The HWPoisoned page may be not in buddy system, and | |
5656 | * page_count() is not 0. | |
5657 | */ | |
5658 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
5659 | continue; | |
5660 | ||
49ac8255 KH |
5661 | if (!PageLRU(page)) |
5662 | found++; | |
5663 | /* | |
5664 | * If there are RECLAIMABLE pages, we need to check it. | |
5665 | * But now, memory offline itself doesn't call shrink_slab() | |
5666 | * and it still to be fixed. | |
5667 | */ | |
5668 | /* | |
5669 | * If the page is not RAM, page_count()should be 0. | |
5670 | * we don't need more check. This is an _used_ not-movable page. | |
5671 | * | |
5672 | * The problematic thing here is PG_reserved pages. PG_reserved | |
5673 | * is set to both of a memory hole page and a _used_ kernel | |
5674 | * page at boot. | |
5675 | */ | |
5676 | if (found > count) | |
80934513 | 5677 | return true; |
49ac8255 | 5678 | } |
80934513 | 5679 | return false; |
49ac8255 KH |
5680 | } |
5681 | ||
5682 | bool is_pageblock_removable_nolock(struct page *page) | |
5683 | { | |
656a0706 MH |
5684 | struct zone *zone; |
5685 | unsigned long pfn; | |
687875fb MH |
5686 | |
5687 | /* | |
5688 | * We have to be careful here because we are iterating over memory | |
5689 | * sections which are not zone aware so we might end up outside of | |
5690 | * the zone but still within the section. | |
656a0706 MH |
5691 | * We have to take care about the node as well. If the node is offline |
5692 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 5693 | */ |
656a0706 MH |
5694 | if (!node_online(page_to_nid(page))) |
5695 | return false; | |
5696 | ||
5697 | zone = page_zone(page); | |
5698 | pfn = page_to_pfn(page); | |
5699 | if (zone->zone_start_pfn > pfn || | |
687875fb MH |
5700 | zone->zone_start_pfn + zone->spanned_pages <= pfn) |
5701 | return false; | |
5702 | ||
b023f468 | 5703 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 5704 | } |
0c0e6195 | 5705 | |
041d3a8c MN |
5706 | #ifdef CONFIG_CMA |
5707 | ||
5708 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
5709 | { | |
5710 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
5711 | pageblock_nr_pages) - 1); | |
5712 | } | |
5713 | ||
5714 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
5715 | { | |
5716 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
5717 | pageblock_nr_pages)); | |
5718 | } | |
5719 | ||
041d3a8c | 5720 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
5721 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
5722 | unsigned long start, unsigned long end) | |
041d3a8c MN |
5723 | { |
5724 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 5725 | unsigned long nr_reclaimed; |
041d3a8c MN |
5726 | unsigned long pfn = start; |
5727 | unsigned int tries = 0; | |
5728 | int ret = 0; | |
5729 | ||
041d3a8c MN |
5730 | migrate_prep_local(); |
5731 | ||
bb13ffeb | 5732 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
5733 | if (fatal_signal_pending(current)) { |
5734 | ret = -EINTR; | |
5735 | break; | |
5736 | } | |
5737 | ||
bb13ffeb MG |
5738 | if (list_empty(&cc->migratepages)) { |
5739 | cc->nr_migratepages = 0; | |
5740 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 5741 | pfn, end, true); |
041d3a8c MN |
5742 | if (!pfn) { |
5743 | ret = -EINTR; | |
5744 | break; | |
5745 | } | |
5746 | tries = 0; | |
5747 | } else if (++tries == 5) { | |
5748 | ret = ret < 0 ? ret : -EBUSY; | |
5749 | break; | |
5750 | } | |
5751 | ||
beb51eaa MK |
5752 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
5753 | &cc->migratepages); | |
5754 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 5755 | |
bb13ffeb | 5756 | ret = migrate_pages(&cc->migratepages, |
723a0644 | 5757 | alloc_migrate_target, |
58f42fd5 | 5758 | 0, false, MIGRATE_SYNC); |
041d3a8c MN |
5759 | } |
5760 | ||
5733c7d1 | 5761 | putback_movable_pages(&cc->migratepages); |
041d3a8c MN |
5762 | return ret > 0 ? 0 : ret; |
5763 | } | |
5764 | ||
5765 | /** | |
5766 | * alloc_contig_range() -- tries to allocate given range of pages | |
5767 | * @start: start PFN to allocate | |
5768 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
5769 | * @migratetype: migratetype of the underlaying pageblocks (either |
5770 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
5771 | * in range must have the same migratetype and it must | |
5772 | * be either of the two. | |
041d3a8c MN |
5773 | * |
5774 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
5775 | * aligned, however it's the caller's responsibility to guarantee that | |
5776 | * we are the only thread that changes migrate type of pageblocks the | |
5777 | * pages fall in. | |
5778 | * | |
5779 | * The PFN range must belong to a single zone. | |
5780 | * | |
5781 | * Returns zero on success or negative error code. On success all | |
5782 | * pages which PFN is in [start, end) are allocated for the caller and | |
5783 | * need to be freed with free_contig_range(). | |
5784 | */ | |
0815f3d8 MN |
5785 | int alloc_contig_range(unsigned long start, unsigned long end, |
5786 | unsigned migratetype) | |
041d3a8c | 5787 | { |
041d3a8c MN |
5788 | unsigned long outer_start, outer_end; |
5789 | int ret = 0, order; | |
5790 | ||
bb13ffeb MG |
5791 | struct compact_control cc = { |
5792 | .nr_migratepages = 0, | |
5793 | .order = -1, | |
5794 | .zone = page_zone(pfn_to_page(start)), | |
5795 | .sync = true, | |
5796 | .ignore_skip_hint = true, | |
5797 | }; | |
5798 | INIT_LIST_HEAD(&cc.migratepages); | |
5799 | ||
041d3a8c MN |
5800 | /* |
5801 | * What we do here is we mark all pageblocks in range as | |
5802 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
5803 | * have different sizes, and due to the way page allocator | |
5804 | * work, we align the range to biggest of the two pages so | |
5805 | * that page allocator won't try to merge buddies from | |
5806 | * different pageblocks and change MIGRATE_ISOLATE to some | |
5807 | * other migration type. | |
5808 | * | |
5809 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
5810 | * migrate the pages from an unaligned range (ie. pages that | |
5811 | * we are interested in). This will put all the pages in | |
5812 | * range back to page allocator as MIGRATE_ISOLATE. | |
5813 | * | |
5814 | * When this is done, we take the pages in range from page | |
5815 | * allocator removing them from the buddy system. This way | |
5816 | * page allocator will never consider using them. | |
5817 | * | |
5818 | * This lets us mark the pageblocks back as | |
5819 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
5820 | * aligned range but not in the unaligned, original range are | |
5821 | * put back to page allocator so that buddy can use them. | |
5822 | */ | |
5823 | ||
5824 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
5825 | pfn_max_align_up(end), migratetype, |
5826 | false); | |
041d3a8c | 5827 | if (ret) |
86a595f9 | 5828 | return ret; |
041d3a8c | 5829 | |
bb13ffeb | 5830 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
5831 | if (ret) |
5832 | goto done; | |
5833 | ||
5834 | /* | |
5835 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
5836 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
5837 | * more, all pages in [start, end) are free in page allocator. | |
5838 | * What we are going to do is to allocate all pages from | |
5839 | * [start, end) (that is remove them from page allocator). | |
5840 | * | |
5841 | * The only problem is that pages at the beginning and at the | |
5842 | * end of interesting range may be not aligned with pages that | |
5843 | * page allocator holds, ie. they can be part of higher order | |
5844 | * pages. Because of this, we reserve the bigger range and | |
5845 | * once this is done free the pages we are not interested in. | |
5846 | * | |
5847 | * We don't have to hold zone->lock here because the pages are | |
5848 | * isolated thus they won't get removed from buddy. | |
5849 | */ | |
5850 | ||
5851 | lru_add_drain_all(); | |
5852 | drain_all_pages(); | |
5853 | ||
5854 | order = 0; | |
5855 | outer_start = start; | |
5856 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
5857 | if (++order >= MAX_ORDER) { | |
5858 | ret = -EBUSY; | |
5859 | goto done; | |
5860 | } | |
5861 | outer_start &= ~0UL << order; | |
5862 | } | |
5863 | ||
5864 | /* Make sure the range is really isolated. */ | |
b023f468 | 5865 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
5866 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
5867 | outer_start, end); | |
5868 | ret = -EBUSY; | |
5869 | goto done; | |
5870 | } | |
5871 | ||
49f223a9 MS |
5872 | |
5873 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 5874 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
5875 | if (!outer_end) { |
5876 | ret = -EBUSY; | |
5877 | goto done; | |
5878 | } | |
5879 | ||
5880 | /* Free head and tail (if any) */ | |
5881 | if (start != outer_start) | |
5882 | free_contig_range(outer_start, start - outer_start); | |
5883 | if (end != outer_end) | |
5884 | free_contig_range(end, outer_end - end); | |
5885 | ||
5886 | done: | |
5887 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 5888 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
5889 | return ret; |
5890 | } | |
5891 | ||
5892 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
5893 | { | |
5894 | for (; nr_pages--; ++pfn) | |
5895 | __free_page(pfn_to_page(pfn)); | |
5896 | } | |
5897 | #endif | |
5898 | ||
4ed7e022 JL |
5899 | #ifdef CONFIG_MEMORY_HOTPLUG |
5900 | static int __meminit __zone_pcp_update(void *data) | |
5901 | { | |
5902 | struct zone *zone = data; | |
5903 | int cpu; | |
5904 | unsigned long batch = zone_batchsize(zone), flags; | |
5905 | ||
5906 | for_each_possible_cpu(cpu) { | |
5907 | struct per_cpu_pageset *pset; | |
5908 | struct per_cpu_pages *pcp; | |
5909 | ||
5910 | pset = per_cpu_ptr(zone->pageset, cpu); | |
5911 | pcp = &pset->pcp; | |
5912 | ||
5913 | local_irq_save(flags); | |
5914 | if (pcp->count > 0) | |
5915 | free_pcppages_bulk(zone, pcp->count, pcp); | |
5a883813 | 5916 | drain_zonestat(zone, pset); |
4ed7e022 JL |
5917 | setup_pageset(pset, batch); |
5918 | local_irq_restore(flags); | |
5919 | } | |
5920 | return 0; | |
5921 | } | |
5922 | ||
5923 | void __meminit zone_pcp_update(struct zone *zone) | |
5924 | { | |
5925 | stop_machine(__zone_pcp_update, zone, NULL); | |
5926 | } | |
5927 | #endif | |
5928 | ||
340175b7 JL |
5929 | void zone_pcp_reset(struct zone *zone) |
5930 | { | |
5931 | unsigned long flags; | |
5a883813 MK |
5932 | int cpu; |
5933 | struct per_cpu_pageset *pset; | |
340175b7 JL |
5934 | |
5935 | /* avoid races with drain_pages() */ | |
5936 | local_irq_save(flags); | |
5937 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
5938 | for_each_online_cpu(cpu) { |
5939 | pset = per_cpu_ptr(zone->pageset, cpu); | |
5940 | drain_zonestat(zone, pset); | |
5941 | } | |
340175b7 JL |
5942 | free_percpu(zone->pageset); |
5943 | zone->pageset = &boot_pageset; | |
5944 | } | |
5945 | local_irq_restore(flags); | |
5946 | } | |
5947 | ||
6dcd73d7 | 5948 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
5949 | /* |
5950 | * All pages in the range must be isolated before calling this. | |
5951 | */ | |
5952 | void | |
5953 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5954 | { | |
5955 | struct page *page; | |
5956 | struct zone *zone; | |
5957 | int order, i; | |
5958 | unsigned long pfn; | |
5959 | unsigned long flags; | |
5960 | /* find the first valid pfn */ | |
5961 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5962 | if (pfn_valid(pfn)) | |
5963 | break; | |
5964 | if (pfn == end_pfn) | |
5965 | return; | |
5966 | zone = page_zone(pfn_to_page(pfn)); | |
5967 | spin_lock_irqsave(&zone->lock, flags); | |
5968 | pfn = start_pfn; | |
5969 | while (pfn < end_pfn) { | |
5970 | if (!pfn_valid(pfn)) { | |
5971 | pfn++; | |
5972 | continue; | |
5973 | } | |
5974 | page = pfn_to_page(pfn); | |
b023f468 WC |
5975 | /* |
5976 | * The HWPoisoned page may be not in buddy system, and | |
5977 | * page_count() is not 0. | |
5978 | */ | |
5979 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
5980 | pfn++; | |
5981 | SetPageReserved(page); | |
5982 | continue; | |
5983 | } | |
5984 | ||
0c0e6195 KH |
5985 | BUG_ON(page_count(page)); |
5986 | BUG_ON(!PageBuddy(page)); | |
5987 | order = page_order(page); | |
5988 | #ifdef CONFIG_DEBUG_VM | |
5989 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5990 | pfn, 1 << order, end_pfn); | |
5991 | #endif | |
5992 | list_del(&page->lru); | |
5993 | rmv_page_order(page); | |
5994 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
5995 | for (i = 0; i < (1 << order); i++) |
5996 | SetPageReserved((page+i)); | |
5997 | pfn += (1 << order); | |
5998 | } | |
5999 | spin_unlock_irqrestore(&zone->lock, flags); | |
6000 | } | |
6001 | #endif | |
8d22ba1b WF |
6002 | |
6003 | #ifdef CONFIG_MEMORY_FAILURE | |
6004 | bool is_free_buddy_page(struct page *page) | |
6005 | { | |
6006 | struct zone *zone = page_zone(page); | |
6007 | unsigned long pfn = page_to_pfn(page); | |
6008 | unsigned long flags; | |
6009 | int order; | |
6010 | ||
6011 | spin_lock_irqsave(&zone->lock, flags); | |
6012 | for (order = 0; order < MAX_ORDER; order++) { | |
6013 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6014 | ||
6015 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6016 | break; | |
6017 | } | |
6018 | spin_unlock_irqrestore(&zone->lock, flags); | |
6019 | ||
6020 | return order < MAX_ORDER; | |
6021 | } | |
6022 | #endif | |
718a3821 | 6023 | |
51300cef | 6024 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6025 | {1UL << PG_locked, "locked" }, |
6026 | {1UL << PG_error, "error" }, | |
6027 | {1UL << PG_referenced, "referenced" }, | |
6028 | {1UL << PG_uptodate, "uptodate" }, | |
6029 | {1UL << PG_dirty, "dirty" }, | |
6030 | {1UL << PG_lru, "lru" }, | |
6031 | {1UL << PG_active, "active" }, | |
6032 | {1UL << PG_slab, "slab" }, | |
6033 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6034 | {1UL << PG_arch_1, "arch_1" }, | |
6035 | {1UL << PG_reserved, "reserved" }, | |
6036 | {1UL << PG_private, "private" }, | |
6037 | {1UL << PG_private_2, "private_2" }, | |
6038 | {1UL << PG_writeback, "writeback" }, | |
6039 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6040 | {1UL << PG_head, "head" }, | |
6041 | {1UL << PG_tail, "tail" }, | |
6042 | #else | |
6043 | {1UL << PG_compound, "compound" }, | |
6044 | #endif | |
6045 | {1UL << PG_swapcache, "swapcache" }, | |
6046 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6047 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6048 | {1UL << PG_swapbacked, "swapbacked" }, |
6049 | {1UL << PG_unevictable, "unevictable" }, | |
6050 | #ifdef CONFIG_MMU | |
6051 | {1UL << PG_mlocked, "mlocked" }, | |
6052 | #endif | |
6053 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6054 | {1UL << PG_uncached, "uncached" }, | |
6055 | #endif | |
6056 | #ifdef CONFIG_MEMORY_FAILURE | |
6057 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6058 | #endif |
6059 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6060 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6061 | #endif |
718a3821 WF |
6062 | }; |
6063 | ||
6064 | static void dump_page_flags(unsigned long flags) | |
6065 | { | |
6066 | const char *delim = ""; | |
6067 | unsigned long mask; | |
6068 | int i; | |
6069 | ||
51300cef | 6070 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6071 | |
718a3821 WF |
6072 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6073 | ||
6074 | /* remove zone id */ | |
6075 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6076 | ||
51300cef | 6077 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6078 | |
6079 | mask = pageflag_names[i].mask; | |
6080 | if ((flags & mask) != mask) | |
6081 | continue; | |
6082 | ||
6083 | flags &= ~mask; | |
6084 | printk("%s%s", delim, pageflag_names[i].name); | |
6085 | delim = "|"; | |
6086 | } | |
6087 | ||
6088 | /* check for left over flags */ | |
6089 | if (flags) | |
6090 | printk("%s%#lx", delim, flags); | |
6091 | ||
6092 | printk(")\n"); | |
6093 | } | |
6094 | ||
6095 | void dump_page(struct page *page) | |
6096 | { | |
6097 | printk(KERN_ALERT | |
6098 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6099 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6100 | page->mapping, page->index); |
6101 | dump_page_flags(page->flags); | |
f212ad7c | 6102 | mem_cgroup_print_bad_page(page); |
718a3821 | 6103 | } |